Original Research

Substance use disorders (SUDs) are an important but understudied aspect of treating patients diagnosed with cancer. Substance use can affect cancer treatment outcomes, including morbidity and mortality.^1,2 Additionally, patients with cancer and SUD may have unique psychosocial needs that require close attention and management. There is a paucity of data regarding the best approach to treating such patients. For example, cocaine use may increase the cardiovascular and hematologic risk of some traditional chemotherapy agents.^3,4 Newer targeted agents and immunotherapies remain understudied with respect to SUD risk.

Although the US Department of Veterans Affairs (VA) has established helpful clinical practice guidelines for the treatment of SUD, there are no guidelines for treating patients with SUD and cancer.⁵ Clinicians have limited confidence in treatment approach, and treatment is inconsistent among oncologists nationwide even within the same practice. Furthermore, it can be challenging to safely prescribe opioids for cancer-related pain in individuals with SUD. There is a high risk of SUD and mental health disorders in veterans, making this population particularly vulnerable. We report a case of a male with metastatic pancreatic cancer, severe opioid use disorder (OUD) and moderate cocaine use disorder (CUD) who received pain management and cancer treatment under the direction of a multidisciplinary team approach.

Case Report

A 63-year-old male with a medical history of HIV treated with highly active antiretroviral therapy (HAART), compensated cirrhosis, severe OUD, moderate CUD, and sedative use disorder in sustained remission was admitted to the West Haven campus of the VA Connecticut Healthcare System (VACHS) with abdominal pain, weight loss and fatigue. He used heroin 1 month prior to his admission and reported regular cocaine and marijuana use (Table 1). He was diagnosed with HIV in 1989, and his medical history included herpes zoster and oral candidiasis but no other opportunistic infections. Several months prior to this admission, he had an undetectable viral load and CD4 count of 688.

At the time of this admission, the patient was adherent to methadone treatment. He reported increased abdominal pain. Computed tomography (CT) showed a 2.4-cm mass in the pancreatic uncinate process, multiple liver metastases, retroperitoneal lymphadenopathy, and small lung nodules. A CT-guided liver biopsy showed adenocarcinoma consistent with a primary cancer of the pancreas. Given the complexity of the case, a multidisciplinary team approach was used to treat his cancer and the sequelae safely, including the oncology team, community living center team, palliative care team, and interprofessional opioid reassessment clinic team (ORC).

Cancer Treatment

Chemotherapy with FOLFIRINOX (leucovorin calcium, fluorouracil, irinotecan hydrochloride, and oxaliplatin) was recommended. The first cycle of treatment originally was planned for the outpatient setting, and a peripherally inserted central catheter (PICC) line was placed. However, after a urine toxicology test was positive for cocaine, the PICC line was removed due to concern for possible use of PICC line for nonprescribed substance use. The patient expressed suicidal ideation at the time and was admitted for psychiatric consult and pain control. Cycle 1 FOLFIRINOX was started during this admission. A PICC line was again put in place and then removed before discharge. A celiac plexus block was performed several days after this admission for pain control.

Given concern about cocaine use increasing the risk of cardiac toxicity with FOLFIRINOX treatment, treating providers sconsulted with the community living center (CLC) about possible admission for future chemotherapy administration and pain management. The CLC at VACHS has 38 beds for rehabilitation, long-term care, and hospice with the mission to restore each veteran to his or her highest level of well-being. After discussion with this patient and CLC staff, he agreed to a CLC admission. The patient agreed to remain in the facility, wear a secure care device, and not leave without staff accompaniment. He was able to obtain a 2-hour pass to pay bills and rent. During the 2 months he was admitted to the CLC he would present to the VACHS Cancer Center for chemotherapy every 2 weeks. He completed 6 cycles of chemotherapy while admitted. During the admission, he was transferred to active medical service for 2 days for fever and malaise, and then returned to the CLC. The patient elected to leave the CLC after 2 months as the inability to see close friends was interfering with his quality of life.

Upon being discharged from the CLC, shared decision making took place with the patient to establish a new treatment plan. In collaboration with the patient, a plan was made to admit him every 2 weeks for continued chemotherapy. A PICC line was placed on each day of admission and removed prior to discharge. It was also agreed that treatment would be delayed if a urine drug test was positive for cocaine on the morning of admission. The patient was also seen by ORC every 2 weeks after being discharged from the CLC.

Imaging after cycle 6 showed decreased size of liver metastases, retroperitoneal lymph nodes, and pancreas mass. Cancer antigen 19-9 (CA19-9) tumor marker was reduced from 3513 U/mL pretreatment to 50 U/mL after cycle 7. Chemotherapy cycle 7 was delayed 6 days due to active cocaine and heroin use. A repeat urine was obtained several days later, which was negative for cocaine, and he was admitted for cycle 7 chemotherapy. Using this treatment approach of admissions for every cycle, the patient was able to receive 11 cycles of FOLFIRINOX with clinical benefit.

Palliative Care/Pain Management

Safely treating the patient’s malignant pain in the context of his OUD was critically important. In order to do this the palliative care team worked closely alongside ORC, is a multidisciplinary team consisting of health care providers (HCPs) from addiction psychiatry, internal medicine, health psychology and pharmacy who are consulted to evaluate veterans’ current opioid regimens and make recommendations to optimize both safety and efficacy. ORC followed this particular veteran as an outpatient and consulted on pain issues during his admission. They recommended the continuation of methadone at 120 mg daily and increased oral oxycodone to 30 mg every 6 hours, and then further increased to 45 mg every 6 hours. He continued to have increased pain despite higher doses of oxycodone, and pain medication was changed to oral hydromorphone 28 mg every 6 hours with the continuation of methadone. ORC and the palliative care team obtained consent from the veteran and a release of Information form signed by the patient to contact his community methadone clinic for further collaboration around pain management throughout the time caring for the veteran.

Even with improvement in disease based on imaging and tumor markers, opioid medications could not be decreased in this case. This is likely in part due to the multidimensional nature of pain. Careful assessment of the biologic, emotional, social, and spiritual contributors to pain is needed in the management of pain, especially at end of life.⁶ Nonpharmacologic pain management strategies used in this case included a transcutaneous electrical nerve stimulation unit, moist heat, celiac plexus block, and emotional support.

Psychosocial Issues/Substance Use

Psychosocial support for the patient was provided by the interdisciplinary palliative care team and the ORC team in both the inpatient and outpatient settings. Despite efforts from case management to get the veteran home services once discharged from the CLC, he declined repeatedly. Thus, the CLC social worker obtained a guardian alert for the veteran on discharge.

Close outpatient follow-up for medical and psychosocial support was very critical. When an outpatient, the veteran was scheduled for biweekly appointments with palliative care or ORC. When admitted to the hospital, the palliative care team medical director and psychologist conducted joint visits with him. Although he denied depressed mood and anxiety throughout his treatment, he often reflected on regrets that he had as he faced the end of his life. Specifically, he shared thoughts about being estranged from his surviving brother given his long struggle with substance use. Although he did not think a relationship was possible with his brother at the end of life, he still cared deeply for him and wanted to make him aware of his pancreatic cancer diagnosis. This was particularly important to him because their late brother had also died of pancreatic cancer. It was the patient’s wish at the end of his life to alert his surviving brother of his diagnosis so he and his children could get adequate screening throughout their lives. Although he had spoken of this desire often, it wasn’t until his disease progressed and he elected to transition to hospice that he felt ready to write the letter. The palliative care team assisted the veteran in writing and mailing a letter to his brother informing him of his diagnosis and transition to hospice as well as communicating that his brother and his family had been in his thoughts at the end of his life. The patient’s brother received this letter and with assistance from the CLC social worker made arrangements to visit the veteran at bedside at the inpatient CLC hospice unit the final days of his life.

Discussion

There are very little data on the safety of cancer-directed therapy in patients with active SUD. The limited studies that have been done showed conflicting results.

A retrospective study among women with co-occurring SUD and locally advanced cervical cancer who were undergoing primary radiation therapy found that SUD was not associated with a difference in toxicity or survival outcomes.⁷ However, other research suggests that SUD may be associated with an increase in all-cause mortality as well as other adverse outcomes for patients and health care systems (eg, emergency department visits, hospitalizations).⁸ A retrospective study of patients with a history of SUD and nonsmall cell lung cancer showed that these patients had higher rates of depression, less family support, increased rates of missed appointments, more emergency department visits and more hospitalizations.⁹ Patients with chronic myeloid leukemia or myelodysplastic syndromes who had long-term cocaine use had a 6-fold increased risk of death, which was not found in patients who had long-term alcohol or marijuana use.²

The limited data highlight the need for careful consideration of ways to mitigate potentially adverse outcomes in this population while still providing clinically indicated cancer treatment. Integrated VA health care systems provide unique resources that can maximize veteran safety during cancer treatment. Utilization of VA resources and close interdisciplinary collaboration across VA HCPs can help to ensure equitable access to state-of-the-art cancer therapies for veterans with comorbid SUD.

VA Services for Patients With Comorbidities

This case highlights several distinct aspects of VA health care that make it possible to safely treat individuals with complex comorbidities. One important aspect of this was collaboration with the CLC to admit the veteran for his initial treatment after a positive cocaine test. CLC admission was nonpunitive and allowed ongoing involvement in the VA community. This provided an essential, safe, and structured environment in which 6 cycles of chemotherapy could be delivered.

Although the patient left the CLC after 2 months due to floor restrictions negatively impacting his quality of life and ability to spend time with close friends, several important events occurred during this stay. First, the patient established close relationships with the CLC staff and the palliative care team; both groups followed him throughout his inpatient and outpatient care. These relationships proved essential throughout his care as they were the foundation of difficult conversations about substance use, treatment adherence, and eventually, transition to hospice.

In addition, the opportunity to administer 6 cycles of chemotherapy at the CLC was enough to lead to clinical benefit and radiographic response to treatment. Clinical benefits while in the CLC included maintenance of a good appetite, 15-lb weight gain and preserved performance status (ECOG [Eastern Cooperative Group]-1), which allowed him to actively participate in multiple social and recreational activities while in the CLC. From early conversations, this patient was clear that he wanted treatment as long as his life could be prolonged with good quality of life. Having evidence of the benefit of treatment, at least initially, increased the patient’s confidence in treatment. There were a few conversations when the challenges of treatment mounted (eg, pain, needs for abstinence from cocaine prior to admission for chemotherapy, frequent doctor appointments), and the patient would remind himself of these data to recommit himself to treatment. The opportunity to admit him to the inpatient VA facility, including bed availability for 3 days during his treatment once he left the CLC was important. This plan to admit the patient following a negative urine toxicology test for cocaine was made collaboratively with the veteran and the oncology and palliative care teams. The plan allowed the patient to achieve his treatment goals while maintaining his safety and reducing theoretical cardiac toxicities with his cancer treatment.

Finally, the availability of a multidisciplinary team approach including palliative care, oncology, psychology, addiction medicine and addiction psychiatry, was critical for addressing the veteran’s malignant pain. Palliative care worked in close collaboration with the ORC to prescribe and renew pain medications. ORC offered ongoing consultation on pain management in the context of OUD. As the veteran’s cancer progressed and functional decline prohibited his daily attendance at the community methadone clinic, palliative care and ORC met with the methadone clinic to arrange a less frequent methadone pickup schedule (the patient previously needed daily pickup). Non-VA settings may not have access to these resources to safely treat the biopsychosocial issues that arise in complex cases.

Substance Use and Cancer Treatments

This case raises several critical questions for oncologic care. Cocaine and fluorouracil are both associated with cardiotoxicity, and many oncologists would not feel it is safe to administer a regimen containing fluorouracil to a patient with active cocaine use. The National Comprehensive Cancer Network (NCCN) panel recommends FOLFIRINOX as a preferred category 1 recommendation for first-line treatment of patients with advanced pancreas cancer with good performance status.¹⁰ This recommendation is based on the PRODIGE trial, which has shown improved overall survival (OS): 11.1 vs 6.8 months for patients who received single-agent gemcitabine.¹¹ If patients are not candidates for FOLFIRINOX and have good performance status, the NCCN recommends gemcitabine plus albumin-bound paclitaxel with category 1 level of evidence based on the IMPACT trial, which showed improvement in OS (8.7 vs 6.6 months compared with single-agent gemcitabine).¹²

Some oncologists may have additional concerns administering fluorouracil treatment alternatives (such as gemcitabine and albumin-bound paclitaxel) to individuals with active SUD because of concerns about altered mental status impacting the ability to report important adverse effects. In the absence of sufficient data, HCPs must determine whether they feel it is safe to administer these agents in individuals with active cocaine use. However, denying these patients the possible benefits of standard-of-care life-prolonging therapies without established data raises concerns regarding the ethics of such practices. There is concern that the stigma surrounding cocaine use might contribute to withholding treatment, while treatment is continued for individuals taking prescribed stimulant medications that also have cardiotoxicity risks. VA health care facilities are uniquely situated to use all available resources to address these issues using interprofessional patient-centered care and determine the most optimal treatment based on a risk/benefit discussion between the patient and the HCP.

SUD Screening

This case begs the question of whether oncologists are adequately screening for a range of SUDs, and when they encounter an issue, how they are addressing it. Many oncologists do not receive adequate training on assessment of current or recent substance use. There are health care and systems-level practices that may increase patient safety for individuals with ongoing substance use who are undergoing cancer treatment. Training on obtaining appropriate substance use histories, motivational interviewing to resolve ambivalence about substance use in the direction of change, and shared decision making about treatment options could increase confidence in understanding and addressing substance use issues. It is also important to educate oncologists on how to address patients who return to or continued substance use during treatment. In this case the collaboration from palliative care, psychology, addiction medicine, and addiction psychiatry through the ORC was essential in assisting with ongoing assessment of substance use, guiding difficult conversations about the impact of substance use on the treatment plan, and identifying risk-mitigation strategies. Close collaboration and full utilization of all VA resources allowed this patient to receive first-line treatment for pancreatic cancer in order to reach his goal of prolonging his life while maintaining acceptable quality of life. Table 2 provides best practices for management of patients with comorbid SUD and cancer.

More research is needed into cancer treatment for patients with SUD, especially in the current era of cancer care using novel cancer treatments leading to significantly improved survival in many cancer types. Ideally, oncologists should be routinely or consistently screening patients for substance use, including alcohol. The patient should participate in this decision-making process after being educated about the risks and benefits. These patients can be followed using a multimodal approach to increase their rates of success and improve their quality of life. Although the literature is limited and no formal guidelines are available, VA oncologists are fortunate to have a range of resources available to them to navigate these difficult cases. Veterans have elevated rates of SUD, making this a critical issue to consider in the VA.¹³ It is the hope that this case can highlight how to take advantage of the many VA resources in order to ensure equitable cancer care for all veterans.

Conclusions

This case demonstrates that cancer-directed treatment is safe and feasible in a patient with advanced pancreatic cancer and coexisting active SUD by using a multidisciplinary approach. The multidisciplinary team included palliative care, oncology, psychology, addiction medicine, and addiction psychiatry. Critical steps for a successful outcome include gathering history about SUD; motivational interviewing to resolve ambivalence about treatment for SUD; shared decision making about cancer treatment; and risk-reduction strategies in pain and SUD management.

Treatment advancements in many cancer types have led to significantly longer survival, and it is critical to develop safe protocols to treat patients with active SUD so they also can derive benefit from these very significant medical advancements.

Acknowledgments

Michal Rose, MD, Director of VACHS Cancer Center, and Chandrika Kumar, MD, Director of VACHS Community Living Center, for their collaboration in care for this veteran.

Substance use disorders (SUDs) are an important but understudied aspect of treating patients diagnosed with cancer. Substance use can affect cancer treatment outcomes, including morbidity and mortality.^1,2 Additionally, patients with cancer and SUD may have unique psychosocial needs that require close attention and management. There is a paucity of data regarding the best approach to treating such patients. For example, cocaine use may increase the cardiovascular and hematologic risk of some traditional chemotherapy agents.^3,4 Newer targeted agents and immunotherapies remain understudied with respect to SUD risk.

Although the US Department of Veterans Affairs (VA) has established helpful clinical practice guidelines for the treatment of SUD, there are no guidelines for treating patients with SUD and cancer.⁵ Clinicians have limited confidence in treatment approach, and treatment is inconsistent among oncologists nationwide even within the same practice. Furthermore, it can be challenging to safely prescribe opioids for cancer-related pain in individuals with SUD. There is a high risk of SUD and mental health disorders in veterans, making this population particularly vulnerable. We report a case of a male with metastatic pancreatic cancer, severe opioid use disorder (OUD) and moderate cocaine use disorder (CUD) who received pain management and cancer treatment under the direction of a multidisciplinary team approach.

Case Report

A 63-year-old male with a medical history of HIV treated with highly active antiretroviral therapy (HAART), compensated cirrhosis, severe OUD, moderate CUD, and sedative use disorder in sustained remission was admitted to the West Haven campus of the VA Connecticut Healthcare System (VACHS) with abdominal pain, weight loss and fatigue. He used heroin 1 month prior to his admission and reported regular cocaine and marijuana use (Table 1). He was diagnosed with HIV in 1989, and his medical history included herpes zoster and oral candidiasis but no other opportunistic infections. Several months prior to this admission, he had an undetectable viral load and CD4 count of 688.

At the time of this admission, the patient was adherent to methadone treatment. He reported increased abdominal pain. Computed tomography (CT) showed a 2.4-cm mass in the pancreatic uncinate process, multiple liver metastases, retroperitoneal lymphadenopathy, and small lung nodules. A CT-guided liver biopsy showed adenocarcinoma consistent with a primary cancer of the pancreas. Given the complexity of the case, a multidisciplinary team approach was used to treat his cancer and the sequelae safely, including the oncology team, community living center team, palliative care team, and interprofessional opioid reassessment clinic team (ORC).

Cancer Treatment

Chemotherapy with FOLFIRINOX (leucovorin calcium, fluorouracil, irinotecan hydrochloride, and oxaliplatin) was recommended. The first cycle of treatment originally was planned for the outpatient setting, and a peripherally inserted central catheter (PICC) line was placed. However, after a urine toxicology test was positive for cocaine, the PICC line was removed due to concern for possible use of PICC line for nonprescribed substance use. The patient expressed suicidal ideation at the time and was admitted for psychiatric consult and pain control. Cycle 1 FOLFIRINOX was started during this admission. A PICC line was again put in place and then removed before discharge. A celiac plexus block was performed several days after this admission for pain control.

Given concern about cocaine use increasing the risk of cardiac toxicity with FOLFIRINOX treatment, treating providers sconsulted with the community living center (CLC) about possible admission for future chemotherapy administration and pain management. The CLC at VACHS has 38 beds for rehabilitation, long-term care, and hospice with the mission to restore each veteran to his or her highest level of well-being. After discussion with this patient and CLC staff, he agreed to a CLC admission. The patient agreed to remain in the facility, wear a secure care device, and not leave without staff accompaniment. He was able to obtain a 2-hour pass to pay bills and rent. During the 2 months he was admitted to the CLC he would present to the VACHS Cancer Center for chemotherapy every 2 weeks. He completed 6 cycles of chemotherapy while admitted. During the admission, he was transferred to active medical service for 2 days for fever and malaise, and then returned to the CLC. The patient elected to leave the CLC after 2 months as the inability to see close friends was interfering with his quality of life.

Upon being discharged from the CLC, shared decision making took place with the patient to establish a new treatment plan. In collaboration with the patient, a plan was made to admit him every 2 weeks for continued chemotherapy. A PICC line was placed on each day of admission and removed prior to discharge. It was also agreed that treatment would be delayed if a urine drug test was positive for cocaine on the morning of admission. The patient was also seen by ORC every 2 weeks after being discharged from the CLC.

Imaging after cycle 6 showed decreased size of liver metastases, retroperitoneal lymph nodes, and pancreas mass. Cancer antigen 19-9 (CA19-9) tumor marker was reduced from 3513 U/mL pretreatment to 50 U/mL after cycle 7. Chemotherapy cycle 7 was delayed 6 days due to active cocaine and heroin use. A repeat urine was obtained several days later, which was negative for cocaine, and he was admitted for cycle 7 chemotherapy. Using this treatment approach of admissions for every cycle, the patient was able to receive 11 cycles of FOLFIRINOX with clinical benefit.

Palliative Care/Pain Management

Safely treating the patient’s malignant pain in the context of his OUD was critically important. In order to do this the palliative care team worked closely alongside ORC, is a multidisciplinary team consisting of health care providers (HCPs) from addiction psychiatry, internal medicine, health psychology and pharmacy who are consulted to evaluate veterans’ current opioid regimens and make recommendations to optimize both safety and efficacy. ORC followed this particular veteran as an outpatient and consulted on pain issues during his admission. They recommended the continuation of methadone at 120 mg daily and increased oral oxycodone to 30 mg every 6 hours, and then further increased to 45 mg every 6 hours. He continued to have increased pain despite higher doses of oxycodone, and pain medication was changed to oral hydromorphone 28 mg every 6 hours with the continuation of methadone. ORC and the palliative care team obtained consent from the veteran and a release of Information form signed by the patient to contact his community methadone clinic for further collaboration around pain management throughout the time caring for the veteran.

Even with improvement in disease based on imaging and tumor markers, opioid medications could not be decreased in this case. This is likely in part due to the multidimensional nature of pain. Careful assessment of the biologic, emotional, social, and spiritual contributors to pain is needed in the management of pain, especially at end of life.⁶ Nonpharmacologic pain management strategies used in this case included a transcutaneous electrical nerve stimulation unit, moist heat, celiac plexus block, and emotional support.

Psychosocial Issues/Substance Use

Psychosocial support for the patient was provided by the interdisciplinary palliative care team and the ORC team in both the inpatient and outpatient settings. Despite efforts from case management to get the veteran home services once discharged from the CLC, he declined repeatedly. Thus, the CLC social worker obtained a guardian alert for the veteran on discharge.

Close outpatient follow-up for medical and psychosocial support was very critical. When an outpatient, the veteran was scheduled for biweekly appointments with palliative care or ORC. When admitted to the hospital, the palliative care team medical director and psychologist conducted joint visits with him. Although he denied depressed mood and anxiety throughout his treatment, he often reflected on regrets that he had as he faced the end of his life. Specifically, he shared thoughts about being estranged from his surviving brother given his long struggle with substance use. Although he did not think a relationship was possible with his brother at the end of life, he still cared deeply for him and wanted to make him aware of his pancreatic cancer diagnosis. This was particularly important to him because their late brother had also died of pancreatic cancer. It was the patient’s wish at the end of his life to alert his surviving brother of his diagnosis so he and his children could get adequate screening throughout their lives. Although he had spoken of this desire often, it wasn’t until his disease progressed and he elected to transition to hospice that he felt ready to write the letter. The palliative care team assisted the veteran in writing and mailing a letter to his brother informing him of his diagnosis and transition to hospice as well as communicating that his brother and his family had been in his thoughts at the end of his life. The patient’s brother received this letter and with assistance from the CLC social worker made arrangements to visit the veteran at bedside at the inpatient CLC hospice unit the final days of his life.

Discussion

There are very little data on the safety of cancer-directed therapy in patients with active SUD. The limited studies that have been done showed conflicting results.

A retrospective study among women with co-occurring SUD and locally advanced cervical cancer who were undergoing primary radiation therapy found that SUD was not associated with a difference in toxicity or survival outcomes.⁷ However, other research suggests that SUD may be associated with an increase in all-cause mortality as well as other adverse outcomes for patients and health care systems (eg, emergency department visits, hospitalizations).⁸ A retrospective study of patients with a history of SUD and nonsmall cell lung cancer showed that these patients had higher rates of depression, less family support, increased rates of missed appointments, more emergency department visits and more hospitalizations.⁹ Patients with chronic myeloid leukemia or myelodysplastic syndromes who had long-term cocaine use had a 6-fold increased risk of death, which was not found in patients who had long-term alcohol or marijuana use.²

The limited data highlight the need for careful consideration of ways to mitigate potentially adverse outcomes in this population while still providing clinically indicated cancer treatment. Integrated VA health care systems provide unique resources that can maximize veteran safety during cancer treatment. Utilization of VA resources and close interdisciplinary collaboration across VA HCPs can help to ensure equitable access to state-of-the-art cancer therapies for veterans with comorbid SUD.

VA Services for Patients With Comorbidities

This case highlights several distinct aspects of VA health care that make it possible to safely treat individuals with complex comorbidities. One important aspect of this was collaboration with the CLC to admit the veteran for his initial treatment after a positive cocaine test. CLC admission was nonpunitive and allowed ongoing involvement in the VA community. This provided an essential, safe, and structured environment in which 6 cycles of chemotherapy could be delivered.

Although the patient left the CLC after 2 months due to floor restrictions negatively impacting his quality of life and ability to spend time with close friends, several important events occurred during this stay. First, the patient established close relationships with the CLC staff and the palliative care team; both groups followed him throughout his inpatient and outpatient care. These relationships proved essential throughout his care as they were the foundation of difficult conversations about substance use, treatment adherence, and eventually, transition to hospice.

In addition, the opportunity to administer 6 cycles of chemotherapy at the CLC was enough to lead to clinical benefit and radiographic response to treatment. Clinical benefits while in the CLC included maintenance of a good appetite, 15-lb weight gain and preserved performance status (ECOG [Eastern Cooperative Group]-1), which allowed him to actively participate in multiple social and recreational activities while in the CLC. From early conversations, this patient was clear that he wanted treatment as long as his life could be prolonged with good quality of life. Having evidence of the benefit of treatment, at least initially, increased the patient’s confidence in treatment. There were a few conversations when the challenges of treatment mounted (eg, pain, needs for abstinence from cocaine prior to admission for chemotherapy, frequent doctor appointments), and the patient would remind himself of these data to recommit himself to treatment. The opportunity to admit him to the inpatient VA facility, including bed availability for 3 days during his treatment once he left the CLC was important. This plan to admit the patient following a negative urine toxicology test for cocaine was made collaboratively with the veteran and the oncology and palliative care teams. The plan allowed the patient to achieve his treatment goals while maintaining his safety and reducing theoretical cardiac toxicities with his cancer treatment.

Finally, the availability of a multidisciplinary team approach including palliative care, oncology, psychology, addiction medicine and addiction psychiatry, was critical for addressing the veteran’s malignant pain. Palliative care worked in close collaboration with the ORC to prescribe and renew pain medications. ORC offered ongoing consultation on pain management in the context of OUD. As the veteran’s cancer progressed and functional decline prohibited his daily attendance at the community methadone clinic, palliative care and ORC met with the methadone clinic to arrange a less frequent methadone pickup schedule (the patient previously needed daily pickup). Non-VA settings may not have access to these resources to safely treat the biopsychosocial issues that arise in complex cases.

Substance Use and Cancer Treatments

This case raises several critical questions for oncologic care. Cocaine and fluorouracil are both associated with cardiotoxicity, and many oncologists would not feel it is safe to administer a regimen containing fluorouracil to a patient with active cocaine use. The National Comprehensive Cancer Network (NCCN) panel recommends FOLFIRINOX as a preferred category 1 recommendation for first-line treatment of patients with advanced pancreas cancer with good performance status.¹⁰ This recommendation is based on the PRODIGE trial, which has shown improved overall survival (OS): 11.1 vs 6.8 months for patients who received single-agent gemcitabine.¹¹ If patients are not candidates for FOLFIRINOX and have good performance status, the NCCN recommends gemcitabine plus albumin-bound paclitaxel with category 1 level of evidence based on the IMPACT trial, which showed improvement in OS (8.7 vs 6.6 months compared with single-agent gemcitabine).¹²

Some oncologists may have additional concerns administering fluorouracil treatment alternatives (such as gemcitabine and albumin-bound paclitaxel) to individuals with active SUD because of concerns about altered mental status impacting the ability to report important adverse effects. In the absence of sufficient data, HCPs must determine whether they feel it is safe to administer these agents in individuals with active cocaine use. However, denying these patients the possible benefits of standard-of-care life-prolonging therapies without established data raises concerns regarding the ethics of such practices. There is concern that the stigma surrounding cocaine use might contribute to withholding treatment, while treatment is continued for individuals taking prescribed stimulant medications that also have cardiotoxicity risks. VA health care facilities are uniquely situated to use all available resources to address these issues using interprofessional patient-centered care and determine the most optimal treatment based on a risk/benefit discussion between the patient and the HCP.

SUD Screening

This case begs the question of whether oncologists are adequately screening for a range of SUDs, and when they encounter an issue, how they are addressing it. Many oncologists do not receive adequate training on assessment of current or recent substance use. There are health care and systems-level practices that may increase patient safety for individuals with ongoing substance use who are undergoing cancer treatment. Training on obtaining appropriate substance use histories, motivational interviewing to resolve ambivalence about substance use in the direction of change, and shared decision making about treatment options could increase confidence in understanding and addressing substance use issues. It is also important to educate oncologists on how to address patients who return to or continued substance use during treatment. In this case the collaboration from palliative care, psychology, addiction medicine, and addiction psychiatry through the ORC was essential in assisting with ongoing assessment of substance use, guiding difficult conversations about the impact of substance use on the treatment plan, and identifying risk-mitigation strategies. Close collaboration and full utilization of all VA resources allowed this patient to receive first-line treatment for pancreatic cancer in order to reach his goal of prolonging his life while maintaining acceptable quality of life. Table 2 provides best practices for management of patients with comorbid SUD and cancer.

More research is needed into cancer treatment for patients with SUD, especially in the current era of cancer care using novel cancer treatments leading to significantly improved survival in many cancer types. Ideally, oncologists should be routinely or consistently screening patients for substance use, including alcohol. The patient should participate in this decision-making process after being educated about the risks and benefits. These patients can be followed using a multimodal approach to increase their rates of success and improve their quality of life. Although the literature is limited and no formal guidelines are available, VA oncologists are fortunate to have a range of resources available to them to navigate these difficult cases. Veterans have elevated rates of SUD, making this a critical issue to consider in the VA.¹³ It is the hope that this case can highlight how to take advantage of the many VA resources in order to ensure equitable cancer care for all veterans.

Conclusions

This case demonstrates that cancer-directed treatment is safe and feasible in a patient with advanced pancreatic cancer and coexisting active SUD by using a multidisciplinary approach. The multidisciplinary team included palliative care, oncology, psychology, addiction medicine, and addiction psychiatry. Critical steps for a successful outcome include gathering history about SUD; motivational interviewing to resolve ambivalence about treatment for SUD; shared decision making about cancer treatment; and risk-reduction strategies in pain and SUD management.

Treatment advancements in many cancer types have led to significantly longer survival, and it is critical to develop safe protocols to treat patients with active SUD so they also can derive benefit from these very significant medical advancements.

Acknowledgments

Michal Rose, MD, Director of VACHS Cancer Center, and Chandrika Kumar, MD, Director of VACHS Community Living Center, for their collaboration in care for this veteran.

Substance use disorders (SUDs) are an important but understudied aspect of treating patients diagnosed with cancer. Substance use can affect cancer treatment outcomes, including morbidity and mortality.^1,2 Additionally, patients with cancer and SUD may have unique psychosocial needs that require close attention and management. There is a paucity of data regarding the best approach to treating such patients. For example, cocaine use may increase the cardiovascular and hematologic risk of some traditional chemotherapy agents.^3,4 Newer targeted agents and immunotherapies remain understudied with respect to SUD risk.

Although the US Department of Veterans Affairs (VA) has established helpful clinical practice guidelines for the treatment of SUD, there are no guidelines for treating patients with SUD and cancer.⁵ Clinicians have limited confidence in treatment approach, and treatment is inconsistent among oncologists nationwide even within the same practice. Furthermore, it can be challenging to safely prescribe opioids for cancer-related pain in individuals with SUD. There is a high risk of SUD and mental health disorders in veterans, making this population particularly vulnerable. We report a case of a male with metastatic pancreatic cancer, severe opioid use disorder (OUD) and moderate cocaine use disorder (CUD) who received pain management and cancer treatment under the direction of a multidisciplinary team approach.

Case Report

A 63-year-old male with a medical history of HIV treated with highly active antiretroviral therapy (HAART), compensated cirrhosis, severe OUD, moderate CUD, and sedative use disorder in sustained remission was admitted to the West Haven campus of the VA Connecticut Healthcare System (VACHS) with abdominal pain, weight loss and fatigue. He used heroin 1 month prior to his admission and reported regular cocaine and marijuana use (Table 1). He was diagnosed with HIV in 1989, and his medical history included herpes zoster and oral candidiasis but no other opportunistic infections. Several months prior to this admission, he had an undetectable viral load and CD4 count of 688.

At the time of this admission, the patient was adherent to methadone treatment. He reported increased abdominal pain. Computed tomography (CT) showed a 2.4-cm mass in the pancreatic uncinate process, multiple liver metastases, retroperitoneal lymphadenopathy, and small lung nodules. A CT-guided liver biopsy showed adenocarcinoma consistent with a primary cancer of the pancreas. Given the complexity of the case, a multidisciplinary team approach was used to treat his cancer and the sequelae safely, including the oncology team, community living center team, palliative care team, and interprofessional opioid reassessment clinic team (ORC).

Cancer Treatment

Chemotherapy with FOLFIRINOX (leucovorin calcium, fluorouracil, irinotecan hydrochloride, and oxaliplatin) was recommended. The first cycle of treatment originally was planned for the outpatient setting, and a peripherally inserted central catheter (PICC) line was placed. However, after a urine toxicology test was positive for cocaine, the PICC line was removed due to concern for possible use of PICC line for nonprescribed substance use. The patient expressed suicidal ideation at the time and was admitted for psychiatric consult and pain control. Cycle 1 FOLFIRINOX was started during this admission. A PICC line was again put in place and then removed before discharge. A celiac plexus block was performed several days after this admission for pain control.

Given concern about cocaine use increasing the risk of cardiac toxicity with FOLFIRINOX treatment, treating providers sconsulted with the community living center (CLC) about possible admission for future chemotherapy administration and pain management. The CLC at VACHS has 38 beds for rehabilitation, long-term care, and hospice with the mission to restore each veteran to his or her highest level of well-being. After discussion with this patient and CLC staff, he agreed to a CLC admission. The patient agreed to remain in the facility, wear a secure care device, and not leave without staff accompaniment. He was able to obtain a 2-hour pass to pay bills and rent. During the 2 months he was admitted to the CLC he would present to the VACHS Cancer Center for chemotherapy every 2 weeks. He completed 6 cycles of chemotherapy while admitted. During the admission, he was transferred to active medical service for 2 days for fever and malaise, and then returned to the CLC. The patient elected to leave the CLC after 2 months as the inability to see close friends was interfering with his quality of life.

Upon being discharged from the CLC, shared decision making took place with the patient to establish a new treatment plan. In collaboration with the patient, a plan was made to admit him every 2 weeks for continued chemotherapy. A PICC line was placed on each day of admission and removed prior to discharge. It was also agreed that treatment would be delayed if a urine drug test was positive for cocaine on the morning of admission. The patient was also seen by ORC every 2 weeks after being discharged from the CLC.

Imaging after cycle 6 showed decreased size of liver metastases, retroperitoneal lymph nodes, and pancreas mass. Cancer antigen 19-9 (CA19-9) tumor marker was reduced from 3513 U/mL pretreatment to 50 U/mL after cycle 7. Chemotherapy cycle 7 was delayed 6 days due to active cocaine and heroin use. A repeat urine was obtained several days later, which was negative for cocaine, and he was admitted for cycle 7 chemotherapy. Using this treatment approach of admissions for every cycle, the patient was able to receive 11 cycles of FOLFIRINOX with clinical benefit.

Palliative Care/Pain Management

Safely treating the patient’s malignant pain in the context of his OUD was critically important. In order to do this the palliative care team worked closely alongside ORC, is a multidisciplinary team consisting of health care providers (HCPs) from addiction psychiatry, internal medicine, health psychology and pharmacy who are consulted to evaluate veterans’ current opioid regimens and make recommendations to optimize both safety and efficacy. ORC followed this particular veteran as an outpatient and consulted on pain issues during his admission. They recommended the continuation of methadone at 120 mg daily and increased oral oxycodone to 30 mg every 6 hours, and then further increased to 45 mg every 6 hours. He continued to have increased pain despite higher doses of oxycodone, and pain medication was changed to oral hydromorphone 28 mg every 6 hours with the continuation of methadone. ORC and the palliative care team obtained consent from the veteran and a release of Information form signed by the patient to contact his community methadone clinic for further collaboration around pain management throughout the time caring for the veteran.

Even with improvement in disease based on imaging and tumor markers, opioid medications could not be decreased in this case. This is likely in part due to the multidimensional nature of pain. Careful assessment of the biologic, emotional, social, and spiritual contributors to pain is needed in the management of pain, especially at end of life.⁶ Nonpharmacologic pain management strategies used in this case included a transcutaneous electrical nerve stimulation unit, moist heat, celiac plexus block, and emotional support.

Psychosocial Issues/Substance Use

Psychosocial support for the patient was provided by the interdisciplinary palliative care team and the ORC team in both the inpatient and outpatient settings. Despite efforts from case management to get the veteran home services once discharged from the CLC, he declined repeatedly. Thus, the CLC social worker obtained a guardian alert for the veteran on discharge.

Close outpatient follow-up for medical and psychosocial support was very critical. When an outpatient, the veteran was scheduled for biweekly appointments with palliative care or ORC. When admitted to the hospital, the palliative care team medical director and psychologist conducted joint visits with him. Although he denied depressed mood and anxiety throughout his treatment, he often reflected on regrets that he had as he faced the end of his life. Specifically, he shared thoughts about being estranged from his surviving brother given his long struggle with substance use. Although he did not think a relationship was possible with his brother at the end of life, he still cared deeply for him and wanted to make him aware of his pancreatic cancer diagnosis. This was particularly important to him because their late brother had also died of pancreatic cancer. It was the patient’s wish at the end of his life to alert his surviving brother of his diagnosis so he and his children could get adequate screening throughout their lives. Although he had spoken of this desire often, it wasn’t until his disease progressed and he elected to transition to hospice that he felt ready to write the letter. The palliative care team assisted the veteran in writing and mailing a letter to his brother informing him of his diagnosis and transition to hospice as well as communicating that his brother and his family had been in his thoughts at the end of his life. The patient’s brother received this letter and with assistance from the CLC social worker made arrangements to visit the veteran at bedside at the inpatient CLC hospice unit the final days of his life.

Discussion

There are very little data on the safety of cancer-directed therapy in patients with active SUD. The limited studies that have been done showed conflicting results.

A retrospective study among women with co-occurring SUD and locally advanced cervical cancer who were undergoing primary radiation therapy found that SUD was not associated with a difference in toxicity or survival outcomes.⁷ However, other research suggests that SUD may be associated with an increase in all-cause mortality as well as other adverse outcomes for patients and health care systems (eg, emergency department visits, hospitalizations).⁸ A retrospective study of patients with a history of SUD and nonsmall cell lung cancer showed that these patients had higher rates of depression, less family support, increased rates of missed appointments, more emergency department visits and more hospitalizations.⁹ Patients with chronic myeloid leukemia or myelodysplastic syndromes who had long-term cocaine use had a 6-fold increased risk of death, which was not found in patients who had long-term alcohol or marijuana use.²

The limited data highlight the need for careful consideration of ways to mitigate potentially adverse outcomes in this population while still providing clinically indicated cancer treatment. Integrated VA health care systems provide unique resources that can maximize veteran safety during cancer treatment. Utilization of VA resources and close interdisciplinary collaboration across VA HCPs can help to ensure equitable access to state-of-the-art cancer therapies for veterans with comorbid SUD.

VA Services for Patients With Comorbidities

This case highlights several distinct aspects of VA health care that make it possible to safely treat individuals with complex comorbidities. One important aspect of this was collaboration with the CLC to admit the veteran for his initial treatment after a positive cocaine test. CLC admission was nonpunitive and allowed ongoing involvement in the VA community. This provided an essential, safe, and structured environment in which 6 cycles of chemotherapy could be delivered.

Although the patient left the CLC after 2 months due to floor restrictions negatively impacting his quality of life and ability to spend time with close friends, several important events occurred during this stay. First, the patient established close relationships with the CLC staff and the palliative care team; both groups followed him throughout his inpatient and outpatient care. These relationships proved essential throughout his care as they were the foundation of difficult conversations about substance use, treatment adherence, and eventually, transition to hospice.

In addition, the opportunity to administer 6 cycles of chemotherapy at the CLC was enough to lead to clinical benefit and radiographic response to treatment. Clinical benefits while in the CLC included maintenance of a good appetite, 15-lb weight gain and preserved performance status (ECOG [Eastern Cooperative Group]-1), which allowed him to actively participate in multiple social and recreational activities while in the CLC. From early conversations, this patient was clear that he wanted treatment as long as his life could be prolonged with good quality of life. Having evidence of the benefit of treatment, at least initially, increased the patient’s confidence in treatment. There were a few conversations when the challenges of treatment mounted (eg, pain, needs for abstinence from cocaine prior to admission for chemotherapy, frequent doctor appointments), and the patient would remind himself of these data to recommit himself to treatment. The opportunity to admit him to the inpatient VA facility, including bed availability for 3 days during his treatment once he left the CLC was important. This plan to admit the patient following a negative urine toxicology test for cocaine was made collaboratively with the veteran and the oncology and palliative care teams. The plan allowed the patient to achieve his treatment goals while maintaining his safety and reducing theoretical cardiac toxicities with his cancer treatment.

Finally, the availability of a multidisciplinary team approach including palliative care, oncology, psychology, addiction medicine and addiction psychiatry, was critical for addressing the veteran’s malignant pain. Palliative care worked in close collaboration with the ORC to prescribe and renew pain medications. ORC offered ongoing consultation on pain management in the context of OUD. As the veteran’s cancer progressed and functional decline prohibited his daily attendance at the community methadone clinic, palliative care and ORC met with the methadone clinic to arrange a less frequent methadone pickup schedule (the patient previously needed daily pickup). Non-VA settings may not have access to these resources to safely treat the biopsychosocial issues that arise in complex cases.

Substance Use and Cancer Treatments

This case raises several critical questions for oncologic care. Cocaine and fluorouracil are both associated with cardiotoxicity, and many oncologists would not feel it is safe to administer a regimen containing fluorouracil to a patient with active cocaine use. The National Comprehensive Cancer Network (NCCN) panel recommends FOLFIRINOX as a preferred category 1 recommendation for first-line treatment of patients with advanced pancreas cancer with good performance status.¹⁰ This recommendation is based on the PRODIGE trial, which has shown improved overall survival (OS): 11.1 vs 6.8 months for patients who received single-agent gemcitabine.¹¹ If patients are not candidates for FOLFIRINOX and have good performance status, the NCCN recommends gemcitabine plus albumin-bound paclitaxel with category 1 level of evidence based on the IMPACT trial, which showed improvement in OS (8.7 vs 6.6 months compared with single-agent gemcitabine).¹²

Some oncologists may have additional concerns administering fluorouracil treatment alternatives (such as gemcitabine and albumin-bound paclitaxel) to individuals with active SUD because of concerns about altered mental status impacting the ability to report important adverse effects. In the absence of sufficient data, HCPs must determine whether they feel it is safe to administer these agents in individuals with active cocaine use. However, denying these patients the possible benefits of standard-of-care life-prolonging therapies without established data raises concerns regarding the ethics of such practices. There is concern that the stigma surrounding cocaine use might contribute to withholding treatment, while treatment is continued for individuals taking prescribed stimulant medications that also have cardiotoxicity risks. VA health care facilities are uniquely situated to use all available resources to address these issues using interprofessional patient-centered care and determine the most optimal treatment based on a risk/benefit discussion between the patient and the HCP.

SUD Screening

This case begs the question of whether oncologists are adequately screening for a range of SUDs, and when they encounter an issue, how they are addressing it. Many oncologists do not receive adequate training on assessment of current or recent substance use. There are health care and systems-level practices that may increase patient safety for individuals with ongoing substance use who are undergoing cancer treatment. Training on obtaining appropriate substance use histories, motivational interviewing to resolve ambivalence about substance use in the direction of change, and shared decision making about treatment options could increase confidence in understanding and addressing substance use issues. It is also important to educate oncologists on how to address patients who return to or continued substance use during treatment. In this case the collaboration from palliative care, psychology, addiction medicine, and addiction psychiatry through the ORC was essential in assisting with ongoing assessment of substance use, guiding difficult conversations about the impact of substance use on the treatment plan, and identifying risk-mitigation strategies. Close collaboration and full utilization of all VA resources allowed this patient to receive first-line treatment for pancreatic cancer in order to reach his goal of prolonging his life while maintaining acceptable quality of life. Table 2 provides best practices for management of patients with comorbid SUD and cancer.

More research is needed into cancer treatment for patients with SUD, especially in the current era of cancer care using novel cancer treatments leading to significantly improved survival in many cancer types. Ideally, oncologists should be routinely or consistently screening patients for substance use, including alcohol. The patient should participate in this decision-making process after being educated about the risks and benefits. These patients can be followed using a multimodal approach to increase their rates of success and improve their quality of life. Although the literature is limited and no formal guidelines are available, VA oncologists are fortunate to have a range of resources available to them to navigate these difficult cases. Veterans have elevated rates of SUD, making this a critical issue to consider in the VA.¹³ It is the hope that this case can highlight how to take advantage of the many VA resources in order to ensure equitable cancer care for all veterans.

Conclusions

This case demonstrates that cancer-directed treatment is safe and feasible in a patient with advanced pancreatic cancer and coexisting active SUD by using a multidisciplinary approach. The multidisciplinary team included palliative care, oncology, psychology, addiction medicine, and addiction psychiatry. Critical steps for a successful outcome include gathering history about SUD; motivational interviewing to resolve ambivalence about treatment for SUD; shared decision making about cancer treatment; and risk-reduction strategies in pain and SUD management.

Treatment advancements in many cancer types have led to significantly longer survival, and it is critical to develop safe protocols to treat patients with active SUD so they also can derive benefit from these very significant medical advancements.

Acknowledgments

Michal Rose, MD, Director of VACHS Cancer Center, and Chandrika Kumar, MD, Director of VACHS Community Living Center, for their collaboration in care for this veteran.

The COVID-19 outbreak in February 2020 at a congregate living facility near Seattle, Washington, signaled the beginning of the pandemic in the United States. In that facility, infected residents had a 54.5% hospitalization rate and 33.7% case-fatality rate.¹ Similar to the experience in Washington, all congregate living facilities have proved particularly vulnerable to the effects of COVID-19,^2-7 with residents at increased risk for disease severity and mortality.^2-7

Due to the COVID-19 emergency, NorthShore University HealthSystem (NUHS), a multihospital, integrated health system in northern Illinois, established a best practice for appropriate use of congregate living facilities after hospitalization. This focused on the safety of discharged patients and mitigation of COVID-19 by putting in place a referral process to a newly established congregate living review committee (CLRC) for review prior to discharge. Although all discharges to congregate living settings are at high risk,² new placements to skilled nursing facilities (SNFs) were the primary focus of the committee and the sole focus of this study. In this study, we sought to determine whether establishment of the CLRC was associated with a reduction in SNF utilization, whether this was safe and efficient, and whether it was associated with a reduction in COVID-19 incidence in the 30 days following discharge.

Setting and Case Review Intervention

We conducted a retrospective cohort study for patients hospitalized within NUHS from March 19, 2019 to July 16, 2020, designed as an interrupted time series. The study was approved by the NUHS Institutional Review Board (EH21-022).

The study exposure was creation of a referral and review process for all patients with expected discharge to a SNF and was implemented as part of usual discharge planning during the COVID-19 pandemic. The key intervention was to establish a multidisciplinary committee, the CLRC, to review all potential discharges to SNFs. The CLRC had dual goals of preventing COVID-19 spread in facilities by limiting placement of new residents and protecting a vulnerable population from a setting that conferred a higher risk of acquiring COVID-19. The CLRC was organized as a multidisciplinary committee with physicians, case managers, social workers, physical therapists, occupational therapists, and the director of NUHS home health agency. Physician members were evenly split as half hospitalists and half ambulatory physicians. The CLRC review was initiated by a patient’s assigned case manager or social worker by consult through a referral in the electronic medical record (EMR). Each case was summarized and then presented to the full CLRC. The CLRC met for 1 hour per day, 6 days per week, to review all planned discharges that met criteria for review. A committee physician chaired each meeting. Three other members were needed for a quorum, with one other member with a title of director or higher. Time required was the 1-hour daily meeting, as well as one full-time position for case review, preparation, and program administration. The case presentation included a clinical summary of the hospitalization as well as COVID-19 status and testing history, previous living situation, level of home support, functional level, psychosocial needs, barrier(s) to discharging home, and long-term residential plans. A structured assessment was then made by each CLRC member in accordance with their professional expertise. Unanimous consensus would be reached before finalizing any recommended adjustments to the discharge, which would be communicated to the inpatient care team via a structured note within the EMR, along with direct communication to the assigned case manager or social worker. When the CLRC suggested adjustments to the discharge, they would work with the assigned case manager or social worker to communicate an appropriate post–acute care plan with the patient or appropriate representative. If there was disagreement or the recommendations could not be followed, the case manager or social worker would place a new referral with additional information for reconsideration. Following a recommendation for SNF, verification would be completed by the CLRC prior to discharge. This process is detailed in Figure 1.

Patient Population

Inclusion criteria for the study were: (1) inpatient hospitalization and (2) eligibility for risk scoring via the organization’s clinical analytics prediction engine (CAPE).⁸ CAPE is a validated predictive model that includes risk of readmission, in-hospital mortality, and out-of-hospital mortality,⁸ with extensive adoption at NUHS. CAPE score eligibility was used as an inclusion criterion so that CAPE could be applied for derivation of a matched control. CAPE eligibility criteria include admission age of at least 18 years and that hospitalization is not psychiatric, rehabilitative, or obstetric. Patients must not be enrolled in hospice and must be discharged alive.

Exclusions were patients who tested positive for SARS-CoV-2 prior to or during index hospitalization. Excluding COVID-19 patients from the analysis eliminated a confounder not present in the preintervention group.

For patients with multiple inpatient admissions, the first admission was the only admission used for analysis. Additionally, if a patient had an admission that occurred in both the preintervention and postintervention periods, they were included only in the postintervention period. This was done to avoid any within-subject correlation and ensure unique patients in each group. Confounding from this approach was mitigated through the process of deriving a matched control.

Outcomes Measurement

The primary outcome of interest was total discharges to SNF across NUHS facilities after hospital admission. Patients were identified as discharging to a SNF if discharge destination codes 03, 64, or 83 appeared on the hospital bill. Additionally, new discharges to SNFs were assessed and identified if documentation indicated that the patient’s living arrangement prior to admission was not a SNF but discharge billing destination codes 03, 64, or 83 appeared on the hospital bill.

Secondary outcomes were measurement of readmissions, days to readmission, and median length of stay (LOS). Readmissions and LOS were balancing measures for the primary outcome, with readmissions measured to evaluate the safety of the CLRC process and LOS measured to evaluate its efficiency. A readmission was any patient who had an unplanned inpatient admission at an NUHS facility within 30 days after an index admission. LOS was measured in days from arrival on a hospital unit to time of discharge.

Additional analysis was done to estimate the effect of the intervention on the incidence of COVID-19 in the 30 days following discharge by comparing the observed to expected incidence of COVID-19 by discharge destination. The expected values were derived by estimating COVID-19 cases that would have been expected to occur with rates of preintervention SNF utilization. This was accomplished by multiplying the observed incidence of COVID-19 in the 30 days following discharge by the number of patients who were discharged to SNFs or home/other in the preintervention period. This expected value was then compared with the observed values to estimate the effect size of the intervention on COVID-19 incidence following discharge. This method of deriving an expected value from the observed incidence was utilized because the preintervention period was before COVID-19 was widespread in the community. It was therefore not possible to directly measure COVID-19 incidence in the preintervention period.

Data Source

Data were retrieved from the NUHS Enterprise Data Warehouse, NUHS’s central data repository, which contains a nightly upload of clinical and financial data from the EMR. Data were collected between March 19, 2019, and July 16, 2020.

The preintervention period was defined as March 19, 2019, to March 18, 2020. Data from that interval were compared with the postintervention period, which was from March 19, 2020, to July 16, 2020. The preintervention period, 1 year immediately prior to the intervention, was chosen to limit any effect of temporal trends while also providing a large sample size. The postintervention period began on the first day NUHS implemented the revised approach to SNF use and ended on the last day before the review process was modified.

Data Analysis

An interrupted time series was used to measure the impact of adoption of the CLRC protocol. A matched control was derived from the preintervention population. To derive this matched control, there was an assessment of covariates in the preintervention and postintervention groups using a standardized mean difference (SMD)⁹ that indicated an imbalance (SMD ≥ 0.1) in some covariates. A propensity score–matching technique¹⁰ was applied to address this imbalance and lack of randomization.

The candidate variables for propensity matching were chosen if they had an association with 30-day readmission. Readmission was chosen to find candidate variables because, of the possible outcomes, this was the only one that was not directly impacted by any CLRC decision. Each covariate was assessed using a logistic regression model while controlling for the postintervention group. If there was an association between a covariate and the outcome, it was chosen for propensity matching. Propensity scores were calculated using a logistic regression model with the treatment (1/0) variable as the dependent variable and the chosen covariates as predictors.

There were no indications of strong multicollinearity. The propensity scores generated were then used to derive a matched control using paired matching. MatchIt package in R (R Foundation for Statistical Computing) was used to create a matched dataset with a logit distance and standard caliper of 0.2 times the standard deviations of the logit of the propensity score. If a match was not found within the caliper, the nearest available match was used.

Regression adjustment¹¹ was then performed using multivariate linear/logistic regression with LOS, readmission rate, days to readmission, total SNF discharges, and new SNF discharges as the outcomes. Treatment (1/0) variable and propensity score were used as the predictors. The adjusted coefficients or odds ratios (ORs) of the intervention variable were thus derived, and their associated P values were used to assess the impact of the intervention on the respective outcomes.

The unmatched preintervention population included 14,468 patients, with 4424 patients in the postintervention population. A matched population was derived and, after matching, the population sizes for pre and post intervention were 4424 each. In the matched population, all measured preintervention characteristics had SMDs and P values that were statistically equivalent. Patient characteristics for the unmatched and matched populations are detailed in Table 1.

During the preintervention period, 1130 (25.5%) patients were discharged to a SNF, with 776 (17.5%) patients being new SNF discharges. In the postintervention period, 568 (12.8%) patients were discharged to a SNF, with 257 (5.8%) patients being new SNF discharges. Total SNF discharges postintervention saw a 49.7% relative reduction (OR, 0.42; 95% CI, 0.38-0.47), while new SNF discharges saw a 66.9% relative reduction (OR, 0.29; 95% CI, 0.25-0.34). These results for both total and new SNF discharges were statistically significant, with P values of <.001, respectively.

Readmissions in the preintervention period were 529 (12.0%) patients, compared with 559 (12.6%) patients in the postintervention period (OR, 1.06; 95% CI, 0.93-1.20; P =.406). An OR was also calculated for readmissions, adjusting for discharge disposition, to account for changes observed in SNF use in the postintervention period. This OR was 1.11 (95% CI, 0.97-1.26; P = .131). Days to readmission in the preintervention and postintervention groups were 11.0 days and 12.0 days, respectively (OR, 0.41; 95% CI, –0.61 to 1.43; P = .429).

LOS was 3.61 days in the preintervention group and 3.64 days in the postintervention group, with an interquartile range (IQR) of 2.14 to 5.69 days in the preintervention group and 2.08 to 5.95 in the postintervention group (OR, 0.09; 95% CI, –0.09 to 0.27; P =.316). These results are summarized in Table 2.

A COVID-19 outbreak in a SNF presents a grave risk to residents and patients discharged to these facilities. It is critical for healthcare systems to do the utmost to protect the health of this vulnerable population and the public in efforts to limit COVID-19 within SNFs.^12-14

In this study, we observed that at NUHS, establishing a multidisciplinary review committee, the CLRC, to assess the appropriateness of discharge to a SNF after hospitalization resulted in a nearly 50% reduction in total SNF discharges and a greater than two-thirds reduction in new SNF discharges, without any increase in LOS or readmissions. Additionally, it was observed that discharging to settings other than a SNF greatly reduced a patient’s risk of being diagnosed with COVID-19 within 30 days, a result that reached statistical significance. Based on the observed 37.2% relative reduction in COVID-19 cases, we estimate that there may have been one COVID-19 infection prevented every 5.6 days from this intervention. Based on published COVID-19 mortality rates for SNF residents,¹ the intervention may have prevented one death every 2.6 weeks. Beyond the risk of COVID-19, other benefits of reducing SNF use are patient and family well-being. Although not measured in this study, others have published about the significant psychological burdens placed on SNF residents, who were at high risk for social isolation, anxiety, and depression during the COVID-19 pandemic^2,15-19 Family members also may have had increased stress, as they were deprived of the opportunity to visit loved ones, advocate for them, and help maintain their identity, humanity, and quality of life.²⁰

Although other hospitals have established a structured approach to reduce COVID-19 in SNFs,²¹ to the best of the authors’ knowledge, the approach described in this article is a unique response to the COVID-19 pandemic. As we have demonstrated, it is highly effective and safe and likely prevented many COVID-19 cases and deaths.

Furthermore, a review committee, such as the one we have described, has value well beyond the COVID-19 pandemic. The health and affordability of care for patients, provider success in value-based care models, and the long-term sustainability of the US healthcare system require close attention to appropriate use of expensive services and to ensuring that their use creates high value. SNF use after a hospitalization is one such service that is frequently targeted and thought to contribute to a substantial portion of wasteful medical spending.^22,23 Additionally, SNFs are known to be high risk for communicable disease outbreaks other than COVID-19,^24,25 as well as a high-risk environment for many other preventable adverse events.^25,26 This review committee ultimately serves to help determine the most appropriate postacute setting for patients being discharged with a determination made through considerations for patient safety, rehabilitation potential, and mental and physical well-being. From a population health perspective, this can lead to better outcomes and lower costs.^22,23 Therefore, although the risks of COVID-19 infection in SNFs are expected to subside, the work of evaluating appropriate use of SNFs after hospitalization at our institution continues. The broader focus now extends beyond postacute level of service toward ensuring a high-value discharge that results in both appropriate resource use and safe patient care transitions.

Limitations of this study include its retrospective nature, results from a single center, and a number of potentially unmeasured confounders that the COVID-19 pandemic created. One possible confounder is that the reduction in SNF use we observed was a temporal trend related to changing preferences. In addressing this, we reviewed Medicare claims data from the US Department of Health and Human Services in April 2020 and July 2020 compared with the same period in 2019. These data demonstrated only a modest reduction in spending on SNFs in April 2020 that was smaller than the reduction seen in Part A inpatient hospital spending during that same month.²⁷ By July 2020, the spending from Medicare on SNFs exceeded the levels seen in 2019,²⁷ suggesting that the percentage of acute care admissions discharging to SNFs was no lower for Medicare patients in response to COVD-19. We also considered more stringent SNF admission standards as another potential confounder; however, this was not seen at the SNFs in the NUHS geography, where the referral process became less stringent because of COVID-19 waivers for a qualifying stay or skilled need from the Centers for Medicare and Medicaid Services. We were also not able to account for readmissions outside of NUHS, and therefore there may have been differences in the readmission rate that were unmeasured. To address this limitation, we reviewed a data extract from the Illinois Health and Hospital Association and found that the percentage of patients who returned for readmission to a NUHS facility in the year prior to the intervention and during the intervention period were 92.8% and 95.3%, respectively. From this we concluded the unmeasured readmission rate appears to be low, stable, and unlikely to have altered the results of this study. Additionally, when calculating potential COVID-19 cases avoided, the expected number was, by necessity, derived from the observed outcome, given the absence of COVID-19 in the preintervention population. This may have introduced unmeasured confounders, limiting the ability to precisely measure the effect size or draw conclusions on causation. Finally, there may be limitations to the generalizability of these results based on the payor mix of the population at NUHS, which is predominantly insured through Medicare or commercial payors.

We believe this model is replicable and the results generalizable and could serve as both a template for reducing the risks of COVID-19 in SNFs and as part of a larger infection-control strategy to mitigate disease spread in vulnerable populations. It could also be applied as a component of value-improvement programs to foster appropriate use of postacute services after an acute care hospitalization, ensuring safe transitions of care through promotion of high-value care practices.

The authors thank Wei Ning Chi for editorial assistance.

The COVID-19 outbreak in February 2020 at a congregate living facility near Seattle, Washington, signaled the beginning of the pandemic in the United States. In that facility, infected residents had a 54.5% hospitalization rate and 33.7% case-fatality rate.¹ Similar to the experience in Washington, all congregate living facilities have proved particularly vulnerable to the effects of COVID-19,^2-7 with residents at increased risk for disease severity and mortality.^2-7

Due to the COVID-19 emergency, NorthShore University HealthSystem (NUHS), a multihospital, integrated health system in northern Illinois, established a best practice for appropriate use of congregate living facilities after hospitalization. This focused on the safety of discharged patients and mitigation of COVID-19 by putting in place a referral process to a newly established congregate living review committee (CLRC) for review prior to discharge. Although all discharges to congregate living settings are at high risk,² new placements to skilled nursing facilities (SNFs) were the primary focus of the committee and the sole focus of this study. In this study, we sought to determine whether establishment of the CLRC was associated with a reduction in SNF utilization, whether this was safe and efficient, and whether it was associated with a reduction in COVID-19 incidence in the 30 days following discharge.

Setting and Case Review Intervention

We conducted a retrospective cohort study for patients hospitalized within NUHS from March 19, 2019 to July 16, 2020, designed as an interrupted time series. The study was approved by the NUHS Institutional Review Board (EH21-022).

The study exposure was creation of a referral and review process for all patients with expected discharge to a SNF and was implemented as part of usual discharge planning during the COVID-19 pandemic. The key intervention was to establish a multidisciplinary committee, the CLRC, to review all potential discharges to SNFs. The CLRC had dual goals of preventing COVID-19 spread in facilities by limiting placement of new residents and protecting a vulnerable population from a setting that conferred a higher risk of acquiring COVID-19. The CLRC was organized as a multidisciplinary committee with physicians, case managers, social workers, physical therapists, occupational therapists, and the director of NUHS home health agency. Physician members were evenly split as half hospitalists and half ambulatory physicians. The CLRC review was initiated by a patient’s assigned case manager or social worker by consult through a referral in the electronic medical record (EMR). Each case was summarized and then presented to the full CLRC. The CLRC met for 1 hour per day, 6 days per week, to review all planned discharges that met criteria for review. A committee physician chaired each meeting. Three other members were needed for a quorum, with one other member with a title of director or higher. Time required was the 1-hour daily meeting, as well as one full-time position for case review, preparation, and program administration. The case presentation included a clinical summary of the hospitalization as well as COVID-19 status and testing history, previous living situation, level of home support, functional level, psychosocial needs, barrier(s) to discharging home, and long-term residential plans. A structured assessment was then made by each CLRC member in accordance with their professional expertise. Unanimous consensus would be reached before finalizing any recommended adjustments to the discharge, which would be communicated to the inpatient care team via a structured note within the EMR, along with direct communication to the assigned case manager or social worker. When the CLRC suggested adjustments to the discharge, they would work with the assigned case manager or social worker to communicate an appropriate post–acute care plan with the patient or appropriate representative. If there was disagreement or the recommendations could not be followed, the case manager or social worker would place a new referral with additional information for reconsideration. Following a recommendation for SNF, verification would be completed by the CLRC prior to discharge. This process is detailed in Figure 1.

Patient Population

Inclusion criteria for the study were: (1) inpatient hospitalization and (2) eligibility for risk scoring via the organization’s clinical analytics prediction engine (CAPE).⁸ CAPE is a validated predictive model that includes risk of readmission, in-hospital mortality, and out-of-hospital mortality,⁸ with extensive adoption at NUHS. CAPE score eligibility was used as an inclusion criterion so that CAPE could be applied for derivation of a matched control. CAPE eligibility criteria include admission age of at least 18 years and that hospitalization is not psychiatric, rehabilitative, or obstetric. Patients must not be enrolled in hospice and must be discharged alive.

Exclusions were patients who tested positive for SARS-CoV-2 prior to or during index hospitalization. Excluding COVID-19 patients from the analysis eliminated a confounder not present in the preintervention group.

For patients with multiple inpatient admissions, the first admission was the only admission used for analysis. Additionally, if a patient had an admission that occurred in both the preintervention and postintervention periods, they were included only in the postintervention period. This was done to avoid any within-subject correlation and ensure unique patients in each group. Confounding from this approach was mitigated through the process of deriving a matched control.

Outcomes Measurement

The primary outcome of interest was total discharges to SNF across NUHS facilities after hospital admission. Patients were identified as discharging to a SNF if discharge destination codes 03, 64, or 83 appeared on the hospital bill. Additionally, new discharges to SNFs were assessed and identified if documentation indicated that the patient’s living arrangement prior to admission was not a SNF but discharge billing destination codes 03, 64, or 83 appeared on the hospital bill.

Secondary outcomes were measurement of readmissions, days to readmission, and median length of stay (LOS). Readmissions and LOS were balancing measures for the primary outcome, with readmissions measured to evaluate the safety of the CLRC process and LOS measured to evaluate its efficiency. A readmission was any patient who had an unplanned inpatient admission at an NUHS facility within 30 days after an index admission. LOS was measured in days from arrival on a hospital unit to time of discharge.

Additional analysis was done to estimate the effect of the intervention on the incidence of COVID-19 in the 30 days following discharge by comparing the observed to expected incidence of COVID-19 by discharge destination. The expected values were derived by estimating COVID-19 cases that would have been expected to occur with rates of preintervention SNF utilization. This was accomplished by multiplying the observed incidence of COVID-19 in the 30 days following discharge by the number of patients who were discharged to SNFs or home/other in the preintervention period. This expected value was then compared with the observed values to estimate the effect size of the intervention on COVID-19 incidence following discharge. This method of deriving an expected value from the observed incidence was utilized because the preintervention period was before COVID-19 was widespread in the community. It was therefore not possible to directly measure COVID-19 incidence in the preintervention period.

Data Source

Data were retrieved from the NUHS Enterprise Data Warehouse, NUHS’s central data repository, which contains a nightly upload of clinical and financial data from the EMR. Data were collected between March 19, 2019, and July 16, 2020.

The preintervention period was defined as March 19, 2019, to March 18, 2020. Data from that interval were compared with the postintervention period, which was from March 19, 2020, to July 16, 2020. The preintervention period, 1 year immediately prior to the intervention, was chosen to limit any effect of temporal trends while also providing a large sample size. The postintervention period began on the first day NUHS implemented the revised approach to SNF use and ended on the last day before the review process was modified.

Data Analysis

An interrupted time series was used to measure the impact of adoption of the CLRC protocol. A matched control was derived from the preintervention population. To derive this matched control, there was an assessment of covariates in the preintervention and postintervention groups using a standardized mean difference (SMD)⁹ that indicated an imbalance (SMD ≥ 0.1) in some covariates. A propensity score–matching technique¹⁰ was applied to address this imbalance and lack of randomization.

The candidate variables for propensity matching were chosen if they had an association with 30-day readmission. Readmission was chosen to find candidate variables because, of the possible outcomes, this was the only one that was not directly impacted by any CLRC decision. Each covariate was assessed using a logistic regression model while controlling for the postintervention group. If there was an association between a covariate and the outcome, it was chosen for propensity matching. Propensity scores were calculated using a logistic regression model with the treatment (1/0) variable as the dependent variable and the chosen covariates as predictors.

There were no indications of strong multicollinearity. The propensity scores generated were then used to derive a matched control using paired matching. MatchIt package in R (R Foundation for Statistical Computing) was used to create a matched dataset with a logit distance and standard caliper of 0.2 times the standard deviations of the logit of the propensity score. If a match was not found within the caliper, the nearest available match was used.

Regression adjustment¹¹ was then performed using multivariate linear/logistic regression with LOS, readmission rate, days to readmission, total SNF discharges, and new SNF discharges as the outcomes. Treatment (1/0) variable and propensity score were used as the predictors. The adjusted coefficients or odds ratios (ORs) of the intervention variable were thus derived, and their associated P values were used to assess the impact of the intervention on the respective outcomes.

The unmatched preintervention population included 14,468 patients, with 4424 patients in the postintervention population. A matched population was derived and, after matching, the population sizes for pre and post intervention were 4424 each. In the matched population, all measured preintervention characteristics had SMDs and P values that were statistically equivalent. Patient characteristics for the unmatched and matched populations are detailed in Table 1.

During the preintervention period, 1130 (25.5%) patients were discharged to a SNF, with 776 (17.5%) patients being new SNF discharges. In the postintervention period, 568 (12.8%) patients were discharged to a SNF, with 257 (5.8%) patients being new SNF discharges. Total SNF discharges postintervention saw a 49.7% relative reduction (OR, 0.42; 95% CI, 0.38-0.47), while new SNF discharges saw a 66.9% relative reduction (OR, 0.29; 95% CI, 0.25-0.34). These results for both total and new SNF discharges were statistically significant, with P values of <.001, respectively.

Readmissions in the preintervention period were 529 (12.0%) patients, compared with 559 (12.6%) patients in the postintervention period (OR, 1.06; 95% CI, 0.93-1.20; P =.406). An OR was also calculated for readmissions, adjusting for discharge disposition, to account for changes observed in SNF use in the postintervention period. This OR was 1.11 (95% CI, 0.97-1.26; P = .131). Days to readmission in the preintervention and postintervention groups were 11.0 days and 12.0 days, respectively (OR, 0.41; 95% CI, –0.61 to 1.43; P = .429).

LOS was 3.61 days in the preintervention group and 3.64 days in the postintervention group, with an interquartile range (IQR) of 2.14 to 5.69 days in the preintervention group and 2.08 to 5.95 in the postintervention group (OR, 0.09; 95% CI, –0.09 to 0.27; P =.316). These results are summarized in Table 2.

A COVID-19 outbreak in a SNF presents a grave risk to residents and patients discharged to these facilities. It is critical for healthcare systems to do the utmost to protect the health of this vulnerable population and the public in efforts to limit COVID-19 within SNFs.^12-14

In this study, we observed that at NUHS, establishing a multidisciplinary review committee, the CLRC, to assess the appropriateness of discharge to a SNF after hospitalization resulted in a nearly 50% reduction in total SNF discharges and a greater than two-thirds reduction in new SNF discharges, without any increase in LOS or readmissions. Additionally, it was observed that discharging to settings other than a SNF greatly reduced a patient’s risk of being diagnosed with COVID-19 within 30 days, a result that reached statistical significance. Based on the observed 37.2% relative reduction in COVID-19 cases, we estimate that there may have been one COVID-19 infection prevented every 5.6 days from this intervention. Based on published COVID-19 mortality rates for SNF residents,¹ the intervention may have prevented one death every 2.6 weeks. Beyond the risk of COVID-19, other benefits of reducing SNF use are patient and family well-being. Although not measured in this study, others have published about the significant psychological burdens placed on SNF residents, who were at high risk for social isolation, anxiety, and depression during the COVID-19 pandemic^2,15-19 Family members also may have had increased stress, as they were deprived of the opportunity to visit loved ones, advocate for them, and help maintain their identity, humanity, and quality of life.²⁰

Although other hospitals have established a structured approach to reduce COVID-19 in SNFs,²¹ to the best of the authors’ knowledge, the approach described in this article is a unique response to the COVID-19 pandemic. As we have demonstrated, it is highly effective and safe and likely prevented many COVID-19 cases and deaths.

Furthermore, a review committee, such as the one we have described, has value well beyond the COVID-19 pandemic. The health and affordability of care for patients, provider success in value-based care models, and the long-term sustainability of the US healthcare system require close attention to appropriate use of expensive services and to ensuring that their use creates high value. SNF use after a hospitalization is one such service that is frequently targeted and thought to contribute to a substantial portion of wasteful medical spending.^22,23 Additionally, SNFs are known to be high risk for communicable disease outbreaks other than COVID-19,^24,25 as well as a high-risk environment for many other preventable adverse events.^25,26 This review committee ultimately serves to help determine the most appropriate postacute setting for patients being discharged with a determination made through considerations for patient safety, rehabilitation potential, and mental and physical well-being. From a population health perspective, this can lead to better outcomes and lower costs.^22,23 Therefore, although the risks of COVID-19 infection in SNFs are expected to subside, the work of evaluating appropriate use of SNFs after hospitalization at our institution continues. The broader focus now extends beyond postacute level of service toward ensuring a high-value discharge that results in both appropriate resource use and safe patient care transitions.

Limitations of this study include its retrospective nature, results from a single center, and a number of potentially unmeasured confounders that the COVID-19 pandemic created. One possible confounder is that the reduction in SNF use we observed was a temporal trend related to changing preferences. In addressing this, we reviewed Medicare claims data from the US Department of Health and Human Services in April 2020 and July 2020 compared with the same period in 2019. These data demonstrated only a modest reduction in spending on SNFs in April 2020 that was smaller than the reduction seen in Part A inpatient hospital spending during that same month.²⁷ By July 2020, the spending from Medicare on SNFs exceeded the levels seen in 2019,²⁷ suggesting that the percentage of acute care admissions discharging to SNFs was no lower for Medicare patients in response to COVD-19. We also considered more stringent SNF admission standards as another potential confounder; however, this was not seen at the SNFs in the NUHS geography, where the referral process became less stringent because of COVID-19 waivers for a qualifying stay or skilled need from the Centers for Medicare and Medicaid Services. We were also not able to account for readmissions outside of NUHS, and therefore there may have been differences in the readmission rate that were unmeasured. To address this limitation, we reviewed a data extract from the Illinois Health and Hospital Association and found that the percentage of patients who returned for readmission to a NUHS facility in the year prior to the intervention and during the intervention period were 92.8% and 95.3%, respectively. From this we concluded the unmeasured readmission rate appears to be low, stable, and unlikely to have altered the results of this study. Additionally, when calculating potential COVID-19 cases avoided, the expected number was, by necessity, derived from the observed outcome, given the absence of COVID-19 in the preintervention population. This may have introduced unmeasured confounders, limiting the ability to precisely measure the effect size or draw conclusions on causation. Finally, there may be limitations to the generalizability of these results based on the payor mix of the population at NUHS, which is predominantly insured through Medicare or commercial payors.

We believe this model is replicable and the results generalizable and could serve as both a template for reducing the risks of COVID-19 in SNFs and as part of a larger infection-control strategy to mitigate disease spread in vulnerable populations. It could also be applied as a component of value-improvement programs to foster appropriate use of postacute services after an acute care hospitalization, ensuring safe transitions of care through promotion of high-value care practices.

The authors thank Wei Ning Chi for editorial assistance.

The COVID-19 outbreak in February 2020 at a congregate living facility near Seattle, Washington, signaled the beginning of the pandemic in the United States. In that facility, infected residents had a 54.5% hospitalization rate and 33.7% case-fatality rate.¹ Similar to the experience in Washington, all congregate living facilities have proved particularly vulnerable to the effects of COVID-19,^2-7 with residents at increased risk for disease severity and mortality.^2-7

Due to the COVID-19 emergency, NorthShore University HealthSystem (NUHS), a multihospital, integrated health system in northern Illinois, established a best practice for appropriate use of congregate living facilities after hospitalization. This focused on the safety of discharged patients and mitigation of COVID-19 by putting in place a referral process to a newly established congregate living review committee (CLRC) for review prior to discharge. Although all discharges to congregate living settings are at high risk,² new placements to skilled nursing facilities (SNFs) were the primary focus of the committee and the sole focus of this study. In this study, we sought to determine whether establishment of the CLRC was associated with a reduction in SNF utilization, whether this was safe and efficient, and whether it was associated with a reduction in COVID-19 incidence in the 30 days following discharge.

Setting and Case Review Intervention

We conducted a retrospective cohort study for patients hospitalized within NUHS from March 19, 2019 to July 16, 2020, designed as an interrupted time series. The study was approved by the NUHS Institutional Review Board (EH21-022).

The study exposure was creation of a referral and review process for all patients with expected discharge to a SNF and was implemented as part of usual discharge planning during the COVID-19 pandemic. The key intervention was to establish a multidisciplinary committee, the CLRC, to review all potential discharges to SNFs. The CLRC had dual goals of preventing COVID-19 spread in facilities by limiting placement of new residents and protecting a vulnerable population from a setting that conferred a higher risk of acquiring COVID-19. The CLRC was organized as a multidisciplinary committee with physicians, case managers, social workers, physical therapists, occupational therapists, and the director of NUHS home health agency. Physician members were evenly split as half hospitalists and half ambulatory physicians. The CLRC review was initiated by a patient’s assigned case manager or social worker by consult through a referral in the electronic medical record (EMR). Each case was summarized and then presented to the full CLRC. The CLRC met for 1 hour per day, 6 days per week, to review all planned discharges that met criteria for review. A committee physician chaired each meeting. Three other members were needed for a quorum, with one other member with a title of director or higher. Time required was the 1-hour daily meeting, as well as one full-time position for case review, preparation, and program administration. The case presentation included a clinical summary of the hospitalization as well as COVID-19 status and testing history, previous living situation, level of home support, functional level, psychosocial needs, barrier(s) to discharging home, and long-term residential plans. A structured assessment was then made by each CLRC member in accordance with their professional expertise. Unanimous consensus would be reached before finalizing any recommended adjustments to the discharge, which would be communicated to the inpatient care team via a structured note within the EMR, along with direct communication to the assigned case manager or social worker. When the CLRC suggested adjustments to the discharge, they would work with the assigned case manager or social worker to communicate an appropriate post–acute care plan with the patient or appropriate representative. If there was disagreement or the recommendations could not be followed, the case manager or social worker would place a new referral with additional information for reconsideration. Following a recommendation for SNF, verification would be completed by the CLRC prior to discharge. This process is detailed in Figure 1.

Patient Population

Inclusion criteria for the study were: (1) inpatient hospitalization and (2) eligibility for risk scoring via the organization’s clinical analytics prediction engine (CAPE).⁸ CAPE is a validated predictive model that includes risk of readmission, in-hospital mortality, and out-of-hospital mortality,⁸ with extensive adoption at NUHS. CAPE score eligibility was used as an inclusion criterion so that CAPE could be applied for derivation of a matched control. CAPE eligibility criteria include admission age of at least 18 years and that hospitalization is not psychiatric, rehabilitative, or obstetric. Patients must not be enrolled in hospice and must be discharged alive.

Exclusions were patients who tested positive for SARS-CoV-2 prior to or during index hospitalization. Excluding COVID-19 patients from the analysis eliminated a confounder not present in the preintervention group.

For patients with multiple inpatient admissions, the first admission was the only admission used for analysis. Additionally, if a patient had an admission that occurred in both the preintervention and postintervention periods, they were included only in the postintervention period. This was done to avoid any within-subject correlation and ensure unique patients in each group. Confounding from this approach was mitigated through the process of deriving a matched control.

Outcomes Measurement

The primary outcome of interest was total discharges to SNF across NUHS facilities after hospital admission. Patients were identified as discharging to a SNF if discharge destination codes 03, 64, or 83 appeared on the hospital bill. Additionally, new discharges to SNFs were assessed and identified if documentation indicated that the patient’s living arrangement prior to admission was not a SNF but discharge billing destination codes 03, 64, or 83 appeared on the hospital bill.

Secondary outcomes were measurement of readmissions, days to readmission, and median length of stay (LOS). Readmissions and LOS were balancing measures for the primary outcome, with readmissions measured to evaluate the safety of the CLRC process and LOS measured to evaluate its efficiency. A readmission was any patient who had an unplanned inpatient admission at an NUHS facility within 30 days after an index admission. LOS was measured in days from arrival on a hospital unit to time of discharge.

Additional analysis was done to estimate the effect of the intervention on the incidence of COVID-19 in the 30 days following discharge by comparing the observed to expected incidence of COVID-19 by discharge destination. The expected values were derived by estimating COVID-19 cases that would have been expected to occur with rates of preintervention SNF utilization. This was accomplished by multiplying the observed incidence of COVID-19 in the 30 days following discharge by the number of patients who were discharged to SNFs or home/other in the preintervention period. This expected value was then compared with the observed values to estimate the effect size of the intervention on COVID-19 incidence following discharge. This method of deriving an expected value from the observed incidence was utilized because the preintervention period was before COVID-19 was widespread in the community. It was therefore not possible to directly measure COVID-19 incidence in the preintervention period.

Data Source

Data were retrieved from the NUHS Enterprise Data Warehouse, NUHS’s central data repository, which contains a nightly upload of clinical and financial data from the EMR. Data were collected between March 19, 2019, and July 16, 2020.

The preintervention period was defined as March 19, 2019, to March 18, 2020. Data from that interval were compared with the postintervention period, which was from March 19, 2020, to July 16, 2020. The preintervention period, 1 year immediately prior to the intervention, was chosen to limit any effect of temporal trends while also providing a large sample size. The postintervention period began on the first day NUHS implemented the revised approach to SNF use and ended on the last day before the review process was modified.

Data Analysis

An interrupted time series was used to measure the impact of adoption of the CLRC protocol. A matched control was derived from the preintervention population. To derive this matched control, there was an assessment of covariates in the preintervention and postintervention groups using a standardized mean difference (SMD)⁹ that indicated an imbalance (SMD ≥ 0.1) in some covariates. A propensity score–matching technique¹⁰ was applied to address this imbalance and lack of randomization.

The candidate variables for propensity matching were chosen if they had an association with 30-day readmission. Readmission was chosen to find candidate variables because, of the possible outcomes, this was the only one that was not directly impacted by any CLRC decision. Each covariate was assessed using a logistic regression model while controlling for the postintervention group. If there was an association between a covariate and the outcome, it was chosen for propensity matching. Propensity scores were calculated using a logistic regression model with the treatment (1/0) variable as the dependent variable and the chosen covariates as predictors.

There were no indications of strong multicollinearity. The propensity scores generated were then used to derive a matched control using paired matching. MatchIt package in R (R Foundation for Statistical Computing) was used to create a matched dataset with a logit distance and standard caliper of 0.2 times the standard deviations of the logit of the propensity score. If a match was not found within the caliper, the nearest available match was used.

Regression adjustment¹¹ was then performed using multivariate linear/logistic regression with LOS, readmission rate, days to readmission, total SNF discharges, and new SNF discharges as the outcomes. Treatment (1/0) variable and propensity score were used as the predictors. The adjusted coefficients or odds ratios (ORs) of the intervention variable were thus derived, and their associated P values were used to assess the impact of the intervention on the respective outcomes.

The unmatched preintervention population included 14,468 patients, with 4424 patients in the postintervention population. A matched population was derived and, after matching, the population sizes for pre and post intervention were 4424 each. In the matched population, all measured preintervention characteristics had SMDs and P values that were statistically equivalent. Patient characteristics for the unmatched and matched populations are detailed in Table 1.

During the preintervention period, 1130 (25.5%) patients were discharged to a SNF, with 776 (17.5%) patients being new SNF discharges. In the postintervention period, 568 (12.8%) patients were discharged to a SNF, with 257 (5.8%) patients being new SNF discharges. Total SNF discharges postintervention saw a 49.7% relative reduction (OR, 0.42; 95% CI, 0.38-0.47), while new SNF discharges saw a 66.9% relative reduction (OR, 0.29; 95% CI, 0.25-0.34). These results for both total and new SNF discharges were statistically significant, with P values of <.001, respectively.

Readmissions in the preintervention period were 529 (12.0%) patients, compared with 559 (12.6%) patients in the postintervention period (OR, 1.06; 95% CI, 0.93-1.20; P =.406). An OR was also calculated for readmissions, adjusting for discharge disposition, to account for changes observed in SNF use in the postintervention period. This OR was 1.11 (95% CI, 0.97-1.26; P = .131). Days to readmission in the preintervention and postintervention groups were 11.0 days and 12.0 days, respectively (OR, 0.41; 95% CI, –0.61 to 1.43; P = .429).

LOS was 3.61 days in the preintervention group and 3.64 days in the postintervention group, with an interquartile range (IQR) of 2.14 to 5.69 days in the preintervention group and 2.08 to 5.95 in the postintervention group (OR, 0.09; 95% CI, –0.09 to 0.27; P =.316). These results are summarized in Table 2.

A COVID-19 outbreak in a SNF presents a grave risk to residents and patients discharged to these facilities. It is critical for healthcare systems to do the utmost to protect the health of this vulnerable population and the public in efforts to limit COVID-19 within SNFs.^12-14

In this study, we observed that at NUHS, establishing a multidisciplinary review committee, the CLRC, to assess the appropriateness of discharge to a SNF after hospitalization resulted in a nearly 50% reduction in total SNF discharges and a greater than two-thirds reduction in new SNF discharges, without any increase in LOS or readmissions. Additionally, it was observed that discharging to settings other than a SNF greatly reduced a patient’s risk of being diagnosed with COVID-19 within 30 days, a result that reached statistical significance. Based on the observed 37.2% relative reduction in COVID-19 cases, we estimate that there may have been one COVID-19 infection prevented every 5.6 days from this intervention. Based on published COVID-19 mortality rates for SNF residents,¹ the intervention may have prevented one death every 2.6 weeks. Beyond the risk of COVID-19, other benefits of reducing SNF use are patient and family well-being. Although not measured in this study, others have published about the significant psychological burdens placed on SNF residents, who were at high risk for social isolation, anxiety, and depression during the COVID-19 pandemic^2,15-19 Family members also may have had increased stress, as they were deprived of the opportunity to visit loved ones, advocate for them, and help maintain their identity, humanity, and quality of life.²⁰

Although other hospitals have established a structured approach to reduce COVID-19 in SNFs,²¹ to the best of the authors’ knowledge, the approach described in this article is a unique response to the COVID-19 pandemic. As we have demonstrated, it is highly effective and safe and likely prevented many COVID-19 cases and deaths.

Furthermore, a review committee, such as the one we have described, has value well beyond the COVID-19 pandemic. The health and affordability of care for patients, provider success in value-based care models, and the long-term sustainability of the US healthcare system require close attention to appropriate use of expensive services and to ensuring that their use creates high value. SNF use after a hospitalization is one such service that is frequently targeted and thought to contribute to a substantial portion of wasteful medical spending.^22,23 Additionally, SNFs are known to be high risk for communicable disease outbreaks other than COVID-19,^24,25 as well as a high-risk environment for many other preventable adverse events.^25,26 This review committee ultimately serves to help determine the most appropriate postacute setting for patients being discharged with a determination made through considerations for patient safety, rehabilitation potential, and mental and physical well-being. From a population health perspective, this can lead to better outcomes and lower costs.^22,23 Therefore, although the risks of COVID-19 infection in SNFs are expected to subside, the work of evaluating appropriate use of SNFs after hospitalization at our institution continues. The broader focus now extends beyond postacute level of service toward ensuring a high-value discharge that results in both appropriate resource use and safe patient care transitions.

Limitations of this study include its retrospective nature, results from a single center, and a number of potentially unmeasured confounders that the COVID-19 pandemic created. One possible confounder is that the reduction in SNF use we observed was a temporal trend related to changing preferences. In addressing this, we reviewed Medicare claims data from the US Department of Health and Human Services in April 2020 and July 2020 compared with the same period in 2019. These data demonstrated only a modest reduction in spending on SNFs in April 2020 that was smaller than the reduction seen in Part A inpatient hospital spending during that same month.²⁷ By July 2020, the spending from Medicare on SNFs exceeded the levels seen in 2019,²⁷ suggesting that the percentage of acute care admissions discharging to SNFs was no lower for Medicare patients in response to COVD-19. We also considered more stringent SNF admission standards as another potential confounder; however, this was not seen at the SNFs in the NUHS geography, where the referral process became less stringent because of COVID-19 waivers for a qualifying stay or skilled need from the Centers for Medicare and Medicaid Services. We were also not able to account for readmissions outside of NUHS, and therefore there may have been differences in the readmission rate that were unmeasured. To address this limitation, we reviewed a data extract from the Illinois Health and Hospital Association and found that the percentage of patients who returned for readmission to a NUHS facility in the year prior to the intervention and during the intervention period were 92.8% and 95.3%, respectively. From this we concluded the unmeasured readmission rate appears to be low, stable, and unlikely to have altered the results of this study. Additionally, when calculating potential COVID-19 cases avoided, the expected number was, by necessity, derived from the observed outcome, given the absence of COVID-19 in the preintervention population. This may have introduced unmeasured confounders, limiting the ability to precisely measure the effect size or draw conclusions on causation. Finally, there may be limitations to the generalizability of these results based on the payor mix of the population at NUHS, which is predominantly insured through Medicare or commercial payors.

We believe this model is replicable and the results generalizable and could serve as both a template for reducing the risks of COVID-19 in SNFs and as part of a larger infection-control strategy to mitigate disease spread in vulnerable populations. It could also be applied as a component of value-improvement programs to foster appropriate use of postacute services after an acute care hospitalization, ensuring safe transitions of care through promotion of high-value care practices.

The authors thank Wei Ning Chi for editorial assistance.

Hospital readmissions are a significant problem in the United States, affecting 15% to 30% of discharges and incurring costs of more than $17 billion annually.¹ Timely posthospitalization follow-up visits are critical to ensure the effective transfer of patients to the outpatient setting; such visits reduce readmission rates as well as hospital length of stay and overall health care resource utilization.^2-4 Patients who receive inadequate follow-up care (ie, within 4 weeks of discharge) are significantly more likely to be readmitted than those who receive close follow-up care^.5

Due to the large clinical and financial consequences associated with hospital readmission, a variety of interventions have been studied, including home visits, telemonitoring, medication management, telephone calls, and postdischarge clinics.^6,7 While studies have not shown postdischarge clinics to be universally efficacious in reducing readmission rates, there is increasing evidence of reduced readmission rates in clinics that target high-risk patients (eg, patients with congestive heart failure [CHF]) rather than the total population.² A study by Hernandez et al that evaluated the relationship between early physician follow-up and 30-day readmissions showed a significantly lower readmission rate among hospitals with higher follow-up rates.⁸ Similarly, patients with CHF in a large, integrated health system who were seen within 7 days of discharge had an odds ratio (OR) of 0.81 (95% CI, 0.70-0.94) for 30-day readmissions.⁹

Transitions-of-care clinics (TOCC), designed to provide early postdischarge follow-up to high-risk patients, have been shown to reduce 30-day readmission rates,^3,4,10,11 especially in clinics that have same-physician follow-up visits rather than follow-up visits with a community primary care physician (PCP).¹² The most pronounced impact of postdischarge follow-up is seen in high-risk patients with high complexity or high severity of disease; however, complex rural patients are less likely to have access to specialty care.¹³ As a result, since rural residents must travel farther for specialty care, they are seen less frequently than their urban counterparts.^14,15

Prior to our TOCC initiative, the Iowa City VA (ICVA) ranked in the fifth quintile of the Veterans Health Administration (VHA) Strategic Analytics for Improvement and Learning model for hospital-wide readmissions (HWR), meaning that HWR at ICVA were higher than 80% of the other VHA healthcare centers. The low score in this metric was in part due to readmission rates in high-risk populations, including patients with CHF and those with high Care Assessment Need (CAN) scores. One concern was that the ICVA system serves many veterans from rural areas, some of whom must travel up to 200 miles to access inpatient and subspecialty care.

To meet these challenges, we implemented a TOCC to deliver timely postdischarge care focusing on high-risk and high-complexity patients. To address access-to-care issues of patients living in rural areas within the ICVA, we included virtual follow-up visits as a key component of our intervention.^16,17 The aim of this project was to decrease 30-day readmission rates of ICVA patients by 20% within 12 months of implementation.

Setting/Study Population

The ICVA serves 184,000 veterans stretched over 50 counties in eastern Iowa, western Illinois, and northern Missouri, with more than 60% of these patients residing in rural areas. Patients were initially eligible for the TOCC if they had an admission diagnosis of CHF and a CAN score > 85 at the time of discharge. The CAN score, developed by the VA to assess the risk of hospital readmission in individual patients, factors in several variables, including demographics, coexisting conditions, vital signs, utilization of services, pharmacy visits, and laboratory results. Patients in the top 5% (95-99) have a readmission rate of 20% at 90 days. Since the CAN is a proprietary tool, it may not be published in full; however, this assessment tool is commonly used and frequently cited in VA research.^18-22 The CAN score is expressed as a percentile ranging from 0 (lowest risk) to 99 (highest risk). Patient eligibility was expanded during subsequent Plan-Do-Study-Act (PDSA) cycles, as outlined below. Patient eligibility was expanded during subsequent PDSA cycles (also outlined below). A review by a local institutional review board was obtained, and the study was classified as exempt due to the use of deidentified data. Standards for Quality Improvement Reporting Excellence 2.0 guidelines were used to construct the manuscript.

Magnitude Assessment

The numbers of discharges, readmissions within 30 days, emergency department (ED) visits by all discharged veterans, and veterans discharged with a CHF hospital diagnosis were recorded from February 2017 to February 2018, which were the 12 months immediately preceding the pilot implementation.

Intervention

The primary intervention was referral to the newly formed ICVA TOCC. The multidisciplinary TOCC team consisted of hospitalists, pharmacists, schedulers, and discharge planners/care managers. Patients were identified by the hospitalist team during admission; prior to hospital discharge, these patients were referred to TOCC discharge planners to schedule appropriate follow-up appointments. Virtual follow-up visits were conducted using a patient’s home technology; in cases where a patient lacked adequate technology capabilities (eg, no computer or internet access), the ICVA provided a tablet device with cellular internet capability for temporary use. Specific clinical activities included medication reconciliation by a pharmacist, follow-up of pending laboratory studies, imaging studies, pathology results, medical diagnosis education, counseling regarding dietary restrictions, and contingency planning outside of an ED visit in the event of a change in clinical status. In addition, the TOCC aimed to facilitate a smooth transition of care back to the PCP by arranging follow-up appointments, providing visit summaries, and scheduling consults with specialty care, as appropriate.

Measures

The primary objective measure was the 30-day readmission rate in the ICVA hospital. Secondary measures included the number of VHA ED visits within 30 days of discharge. The main process measures were the number of hospital discharges per month, the number of TOCC referrals, the number of TOCC appointments made, the number of virtual and in-person visits, and the percentage of appointment “no-shows.”

Implementation

The TOCC was piloted from April 2018 to October 2018. During the pilot phase, TOCC enrollment was limited to virtual appointments and to patients with an admission diagnosis of CHF and a CAN score of > 85. The TOCC had staff on-site 2 days a week; this included pharmacists to reconcile medications and hospitalists to address follow-up care needs.

The TOCC clinic was temporarily closed at the end of October 2018 to analyze pilot results. Based on stakeholder feedback, changes made as part of the second PDSA cycle included expanding eligibility criteria to any hospital admission diagnosis and to patients with a CAN score < 85 if the hospitalist team felt the patient was likely to benefit from TOCC follow-up. In addition, on-site clinic staffing was expanded from 2 to 5 days per week to improve access, and the option for an in-person visit was added based on concerns some veterans expressed regarding the use of the technology at home. Finally, a formal resident program was added, and the order set for referrals was simplified. The TOCC was restarted in February 2019, and TOCC metrics were reviewed monthly. By July 2019, we identified issues with TOCC referrals and appointment creation that required additional modifications to the intervention.

A third PDSA cycle was initiated in July 2019 and included major changes, notably the formation of a designated TOCC committee. The committee appointed a dedicated TOCC scheduler whose role was to reduce confusion regarding scheduling, to update the discharge instructions/orders template to lower incidences of “double-booking” that occurred with PCP and TOCC appointments, to modify discharge educational instruction regarding virtual visits and tablet use, to adjust the TOCC-PCP handoff, and to formalize interactions between discharge coordinators and residents to review possible referrals every morning (Appendix Figure 1).

Statistical Analysis

Run charts were constructed by plotting monthly primary outcome values and monthly process metrics (Figure, Appendix Figure 2, Appendix Figure 3). Chi-square tests were used to compare 30-day readmission rates before and after the intervention. Statistical modeling was used to determine differences in outcomes between referred patients seen and referred patients not seen by the TOCC. In these statistical models, the outcome measures were 30-day readmissions, 30-day ED visits, and 6-month mortality. Covariates included in the final analysis were age, gender, race, CAN score, rural-urban commuting area code, referral service (resident vs nonresident), and admission diagnosis. Admission diagnoses were sorted by the investigators into one of the following seven categories: cardiac, infectious, pulmonary, gastrointestinal, neurologic, renal, and other.

Mean (SD) or counts and percentages were used to describe the distribution of continuous and categorical variables, respectively. Kruskal-Wallis test, t test, or chi-square tests were used, as appropriate, across categories. Generalized linear models with a logistic link function were used to test for differences between patients who kept their appointment at the TOCC and those who did not keep their TOCC appointment (both unadjusted and adjusted for all of the covariates previously mentioned). In addition, generalized linear models were also used to compare outcomes between TOCC patients seen virtually vs those seen in-person (both unadjusted and adjusted for all the covariates previously mentioned). All statistical tests were considered significant at a two-sided P < .05. All analyses were performed using SAS software version 9.4 (SAS Institute Inc).

Magnitude Assessment

During the preimplementation period (February 2017-February 2018), there were 3014 patient discharges from ICVA and 343 readmissions, resulting in a readmission rate of 11.4%. Among patients with a hospital-admission diagnosis of cardiorespiratory disease, which included patients with CHF, there were 381 discharges and 46 readmissions, resulting in a readmission rate of 12.1%.

Primary Outcome

During the pilot phase, which was conducted from April 2018 to October 2018, 142 patients who met inclusion criteria (CHF diagnosis and a CAN score > 85) were discharged from ICVA, and 56 referrals to the TOCC were placed. The readmission rate among the cardiorespiratory cohort of veterans was 9.5%.

During the expansion of the intervention from February 2019 to February 2020, there were 2844 discharges from the ICVA and 291 readmissions, resulting in a readmission rate of 10.2%. However, there was a further decrease in the readmission rate after the third PDSA cycle was initiated in July 2019 (Appendix Figure 1). The readmission rate was 9.2% in the final 6 months of the intervention period, and 7.9% in the final 3 months. Of note, in the group of 1948 patients who did not meet the eligibility criteria to participate in our study, the readmission rate during the same time period was 8.6% (161 readmissions).

When comparing the 6 months following the third PDSA cycle to the magnitude assessment period, there was a relative readmission reduction of 19.3% (P = .04), and an absolute reduction of 2.2%. If the final 3 months of the intervention period are included, there was an absolute reduction of 3.5% and a relative reduction of 30.7% (P = .01). Notably, before the pilot phase, ICVA was in the fifth quintile for HWR among VA hospitals but improved to the second quintile by the end of the expansion phase.

Process Outcomes

Process metrics for TOCC referrals, the number of patients seen, and the number of virtual and in-person visits over time are shown in Appendix Figure 3. Rates of TOCC referrals and the number of TOCC visits were lower than anticipated during the first 5 months of the intervention. However, TOCC referrals increased significantly after we implemented the previously described changes as part of the third PDSA cycle. As a result, total, virtual, and in-person visits also significantly increased from July 2019 to February 2020. The proportion of patients choosing virtual vs in-person visits fluctuated over time, but virtual visits were generally chosen more often than in-person visits.

Statistical Modeling

Baseline Data

Cohort characteristics are shown in Table 1. The cohort, which reflected the ICVA population, was predominantly male (96%) and White (93%), with a mean age of 67 years. The population was approximately half urban and half rural in composition, and the most common reason for hospital admission was cardiac. Other than a small but statistically significant difference in CAN scores, there were no significant differences between patients who kept their TOCC appointment and those who did not. There were also no differences in baseline characteristics between patients who chose virtual follow-up and patients who chose in-person follow-up, including the proportion of urban and rural patients.

Outcomes

Patients who kept their TOCC appointments had a 30-day readmission rate of 9.6%, which was significantly lower than the 30-day readmission rate of 27% in the group that did not keep their TOCC appointment (P < .001). Similarly, the percentage of patients treated in the ED was 15% in the TOCC group compared to 31.2% in the group that canceled their appointment (P < .001) (Table 1). In the multivariable analysis, patients who were seen in the TOCC group had an OR for 30-day readmission of 0.35 (95% CI, 0.19-0.62, P < .001), and an OR for ED visits of 0.39 (95% CI, 0.23-0.65; P < .001) (Table 2). There was no statistically significant difference in 6-month mortality between the two groups. In the virtual group compared to the in-person group, there were no statistically significant differences in outcomes between the two groups in the unadjusted or adjusted analysis (Table 2).

In this quality improvement initiative, we found that a TOCC targeting high-risk patients and offering virtual follow-up visits significantly decreased the 30-day readmission rates among veterans at ICVA. Statistical comparisons of patients seen at the TOCC vs those not seen at the TOCC showed a dramatic reduction in 30-day readmissions and ED visits. Finally, virtual follow-ups were more popular than in-person visits, and patients who followed up virtually had equivalent outcomes to those with the more traditional follow-up.

In the expansion phase, eligibility was expanded to include any hospital indication but continued to focus on high-risk patients. Existing literature suggests that providing postdischarge care to all patients, including low- or medium-risk patients, may not be as impactful as enrolling high-risk patients only. For instance, a postdischarge clinic offered to all patients at a VA system in Colorado did not reduce readmission rates compared to PCP follow-up.²³ In contrast, a study of more than 10,000 high-risk urban patients demonstrated that postdischarge care resulted in a 9.3% reduction in readmission risk.²⁴ Our data are consistent with the previously published studies, as the average CAN score of patients seen in TOCC was 90, suggesting a high risk of readmission. In the final 12 months of the intervention, 15% of discharged patients were seen at the TOCC clinic, suggesting that targeted intervention within the small subset of high-risk patients was sufficient to achieve our primary aim. Of note, among patients who did not meet the inclusion criteria for TOCC referral (ie, patients not considered high risk [CAN score ≤ 85]), the rate of readmissions was 8.6%.

Most of the available research on the efficacy of postdischarge clinics was conducted in urban environments. Our ICVA population sees a large proportion of rural veterans, who account for just over 50% of the discharge population. In a study of more than 2 million Medicare patients discharged from US hospitals, the 30-day readmission rates and adjusted mortality rates were higher among patients in rural counties, and post–acute care seemed to have a greater impact in rural rather than urban settings.²⁵ Previous studies have demonstrated that virtual visits have the potential to improve readmission rates, especially in patients with CHF²⁶ and in patients at the highest risk for readmission.²⁷ In our study, the pilot phase offered only virtual visits, but we subsequently added an in-person option based on veteran feedback. Interestingly, over the next 12 months, virtual visits were more popular with both urban and rural veterans, and there were no differences in the number of rural patients in the in-person vs the virtual group. These findings suggest factors other than rurality influenced the decision to choose virtual follow-up visits over in-person visits. Future studies should seek to determine the extent to which factors such as age, race, educational level, and socioeconomic circumstances impact veterans’ follow-up decisions. Not only were outcomes among patients who chose virtual visits the same as those of patients who chose in-person visits, but both of these groups had better outcomes compared to the non-TOCC group (Table 2). This finding demonstrating the efficacy of virtual visits among rural and urban patients has taken on increased significance due to the COVID-19 pandemic, as virtual visits offer a safer option, one that minimizes physical contact.

Our quality improvement analysis included a statistical comparison of patients seen vs those not seen at the TOCC. Patients who were referred to the TOCC but chose not to keep their appointment were similar to those seen in TOCC in terms of age, CAN score, rurality, and hospital diagnosis, but readmission rates were substantially higher in this group even after adjustments for covariates (Table 2). Evaluating causality in interventions aimed to reduce hospital readmission rates is complicated.²⁸ Our findings add greater plausibility to the utility of TOCC in accounting for at least a portion of the reported decrease in ICVA 30-day readmissions.

Our study has several strengths, including an observation period longer than 2 years, a large population of discharged veterans within an integrated healthcare system, and a large proportion of patients living in rural areas. Another strength of our study is the innovative nature of the intervention, which features a multidisciplinary team and the option of virtual or in-person visits. Nevertheless, this study also has several important limitations. As a single-center study, our findings may not be generalizable to other institutions, especially those outside the VHA system. Similarly, our study population reflected that of the ICVA, which may limit generalizability to a more diverse population. While we attempted to account in our statistical modeling for baseline differences between referred patients seen by the TOCC and those referred but not seen, we cannot exclude residual confounding between the groups. Specifically, the comparison of patients who did and did not choose TOCC follow-up introduces the possibility of selection bias. Future randomized/controlled studies will need to evaluate whether TOCC is more effective than the standard of care to reduce readmissions. Finally, since the analysis period following the final PDSA cycle was compressed due to the onset of the COVID-19 pandemic in the United States, no data are available regarding the sustained impacts of changes made during this cycle.

A multidisciplinary TOCC within the ICVA, featuring both virtual and in-person visits, reduced 30-day readmission rates by 19.3%; this approach to care was especially effective in patients with CHF. Virtual visits were the follow-up mode of choice for both urban and rural veterans, and there was no difference in outcomes between these two follow-up options. Future studies will focus on additional quality metrics, including cost-effectiveness and patient satisfaction.

Hospital readmissions are a significant problem in the United States, affecting 15% to 30% of discharges and incurring costs of more than $17 billion annually.¹ Timely posthospitalization follow-up visits are critical to ensure the effective transfer of patients to the outpatient setting; such visits reduce readmission rates as well as hospital length of stay and overall health care resource utilization.^2-4 Patients who receive inadequate follow-up care (ie, within 4 weeks of discharge) are significantly more likely to be readmitted than those who receive close follow-up care^.5

Due to the large clinical and financial consequences associated with hospital readmission, a variety of interventions have been studied, including home visits, telemonitoring, medication management, telephone calls, and postdischarge clinics.^6,7 While studies have not shown postdischarge clinics to be universally efficacious in reducing readmission rates, there is increasing evidence of reduced readmission rates in clinics that target high-risk patients (eg, patients with congestive heart failure [CHF]) rather than the total population.² A study by Hernandez et al that evaluated the relationship between early physician follow-up and 30-day readmissions showed a significantly lower readmission rate among hospitals with higher follow-up rates.⁸ Similarly, patients with CHF in a large, integrated health system who were seen within 7 days of discharge had an odds ratio (OR) of 0.81 (95% CI, 0.70-0.94) for 30-day readmissions.⁹

Transitions-of-care clinics (TOCC), designed to provide early postdischarge follow-up to high-risk patients, have been shown to reduce 30-day readmission rates,^3,4,10,11 especially in clinics that have same-physician follow-up visits rather than follow-up visits with a community primary care physician (PCP).¹² The most pronounced impact of postdischarge follow-up is seen in high-risk patients with high complexity or high severity of disease; however, complex rural patients are less likely to have access to specialty care.¹³ As a result, since rural residents must travel farther for specialty care, they are seen less frequently than their urban counterparts.^14,15

Prior to our TOCC initiative, the Iowa City VA (ICVA) ranked in the fifth quintile of the Veterans Health Administration (VHA) Strategic Analytics for Improvement and Learning model for hospital-wide readmissions (HWR), meaning that HWR at ICVA were higher than 80% of the other VHA healthcare centers. The low score in this metric was in part due to readmission rates in high-risk populations, including patients with CHF and those with high Care Assessment Need (CAN) scores. One concern was that the ICVA system serves many veterans from rural areas, some of whom must travel up to 200 miles to access inpatient and subspecialty care.

To meet these challenges, we implemented a TOCC to deliver timely postdischarge care focusing on high-risk and high-complexity patients. To address access-to-care issues of patients living in rural areas within the ICVA, we included virtual follow-up visits as a key component of our intervention.^16,17 The aim of this project was to decrease 30-day readmission rates of ICVA patients by 20% within 12 months of implementation.

Setting/Study Population

The ICVA serves 184,000 veterans stretched over 50 counties in eastern Iowa, western Illinois, and northern Missouri, with more than 60% of these patients residing in rural areas. Patients were initially eligible for the TOCC if they had an admission diagnosis of CHF and a CAN score > 85 at the time of discharge. The CAN score, developed by the VA to assess the risk of hospital readmission in individual patients, factors in several variables, including demographics, coexisting conditions, vital signs, utilization of services, pharmacy visits, and laboratory results. Patients in the top 5% (95-99) have a readmission rate of 20% at 90 days. Since the CAN is a proprietary tool, it may not be published in full; however, this assessment tool is commonly used and frequently cited in VA research.^18-22 The CAN score is expressed as a percentile ranging from 0 (lowest risk) to 99 (highest risk). Patient eligibility was expanded during subsequent Plan-Do-Study-Act (PDSA) cycles, as outlined below. Patient eligibility was expanded during subsequent PDSA cycles (also outlined below). A review by a local institutional review board was obtained, and the study was classified as exempt due to the use of deidentified data. Standards for Quality Improvement Reporting Excellence 2.0 guidelines were used to construct the manuscript.

Magnitude Assessment

The numbers of discharges, readmissions within 30 days, emergency department (ED) visits by all discharged veterans, and veterans discharged with a CHF hospital diagnosis were recorded from February 2017 to February 2018, which were the 12 months immediately preceding the pilot implementation.

Intervention

The primary intervention was referral to the newly formed ICVA TOCC. The multidisciplinary TOCC team consisted of hospitalists, pharmacists, schedulers, and discharge planners/care managers. Patients were identified by the hospitalist team during admission; prior to hospital discharge, these patients were referred to TOCC discharge planners to schedule appropriate follow-up appointments. Virtual follow-up visits were conducted using a patient’s home technology; in cases where a patient lacked adequate technology capabilities (eg, no computer or internet access), the ICVA provided a tablet device with cellular internet capability for temporary use. Specific clinical activities included medication reconciliation by a pharmacist, follow-up of pending laboratory studies, imaging studies, pathology results, medical diagnosis education, counseling regarding dietary restrictions, and contingency planning outside of an ED visit in the event of a change in clinical status. In addition, the TOCC aimed to facilitate a smooth transition of care back to the PCP by arranging follow-up appointments, providing visit summaries, and scheduling consults with specialty care, as appropriate.

Measures

The primary objective measure was the 30-day readmission rate in the ICVA hospital. Secondary measures included the number of VHA ED visits within 30 days of discharge. The main process measures were the number of hospital discharges per month, the number of TOCC referrals, the number of TOCC appointments made, the number of virtual and in-person visits, and the percentage of appointment “no-shows.”

Implementation

The TOCC was piloted from April 2018 to October 2018. During the pilot phase, TOCC enrollment was limited to virtual appointments and to patients with an admission diagnosis of CHF and a CAN score of > 85. The TOCC had staff on-site 2 days a week; this included pharmacists to reconcile medications and hospitalists to address follow-up care needs.

The TOCC clinic was temporarily closed at the end of October 2018 to analyze pilot results. Based on stakeholder feedback, changes made as part of the second PDSA cycle included expanding eligibility criteria to any hospital admission diagnosis and to patients with a CAN score < 85 if the hospitalist team felt the patient was likely to benefit from TOCC follow-up. In addition, on-site clinic staffing was expanded from 2 to 5 days per week to improve access, and the option for an in-person visit was added based on concerns some veterans expressed regarding the use of the technology at home. Finally, a formal resident program was added, and the order set for referrals was simplified. The TOCC was restarted in February 2019, and TOCC metrics were reviewed monthly. By July 2019, we identified issues with TOCC referrals and appointment creation that required additional modifications to the intervention.

A third PDSA cycle was initiated in July 2019 and included major changes, notably the formation of a designated TOCC committee. The committee appointed a dedicated TOCC scheduler whose role was to reduce confusion regarding scheduling, to update the discharge instructions/orders template to lower incidences of “double-booking” that occurred with PCP and TOCC appointments, to modify discharge educational instruction regarding virtual visits and tablet use, to adjust the TOCC-PCP handoff, and to formalize interactions between discharge coordinators and residents to review possible referrals every morning (Appendix Figure 1).

Statistical Analysis

Run charts were constructed by plotting monthly primary outcome values and monthly process metrics (Figure, Appendix Figure 2, Appendix Figure 3). Chi-square tests were used to compare 30-day readmission rates before and after the intervention. Statistical modeling was used to determine differences in outcomes between referred patients seen and referred patients not seen by the TOCC. In these statistical models, the outcome measures were 30-day readmissions, 30-day ED visits, and 6-month mortality. Covariates included in the final analysis were age, gender, race, CAN score, rural-urban commuting area code, referral service (resident vs nonresident), and admission diagnosis. Admission diagnoses were sorted by the investigators into one of the following seven categories: cardiac, infectious, pulmonary, gastrointestinal, neurologic, renal, and other.

Mean (SD) or counts and percentages were used to describe the distribution of continuous and categorical variables, respectively. Kruskal-Wallis test, t test, or chi-square tests were used, as appropriate, across categories. Generalized linear models with a logistic link function were used to test for differences between patients who kept their appointment at the TOCC and those who did not keep their TOCC appointment (both unadjusted and adjusted for all of the covariates previously mentioned). In addition, generalized linear models were also used to compare outcomes between TOCC patients seen virtually vs those seen in-person (both unadjusted and adjusted for all the covariates previously mentioned). All statistical tests were considered significant at a two-sided P < .05. All analyses were performed using SAS software version 9.4 (SAS Institute Inc).

Magnitude Assessment

During the preimplementation period (February 2017-February 2018), there were 3014 patient discharges from ICVA and 343 readmissions, resulting in a readmission rate of 11.4%. Among patients with a hospital-admission diagnosis of cardiorespiratory disease, which included patients with CHF, there were 381 discharges and 46 readmissions, resulting in a readmission rate of 12.1%.

Primary Outcome

During the pilot phase, which was conducted from April 2018 to October 2018, 142 patients who met inclusion criteria (CHF diagnosis and a CAN score > 85) were discharged from ICVA, and 56 referrals to the TOCC were placed. The readmission rate among the cardiorespiratory cohort of veterans was 9.5%.

During the expansion of the intervention from February 2019 to February 2020, there were 2844 discharges from the ICVA and 291 readmissions, resulting in a readmission rate of 10.2%. However, there was a further decrease in the readmission rate after the third PDSA cycle was initiated in July 2019 (Appendix Figure 1). The readmission rate was 9.2% in the final 6 months of the intervention period, and 7.9% in the final 3 months. Of note, in the group of 1948 patients who did not meet the eligibility criteria to participate in our study, the readmission rate during the same time period was 8.6% (161 readmissions).

When comparing the 6 months following the third PDSA cycle to the magnitude assessment period, there was a relative readmission reduction of 19.3% (P = .04), and an absolute reduction of 2.2%. If the final 3 months of the intervention period are included, there was an absolute reduction of 3.5% and a relative reduction of 30.7% (P = .01). Notably, before the pilot phase, ICVA was in the fifth quintile for HWR among VA hospitals but improved to the second quintile by the end of the expansion phase.

Process Outcomes

Process metrics for TOCC referrals, the number of patients seen, and the number of virtual and in-person visits over time are shown in Appendix Figure 3. Rates of TOCC referrals and the number of TOCC visits were lower than anticipated during the first 5 months of the intervention. However, TOCC referrals increased significantly after we implemented the previously described changes as part of the third PDSA cycle. As a result, total, virtual, and in-person visits also significantly increased from July 2019 to February 2020. The proportion of patients choosing virtual vs in-person visits fluctuated over time, but virtual visits were generally chosen more often than in-person visits.

Statistical Modeling

Baseline Data

Cohort characteristics are shown in Table 1. The cohort, which reflected the ICVA population, was predominantly male (96%) and White (93%), with a mean age of 67 years. The population was approximately half urban and half rural in composition, and the most common reason for hospital admission was cardiac. Other than a small but statistically significant difference in CAN scores, there were no significant differences between patients who kept their TOCC appointment and those who did not. There were also no differences in baseline characteristics between patients who chose virtual follow-up and patients who chose in-person follow-up, including the proportion of urban and rural patients.

Outcomes

Patients who kept their TOCC appointments had a 30-day readmission rate of 9.6%, which was significantly lower than the 30-day readmission rate of 27% in the group that did not keep their TOCC appointment (P < .001). Similarly, the percentage of patients treated in the ED was 15% in the TOCC group compared to 31.2% in the group that canceled their appointment (P < .001) (Table 1). In the multivariable analysis, patients who were seen in the TOCC group had an OR for 30-day readmission of 0.35 (95% CI, 0.19-0.62, P < .001), and an OR for ED visits of 0.39 (95% CI, 0.23-0.65; P < .001) (Table 2). There was no statistically significant difference in 6-month mortality between the two groups. In the virtual group compared to the in-person group, there were no statistically significant differences in outcomes between the two groups in the unadjusted or adjusted analysis (Table 2).

In this quality improvement initiative, we found that a TOCC targeting high-risk patients and offering virtual follow-up visits significantly decreased the 30-day readmission rates among veterans at ICVA. Statistical comparisons of patients seen at the TOCC vs those not seen at the TOCC showed a dramatic reduction in 30-day readmissions and ED visits. Finally, virtual follow-ups were more popular than in-person visits, and patients who followed up virtually had equivalent outcomes to those with the more traditional follow-up.

In the expansion phase, eligibility was expanded to include any hospital indication but continued to focus on high-risk patients. Existing literature suggests that providing postdischarge care to all patients, including low- or medium-risk patients, may not be as impactful as enrolling high-risk patients only. For instance, a postdischarge clinic offered to all patients at a VA system in Colorado did not reduce readmission rates compared to PCP follow-up.²³ In contrast, a study of more than 10,000 high-risk urban patients demonstrated that postdischarge care resulted in a 9.3% reduction in readmission risk.²⁴ Our data are consistent with the previously published studies, as the average CAN score of patients seen in TOCC was 90, suggesting a high risk of readmission. In the final 12 months of the intervention, 15% of discharged patients were seen at the TOCC clinic, suggesting that targeted intervention within the small subset of high-risk patients was sufficient to achieve our primary aim. Of note, among patients who did not meet the inclusion criteria for TOCC referral (ie, patients not considered high risk [CAN score ≤ 85]), the rate of readmissions was 8.6%.

Most of the available research on the efficacy of postdischarge clinics was conducted in urban environments. Our ICVA population sees a large proportion of rural veterans, who account for just over 50% of the discharge population. In a study of more than 2 million Medicare patients discharged from US hospitals, the 30-day readmission rates and adjusted mortality rates were higher among patients in rural counties, and post–acute care seemed to have a greater impact in rural rather than urban settings.²⁵ Previous studies have demonstrated that virtual visits have the potential to improve readmission rates, especially in patients with CHF²⁶ and in patients at the highest risk for readmission.²⁷ In our study, the pilot phase offered only virtual visits, but we subsequently added an in-person option based on veteran feedback. Interestingly, over the next 12 months, virtual visits were more popular with both urban and rural veterans, and there were no differences in the number of rural patients in the in-person vs the virtual group. These findings suggest factors other than rurality influenced the decision to choose virtual follow-up visits over in-person visits. Future studies should seek to determine the extent to which factors such as age, race, educational level, and socioeconomic circumstances impact veterans’ follow-up decisions. Not only were outcomes among patients who chose virtual visits the same as those of patients who chose in-person visits, but both of these groups had better outcomes compared to the non-TOCC group (Table 2). This finding demonstrating the efficacy of virtual visits among rural and urban patients has taken on increased significance due to the COVID-19 pandemic, as virtual visits offer a safer option, one that minimizes physical contact.

Our quality improvement analysis included a statistical comparison of patients seen vs those not seen at the TOCC. Patients who were referred to the TOCC but chose not to keep their appointment were similar to those seen in TOCC in terms of age, CAN score, rurality, and hospital diagnosis, but readmission rates were substantially higher in this group even after adjustments for covariates (Table 2). Evaluating causality in interventions aimed to reduce hospital readmission rates is complicated.²⁸ Our findings add greater plausibility to the utility of TOCC in accounting for at least a portion of the reported decrease in ICVA 30-day readmissions.

Our study has several strengths, including an observation period longer than 2 years, a large population of discharged veterans within an integrated healthcare system, and a large proportion of patients living in rural areas. Another strength of our study is the innovative nature of the intervention, which features a multidisciplinary team and the option of virtual or in-person visits. Nevertheless, this study also has several important limitations. As a single-center study, our findings may not be generalizable to other institutions, especially those outside the VHA system. Similarly, our study population reflected that of the ICVA, which may limit generalizability to a more diverse population. While we attempted to account in our statistical modeling for baseline differences between referred patients seen by the TOCC and those referred but not seen, we cannot exclude residual confounding between the groups. Specifically, the comparison of patients who did and did not choose TOCC follow-up introduces the possibility of selection bias. Future randomized/controlled studies will need to evaluate whether TOCC is more effective than the standard of care to reduce readmissions. Finally, since the analysis period following the final PDSA cycle was compressed due to the onset of the COVID-19 pandemic in the United States, no data are available regarding the sustained impacts of changes made during this cycle.

A multidisciplinary TOCC within the ICVA, featuring both virtual and in-person visits, reduced 30-day readmission rates by 19.3%; this approach to care was especially effective in patients with CHF. Virtual visits were the follow-up mode of choice for both urban and rural veterans, and there was no difference in outcomes between these two follow-up options. Future studies will focus on additional quality metrics, including cost-effectiveness and patient satisfaction.

Hospital readmissions are a significant problem in the United States, affecting 15% to 30% of discharges and incurring costs of more than $17 billion annually.¹ Timely posthospitalization follow-up visits are critical to ensure the effective transfer of patients to the outpatient setting; such visits reduce readmission rates as well as hospital length of stay and overall health care resource utilization.^2-4 Patients who receive inadequate follow-up care (ie, within 4 weeks of discharge) are significantly more likely to be readmitted than those who receive close follow-up care^.5

Due to the large clinical and financial consequences associated with hospital readmission, a variety of interventions have been studied, including home visits, telemonitoring, medication management, telephone calls, and postdischarge clinics.^6,7 While studies have not shown postdischarge clinics to be universally efficacious in reducing readmission rates, there is increasing evidence of reduced readmission rates in clinics that target high-risk patients (eg, patients with congestive heart failure [CHF]) rather than the total population.² A study by Hernandez et al that evaluated the relationship between early physician follow-up and 30-day readmissions showed a significantly lower readmission rate among hospitals with higher follow-up rates.⁸ Similarly, patients with CHF in a large, integrated health system who were seen within 7 days of discharge had an odds ratio (OR) of 0.81 (95% CI, 0.70-0.94) for 30-day readmissions.⁹

Transitions-of-care clinics (TOCC), designed to provide early postdischarge follow-up to high-risk patients, have been shown to reduce 30-day readmission rates,^3,4,10,11 especially in clinics that have same-physician follow-up visits rather than follow-up visits with a community primary care physician (PCP).¹² The most pronounced impact of postdischarge follow-up is seen in high-risk patients with high complexity or high severity of disease; however, complex rural patients are less likely to have access to specialty care.¹³ As a result, since rural residents must travel farther for specialty care, they are seen less frequently than their urban counterparts.^14,15

Prior to our TOCC initiative, the Iowa City VA (ICVA) ranked in the fifth quintile of the Veterans Health Administration (VHA) Strategic Analytics for Improvement and Learning model for hospital-wide readmissions (HWR), meaning that HWR at ICVA were higher than 80% of the other VHA healthcare centers. The low score in this metric was in part due to readmission rates in high-risk populations, including patients with CHF and those with high Care Assessment Need (CAN) scores. One concern was that the ICVA system serves many veterans from rural areas, some of whom must travel up to 200 miles to access inpatient and subspecialty care.

To meet these challenges, we implemented a TOCC to deliver timely postdischarge care focusing on high-risk and high-complexity patients. To address access-to-care issues of patients living in rural areas within the ICVA, we included virtual follow-up visits as a key component of our intervention.^16,17 The aim of this project was to decrease 30-day readmission rates of ICVA patients by 20% within 12 months of implementation.

Setting/Study Population

The ICVA serves 184,000 veterans stretched over 50 counties in eastern Iowa, western Illinois, and northern Missouri, with more than 60% of these patients residing in rural areas. Patients were initially eligible for the TOCC if they had an admission diagnosis of CHF and a CAN score > 85 at the time of discharge. The CAN score, developed by the VA to assess the risk of hospital readmission in individual patients, factors in several variables, including demographics, coexisting conditions, vital signs, utilization of services, pharmacy visits, and laboratory results. Patients in the top 5% (95-99) have a readmission rate of 20% at 90 days. Since the CAN is a proprietary tool, it may not be published in full; however, this assessment tool is commonly used and frequently cited in VA research.^18-22 The CAN score is expressed as a percentile ranging from 0 (lowest risk) to 99 (highest risk). Patient eligibility was expanded during subsequent Plan-Do-Study-Act (PDSA) cycles, as outlined below. Patient eligibility was expanded during subsequent PDSA cycles (also outlined below). A review by a local institutional review board was obtained, and the study was classified as exempt due to the use of deidentified data. Standards for Quality Improvement Reporting Excellence 2.0 guidelines were used to construct the manuscript.

Magnitude Assessment

The numbers of discharges, readmissions within 30 days, emergency department (ED) visits by all discharged veterans, and veterans discharged with a CHF hospital diagnosis were recorded from February 2017 to February 2018, which were the 12 months immediately preceding the pilot implementation.

Intervention

The primary intervention was referral to the newly formed ICVA TOCC. The multidisciplinary TOCC team consisted of hospitalists, pharmacists, schedulers, and discharge planners/care managers. Patients were identified by the hospitalist team during admission; prior to hospital discharge, these patients were referred to TOCC discharge planners to schedule appropriate follow-up appointments. Virtual follow-up visits were conducted using a patient’s home technology; in cases where a patient lacked adequate technology capabilities (eg, no computer or internet access), the ICVA provided a tablet device with cellular internet capability for temporary use. Specific clinical activities included medication reconciliation by a pharmacist, follow-up of pending laboratory studies, imaging studies, pathology results, medical diagnosis education, counseling regarding dietary restrictions, and contingency planning outside of an ED visit in the event of a change in clinical status. In addition, the TOCC aimed to facilitate a smooth transition of care back to the PCP by arranging follow-up appointments, providing visit summaries, and scheduling consults with specialty care, as appropriate.

Measures

The primary objective measure was the 30-day readmission rate in the ICVA hospital. Secondary measures included the number of VHA ED visits within 30 days of discharge. The main process measures were the number of hospital discharges per month, the number of TOCC referrals, the number of TOCC appointments made, the number of virtual and in-person visits, and the percentage of appointment “no-shows.”

Implementation

The TOCC was piloted from April 2018 to October 2018. During the pilot phase, TOCC enrollment was limited to virtual appointments and to patients with an admission diagnosis of CHF and a CAN score of > 85. The TOCC had staff on-site 2 days a week; this included pharmacists to reconcile medications and hospitalists to address follow-up care needs.

The TOCC clinic was temporarily closed at the end of October 2018 to analyze pilot results. Based on stakeholder feedback, changes made as part of the second PDSA cycle included expanding eligibility criteria to any hospital admission diagnosis and to patients with a CAN score < 85 if the hospitalist team felt the patient was likely to benefit from TOCC follow-up. In addition, on-site clinic staffing was expanded from 2 to 5 days per week to improve access, and the option for an in-person visit was added based on concerns some veterans expressed regarding the use of the technology at home. Finally, a formal resident program was added, and the order set for referrals was simplified. The TOCC was restarted in February 2019, and TOCC metrics were reviewed monthly. By July 2019, we identified issues with TOCC referrals and appointment creation that required additional modifications to the intervention.

A third PDSA cycle was initiated in July 2019 and included major changes, notably the formation of a designated TOCC committee. The committee appointed a dedicated TOCC scheduler whose role was to reduce confusion regarding scheduling, to update the discharge instructions/orders template to lower incidences of “double-booking” that occurred with PCP and TOCC appointments, to modify discharge educational instruction regarding virtual visits and tablet use, to adjust the TOCC-PCP handoff, and to formalize interactions between discharge coordinators and residents to review possible referrals every morning (Appendix Figure 1).

Statistical Analysis

Run charts were constructed by plotting monthly primary outcome values and monthly process metrics (Figure, Appendix Figure 2, Appendix Figure 3). Chi-square tests were used to compare 30-day readmission rates before and after the intervention. Statistical modeling was used to determine differences in outcomes between referred patients seen and referred patients not seen by the TOCC. In these statistical models, the outcome measures were 30-day readmissions, 30-day ED visits, and 6-month mortality. Covariates included in the final analysis were age, gender, race, CAN score, rural-urban commuting area code, referral service (resident vs nonresident), and admission diagnosis. Admission diagnoses were sorted by the investigators into one of the following seven categories: cardiac, infectious, pulmonary, gastrointestinal, neurologic, renal, and other.

Mean (SD) or counts and percentages were used to describe the distribution of continuous and categorical variables, respectively. Kruskal-Wallis test, t test, or chi-square tests were used, as appropriate, across categories. Generalized linear models with a logistic link function were used to test for differences between patients who kept their appointment at the TOCC and those who did not keep their TOCC appointment (both unadjusted and adjusted for all of the covariates previously mentioned). In addition, generalized linear models were also used to compare outcomes between TOCC patients seen virtually vs those seen in-person (both unadjusted and adjusted for all the covariates previously mentioned). All statistical tests were considered significant at a two-sided P < .05. All analyses were performed using SAS software version 9.4 (SAS Institute Inc).

Magnitude Assessment

During the preimplementation period (February 2017-February 2018), there were 3014 patient discharges from ICVA and 343 readmissions, resulting in a readmission rate of 11.4%. Among patients with a hospital-admission diagnosis of cardiorespiratory disease, which included patients with CHF, there were 381 discharges and 46 readmissions, resulting in a readmission rate of 12.1%.

Primary Outcome

During the pilot phase, which was conducted from April 2018 to October 2018, 142 patients who met inclusion criteria (CHF diagnosis and a CAN score > 85) were discharged from ICVA, and 56 referrals to the TOCC were placed. The readmission rate among the cardiorespiratory cohort of veterans was 9.5%.

During the expansion of the intervention from February 2019 to February 2020, there were 2844 discharges from the ICVA and 291 readmissions, resulting in a readmission rate of 10.2%. However, there was a further decrease in the readmission rate after the third PDSA cycle was initiated in July 2019 (Appendix Figure 1). The readmission rate was 9.2% in the final 6 months of the intervention period, and 7.9% in the final 3 months. Of note, in the group of 1948 patients who did not meet the eligibility criteria to participate in our study, the readmission rate during the same time period was 8.6% (161 readmissions).

When comparing the 6 months following the third PDSA cycle to the magnitude assessment period, there was a relative readmission reduction of 19.3% (P = .04), and an absolute reduction of 2.2%. If the final 3 months of the intervention period are included, there was an absolute reduction of 3.5% and a relative reduction of 30.7% (P = .01). Notably, before the pilot phase, ICVA was in the fifth quintile for HWR among VA hospitals but improved to the second quintile by the end of the expansion phase.

Process Outcomes

Process metrics for TOCC referrals, the number of patients seen, and the number of virtual and in-person visits over time are shown in Appendix Figure 3. Rates of TOCC referrals and the number of TOCC visits were lower than anticipated during the first 5 months of the intervention. However, TOCC referrals increased significantly after we implemented the previously described changes as part of the third PDSA cycle. As a result, total, virtual, and in-person visits also significantly increased from July 2019 to February 2020. The proportion of patients choosing virtual vs in-person visits fluctuated over time, but virtual visits were generally chosen more often than in-person visits.

Statistical Modeling

Baseline Data

Cohort characteristics are shown in Table 1. The cohort, which reflected the ICVA population, was predominantly male (96%) and White (93%), with a mean age of 67 years. The population was approximately half urban and half rural in composition, and the most common reason for hospital admission was cardiac. Other than a small but statistically significant difference in CAN scores, there were no significant differences between patients who kept their TOCC appointment and those who did not. There were also no differences in baseline characteristics between patients who chose virtual follow-up and patients who chose in-person follow-up, including the proportion of urban and rural patients.

Outcomes

Patients who kept their TOCC appointments had a 30-day readmission rate of 9.6%, which was significantly lower than the 30-day readmission rate of 27% in the group that did not keep their TOCC appointment (P < .001). Similarly, the percentage of patients treated in the ED was 15% in the TOCC group compared to 31.2% in the group that canceled their appointment (P < .001) (Table 1). In the multivariable analysis, patients who were seen in the TOCC group had an OR for 30-day readmission of 0.35 (95% CI, 0.19-0.62, P < .001), and an OR for ED visits of 0.39 (95% CI, 0.23-0.65; P < .001) (Table 2). There was no statistically significant difference in 6-month mortality between the two groups. In the virtual group compared to the in-person group, there were no statistically significant differences in outcomes between the two groups in the unadjusted or adjusted analysis (Table 2).

In this quality improvement initiative, we found that a TOCC targeting high-risk patients and offering virtual follow-up visits significantly decreased the 30-day readmission rates among veterans at ICVA. Statistical comparisons of patients seen at the TOCC vs those not seen at the TOCC showed a dramatic reduction in 30-day readmissions and ED visits. Finally, virtual follow-ups were more popular than in-person visits, and patients who followed up virtually had equivalent outcomes to those with the more traditional follow-up.

In the expansion phase, eligibility was expanded to include any hospital indication but continued to focus on high-risk patients. Existing literature suggests that providing postdischarge care to all patients, including low- or medium-risk patients, may not be as impactful as enrolling high-risk patients only. For instance, a postdischarge clinic offered to all patients at a VA system in Colorado did not reduce readmission rates compared to PCP follow-up.²³ In contrast, a study of more than 10,000 high-risk urban patients demonstrated that postdischarge care resulted in a 9.3% reduction in readmission risk.²⁴ Our data are consistent with the previously published studies, as the average CAN score of patients seen in TOCC was 90, suggesting a high risk of readmission. In the final 12 months of the intervention, 15% of discharged patients were seen at the TOCC clinic, suggesting that targeted intervention within the small subset of high-risk patients was sufficient to achieve our primary aim. Of note, among patients who did not meet the inclusion criteria for TOCC referral (ie, patients not considered high risk [CAN score ≤ 85]), the rate of readmissions was 8.6%.

Most of the available research on the efficacy of postdischarge clinics was conducted in urban environments. Our ICVA population sees a large proportion of rural veterans, who account for just over 50% of the discharge population. In a study of more than 2 million Medicare patients discharged from US hospitals, the 30-day readmission rates and adjusted mortality rates were higher among patients in rural counties, and post–acute care seemed to have a greater impact in rural rather than urban settings.²⁵ Previous studies have demonstrated that virtual visits have the potential to improve readmission rates, especially in patients with CHF²⁶ and in patients at the highest risk for readmission.²⁷ In our study, the pilot phase offered only virtual visits, but we subsequently added an in-person option based on veteran feedback. Interestingly, over the next 12 months, virtual visits were more popular with both urban and rural veterans, and there were no differences in the number of rural patients in the in-person vs the virtual group. These findings suggest factors other than rurality influenced the decision to choose virtual follow-up visits over in-person visits. Future studies should seek to determine the extent to which factors such as age, race, educational level, and socioeconomic circumstances impact veterans’ follow-up decisions. Not only were outcomes among patients who chose virtual visits the same as those of patients who chose in-person visits, but both of these groups had better outcomes compared to the non-TOCC group (Table 2). This finding demonstrating the efficacy of virtual visits among rural and urban patients has taken on increased significance due to the COVID-19 pandemic, as virtual visits offer a safer option, one that minimizes physical contact.

Our quality improvement analysis included a statistical comparison of patients seen vs those not seen at the TOCC. Patients who were referred to the TOCC but chose not to keep their appointment were similar to those seen in TOCC in terms of age, CAN score, rurality, and hospital diagnosis, but readmission rates were substantially higher in this group even after adjustments for covariates (Table 2). Evaluating causality in interventions aimed to reduce hospital readmission rates is complicated.²⁸ Our findings add greater plausibility to the utility of TOCC in accounting for at least a portion of the reported decrease in ICVA 30-day readmissions.

Our study has several strengths, including an observation period longer than 2 years, a large population of discharged veterans within an integrated healthcare system, and a large proportion of patients living in rural areas. Another strength of our study is the innovative nature of the intervention, which features a multidisciplinary team and the option of virtual or in-person visits. Nevertheless, this study also has several important limitations. As a single-center study, our findings may not be generalizable to other institutions, especially those outside the VHA system. Similarly, our study population reflected that of the ICVA, which may limit generalizability to a more diverse population. While we attempted to account in our statistical modeling for baseline differences between referred patients seen by the TOCC and those referred but not seen, we cannot exclude residual confounding between the groups. Specifically, the comparison of patients who did and did not choose TOCC follow-up introduces the possibility of selection bias. Future randomized/controlled studies will need to evaluate whether TOCC is more effective than the standard of care to reduce readmissions. Finally, since the analysis period following the final PDSA cycle was compressed due to the onset of the COVID-19 pandemic in the United States, no data are available regarding the sustained impacts of changes made during this cycle.

A multidisciplinary TOCC within the ICVA, featuring both virtual and in-person visits, reduced 30-day readmission rates by 19.3%; this approach to care was especially effective in patients with CHF. Virtual visits were the follow-up mode of choice for both urban and rural veterans, and there was no difference in outcomes between these two follow-up options. Future studies will focus on additional quality metrics, including cost-effectiveness and patient satisfaction.

Type 1 diabetes mellitus (T1D) is a common chronic condition of childhood. Its incidence has risen steadily over the past two decades.¹ Treatment requires complex daily tasks, including blood glucose monitoring and insulin administration. The potential long-term complications of T1D—kidney failure, retinopathy, cardiovascular disease, and death—are grave but can be attenuated with effective glycemic management.² Still, less than 25% of youths aged 13 to 19 years achieve target hemoglobin A1c (HbA1c) levels.³

Diabetic ketoacidosis (DKA) is an acute, life-threatening complication of T1D associated with suboptimal glycemic management. In the United States, DKA hospitalizations increased during 2009-2014.⁴ One study found the average standardized cost of a DKA-related pediatric hospitalization exceeded $7000,⁵ an amount not including missed school days, parent/caregiver workdays, and social and family disruptions. Costs extend to psychological well-being, with patients reporting that they “keep thinking [DKA] may happen again.”⁶

The burden of T1D-related morbidity is not equitably distributed. Recent evidence suggests that racial/ethnic minorities and those with public insurance experience disproportionately high rates of DKA.^7,8 Neighborhood socioeconomic measures, like poverty, add an important dimension illustrative of ecological conditions in which patients reside. These measures have been studied in relation to hospitalization rates for chronic conditions like asthma and heart failure.^9,10 For example, one recent study found links between area-level poverty in the United States and the likelihood of readmissions for pediatric patients following an admission for DKA.¹¹ We sought to build on this finding by investigating whether area-level poverty, which we measured as the census tract poverty rate, patient race, and insurance status were associated with the likelihood of initial DKA hospitalization and severity of DKA presentation.

Design

We conducted a single-center, retrospective cohort study that examined data on youth with T1D extracted from the Cincinnati Children’s Hospital Medical Center (CCHMC) electronic medical record (EMR). CCHMC is an urban, tertiary-care, freestanding pediatric hospital, with near-complete market penetration in Hamilton County, Ohio, particularly for subspecialty care. The racial and ethnic breakdown of Hamilton County’s general population is 68% White, 27% Black/African American, 3% Asian, and 4% Hispanic or Latino (adds to more than 100% due to rounding).¹² The CCHMC Institutional Review Board approved this study.

Study Population

We identified eligible patients through a clinical registry within the EMR, which includes all T1D patients seen at CCHMC within the preceding 24 months. Patients are added to this registry upon initial T1D diagnosis by a research nurse who evaluates autoantibody status and clinical presentation. All patients ≤18 years old who had T1D, were active on the registry, and had an address within Hamilton County as of December 31, 2017, were included. Those with type 2 diabetes mellitus or secondary causes of diabetes were excluded. We captured patients on the registry as of December 31, 2017, and then examined whether they had been hospitalized from January 1, 2011, to December 31, 2017.

Covariates

We extracted demographic and clinical data for patients within the T1D registry as of December 31, 2017, and, for those hospitalized, at the time of their admission. Specifically, we extracted date of birth (to calculate age at encounter), sex, and ethnicity. Age was treated as a continuous variable. We defined ethnicity as Latinx (inclusive of Hispanic) or non-Latinx. We calculated duration of T1D diagnosis for patients within the registry by taking the difference of the diabetes onset date and the end of the study period. We extracted whether registry patients had an insulin pump or a continuous glucose monitor (CGM). Hospitalized patients were identified as new-onset or established T1D by comparing their hospitalization with the T1D diagnosis onset date.

For the entire T1D registry, we also assessed a key clinical characteristic: a patient’s median HbA1c. We enumerated the median HbA1c of all patients’ medians within the calendar year 2017, consistent with the methodology of a large international T1D registry.¹³ Among the subset of patients who were hospitalized, we obtained the HbA1c value during or prior to hospitalization.

Exposures

We captured exposure variables from EMR documentation of patients’ address, race, and insurance status. Given the small regional population of Latinx patients, we did not include ethnicity as an exposure. We used the address and insurance status at the beginning of the study period for established T1D patients, or at the time of diagnosis for new-onset patients, for analyses related to the likelihood of DKA admission. We used the address at the time of hospitalization for analyses of DKA presentation severity. We geocoded home addresses using ArcGIS software (Version 10.5.1; Esri), successfully geocoding >95% to the street level and assigning patients to a corresponding census tract—a census-defined geography consisting of approximately 4000 people.¹⁴ Census tracts, when drawn after the decennial (every 10 years) census, are designed to be sociodemographically homogenous¹⁵ and have long been used in medical and public health research.¹⁶ We connected geocoded addresses to the census tract poverty rate, defined as the percentage of individuals within a tract living below the federal poverty level. For our analyses, we primarily treated census tract poverty as a continuous variable ranging from 0% to 100%. We visualized the distribution of census tract poverty across Hamilton County using a choropleth map (Figure, part A). We also grouped the registry population into quartiles, ordered by census tract poverty levels, to ease the graphical display for Figure, part B. We defined race as White, Black (inclusive of African American), or other. We categorized insurance as Medicaid/public or Non-Medicaid/private.

Outcomes

The primary outcome was a hospitalization for DKA during the study period. We identified eligible events by admission diagnosis of DKA (International Classification of Disease, 10th Revision, Clinical Modification, E13.10 or E10.10). Secondarily, we assessed admission severity for each patient’s first DKA hospitalization using initial pH from a venous blood gas and initial bicarbonate from the basic metabolic panel. These measures were chosen in accordance with the American Diabetes Association’s published classification of DKA severity, which uses pH, bicarbonate, and mental status.¹⁷ We defined the time for correction as the minutes from admission time until the anion gap was ≤12 or the bicarbonate was ≥18. We calculated total inpatient bed-days and pediatric intensive care unit (PICU) bed-days, and identified whether a patient was readmitted for DKA during the study period.

Statistical Analyses

We used descriptive statistics to characterize demographic and clinical information for patients within the T1D registry and for the hospitalized subset. We examined DKA severity data for all admitted patients by diagnosis status (new-onset T1D vs established). Associations between categorical outcomes were assessed using the Chi-square or Fisher exact test, and continuous outcomes were evaluated using the Mann-Whitney Rank Sum test.

We individually examined the effects of census tract poverty, race, and insurance on the odds of hospitalization through logistic regression. After confirming overlap and variability in census tract poverty rates by race and admission status (Appendix Table 1), we included these exposures in a multivariable model to capture their independent effects. Because the impact of census tract poverty, race, and insurance on hospitalization odds could vary depending on T1D diagnosis status, we conducted a sensitivity analysis by repeating the multivariable model with only established T1D patients.

Given that median HbA1c levels differed for T1D registry patients based on census tract poverty, race, and insurance, we used linear regression in a post-hoc analysis to evaluate associations between those exposures and median HbA1c. Age, gender, ethnicity, and duration of diabetes were not included in the models given that these variables did not differ by outcome variables. Diabetes technology—CGM and insulin pumps—was not included as a confounder because we hypothesized it mediated the effect of area-level poverty. For analyses, SigmaPlot (Version 14.0; SPSS Inc.) and R statistical software were used (The R Project for Statistical Computing).

A total of 439 patients were in the T1D registry and lived in Hamilton County as of December 31, 2017. Their demographic characteristics are listed in Table 1. Their median age was 14 years; 48% were female. The median poverty rate of the census tracts in which these youth lived was 11%. A total of 24.6% of patients identified as Black, 73.1% as White, and 2.3% as Hispanic; 35.8% were publicly insured. Patients had a median duration of T1D of 61 months (5 years, 1 month), with a median HbA1c (of all patients’ medians) of 8.7%. A total of 58.1% had a CGM, and 56% used an insulin pump.

Hospitalization Characteristics of Those in the T1D Registry

Approximately one-third of registry patients (n = 152) experienced ≥1 hospitalization for DKA during the study period (inclusive of new-onset and established diagnosis). Age, gender, ethnicity, and duration of T1D diagnosis were similar between those who experienced a hospitalization and those who did not. The median census tract poverty rate was higher among those who were hospitalized compared to those who were not (15% vs 8%, P < .01). Among hospitalized patients, 41% identified as Black; among all who were not hospitalized, 16% identified as Black (P < .01). Among all hospitalized patients, 56.6% were covered by public insurance, while 24.7% of non-hospitalized patients were covered by public insurance (P < .01). Hospitalized patients were more likely to have a higher median HbA1c (9.5% vs 8.4%, P < .01) and were significantly less likely to have a CGM or insulin pump (both P < .01). These associations were magnified among the roughly 10% of the T1D registry population (n = 42) who experienced ≥2 hospitalizations.

Figure, part A is a choropleth map depicting the distribution of poverty by census tract across Hamilton County. The map is complemented by part B of the Figure, which displays the T1D registry population split into poverty quartiles and each quartile’s DKA admission rate per T1D registry population. As census tract poverty increases, so too does the rate of DKA hospitalizations. There is an almost threefold difference between the hospitalization rate of the highest poverty quartile and that of the lowest.

Likelihood of Hospitalization

We examined the exposures of interest—census tract poverty, race, and insurance—and odds of hospitalization at a patient level using logistic regression models (Table 2). In the multivariable model that included these exposures, we found that for every 10% increase in the census tract poverty rate, the odds of being hospitalized for DKA increased by 22% (95% CI, 1.03-1.47). Race was not significantly associated with odds of DKA hospitalization in the multivariable model. Public insurance was associated with a 2.71-times higher odds of hospitalization compared to those with private insurance (95% CI, 1.62-4.55). Our sensitivity analysis of only those with established T1D diagnoses showed similar results to that of the entire T1D registry population.

Severity of DKA Presentation

A total of 152 registry patients were admitted; 89 patients had new-onset T1D, and 63 had an established T1D diagnosis. We examined presenting factors (initial pH, initial bicarbonate, HbA1c) and clinical characteristics (time to correction, inpatient bed-days, PICU bed-days) by census tract poverty, race, and insurance for all hospitalized patients. There were no significant differences in presenting factors or outcomes by census tract poverty or insurance. However, we did note certain differences by race. Indeed, though initial pH, bicarbonate, and time to correction did not meaningfully differ by race, Black patients had a longer length of stay than their White peers (Appendix Table 2; 3.06 vs 2.16 days, P < .01). We further examined this by diagnosis status, as newly diagnosed patients often stay longer for diabetes education.¹⁸ Black patients had a longer length of stay than White patients, regardless of whether they had new-onset (3.85 vs 2.99 days, P < .01) or established T1D (1.83 vs 0.97 days, P < .01). Additionally, we found that the HbA1c was significantly higher in Black versus White patients (12.4% vs 10.8%, P = .01). HbA1c was significantly different between established T1D patients identifying as Black and those identifying as White (11.7% vs 9.5%, P < .01); however, HbA1c did not differ by race for newly diagnosed patients (12.5% vs 13.0%, P = .63).

Median HbA1c of All T1D Registry Patients

Given the suggestion of HbA1c variability by exposure variables, we performed a post-hoc linear regression analysis. Table 3 displays the linear regression models assessing associations of census tract poverty, race, and insurance on median HbA1c for all T1D registry patients. In the multivariable model, census tract poverty was no longer associated with higher HbA1c levels (0.13, 95% CI, –0.01 to 0.26). However, Black patients continued to have significantly higher HbA1c levels than White patients (1.09% higher; 95% CI, 0.59-1.59). Also, those with public insurance had significantly higher HbA1c levels than those with private insurance (0.93% higher; 95% CI, 0.51-1.35).

Living in high poverty areas, identifying as Black, and having public insurance were each associated with a higher likelihood of hospitalization for DKA in an unadjusted model. When we examined these exposures together in a multivariable model, census tract poverty and insurance remained associated with a higher likelihood of hospitalization for DKA. Race—with a wide confidence interval—was no longer associated with a higher likelihood of hospitalization. The increased likelihood of hospitalization initially seen for Black patients may be because they are more likely to reside in high poverty areas and be on public insurance—inequities associated with structural racism.

These results mirror those of other studies that have found area-level poverty to be associated with higher rates of intensive care admissions and hospitalizations for conditions like asthma.^10,19 Our findings are also similar to those of a previous study that found that area-level deprivation was associated with a greater likelihood of DKA readmissions in pediatric and adult patients, as well as an adult study that showed increased neighborhood deprivation was associated with higher odds of DKA hospitalization.^11,20 Further examination of the mechanisms underlying DKA hospitalization is clearly warranted and would be informed by a deeper awareness of the social determinants of health—such as community context and the neighborhood/built environment²¹—and their links to racial inequities.

Neighborhood socioeconomic status, here measured as census tract poverty, is likely linked to hospitalization for DKA through contextual factors like food insecurity, limited access to supportive care (eg, presence of a school nurse who can co-manage T1D with the patient/family and medical team), and adverse environmental exposures.^22-24 To improve the health of patients, prevent hospitalizations, and achieve health equity, we suggest that both individual factors and these contextual factors need to be evaluated and addressed through community-connected interventions.^25,26 For example, interventions that deploy community health workers to navigate the complexities of health systems have been effective in reducing HbA1c levels in adults with type 2 diabetes mellitus²⁷ and are beginning to be studied in pediatrics.²⁸

Black patients had a higher median HbA1c compared to their White counterparts in the T1D registry. One study has previously shown that hemoglobin glycation biologically differed by around 0.5% in Black patients compared to White patients.²⁹ Despite this purported biological difference, a greater difference in HbA1c—nearly 2 points—was seen in the univariate model, illustrating that contextual factors such as socioeconomic status and treatment patterns, like the use of diabetes technology, exceed this biological difference.^30-32 Our multivariable results support the importance of these contextual factors, as the HbA1c difference decreased from 1.9 to 1.09 after census tract poverty and insurance were included.

Such differences by race also emerge from experiences of discrimination and bias. Thus, to improve outcomes and equity, we must understand and address racism and its many ill effects. Different levels of racism, such as structural or personally mediated racism, interact with social determinants of health in various ways.^33,34 For example, redlining—racially discriminatory grading of neighborhoods’ creditworthiness in the 1930s—is an example of structural racism that continues to affect the health of patients today.^35,36 Black families are more likely to live in areas with higher poverty, even after controlling for household income level.³⁷ The interaction between race and poverty is complex given these historical underpinnings of structural racism, as evidenced by race no longer being associated with an increased likelihood of DKA hospitalization in our multivariable analysis. The longer length of stay for Black patients despite similar presenting severity also exemplifies this complexity. For example, longer stays could result from explicit or implicit biases that lead providers to keep patients longer due to preconceived beliefs that the caregiver does not understand diabetes management.³⁸ However, it also could be that the caregiver takes longer to complete diabetes education due to systemic barriers, such as work-related obligations during business hours when diabetes educators are present, or transportation and childcare needs.³⁹ These barriers are often tied to socioeconomic status, making it more difficult for those in poverty to complete required education. We cannot conclude from our study the reasons for the difference in length of stay, just that differences exist. Moving forward, we suggest that design of interventions aimed at achieving health equity consider the complex interplay between health, healthcare systems, racism, and neighborhood context.

There were limitations to our study. First, given that this was an observational study, unmeasured confounders like patient or parental education and personal income could affect the strength of associations seen in our models. Second, although the CCHMC’s Diabetes Center treats nearly all children with T1D in Hamilton County, some patients, or hospitalizations at other centers, could have been missed. Third, the demographics of Hamilton County may not generalize to other diabetes centers. Fourth, although more granular data—such as census block groups or blocks—are available from the US Census Bureau, we chose census tracts as our unit of analysis to reduce sampling error that emerges from the use of such smaller geographies. Last, we measured census tract poverty rather than other contextual markers of one’s neighborhood, like vacant housing or the proportion of those receiving public benefits. However, these measures are often highly correlated with census tract poverty.⁴⁰

Type 1 diabetes mellitus (T1D) is a common chronic condition of childhood. Its incidence has risen steadily over the past two decades.¹ Treatment requires complex daily tasks, including blood glucose monitoring and insulin administration. The potential long-term complications of T1D—kidney failure, retinopathy, cardiovascular disease, and death—are grave but can be attenuated with effective glycemic management.² Still, less than 25% of youths aged 13 to 19 years achieve target hemoglobin A1c (HbA1c) levels.³

Diabetic ketoacidosis (DKA) is an acute, life-threatening complication of T1D associated with suboptimal glycemic management. In the United States, DKA hospitalizations increased during 2009-2014.⁴ One study found the average standardized cost of a DKA-related pediatric hospitalization exceeded $7000,⁵ an amount not including missed school days, parent/caregiver workdays, and social and family disruptions. Costs extend to psychological well-being, with patients reporting that they “keep thinking [DKA] may happen again.”⁶

The burden of T1D-related morbidity is not equitably distributed. Recent evidence suggests that racial/ethnic minorities and those with public insurance experience disproportionately high rates of DKA.^7,8 Neighborhood socioeconomic measures, like poverty, add an important dimension illustrative of ecological conditions in which patients reside. These measures have been studied in relation to hospitalization rates for chronic conditions like asthma and heart failure.^9,10 For example, one recent study found links between area-level poverty in the United States and the likelihood of readmissions for pediatric patients following an admission for DKA.¹¹ We sought to build on this finding by investigating whether area-level poverty, which we measured as the census tract poverty rate, patient race, and insurance status were associated with the likelihood of initial DKA hospitalization and severity of DKA presentation.

Design

We conducted a single-center, retrospective cohort study that examined data on youth with T1D extracted from the Cincinnati Children’s Hospital Medical Center (CCHMC) electronic medical record (EMR). CCHMC is an urban, tertiary-care, freestanding pediatric hospital, with near-complete market penetration in Hamilton County, Ohio, particularly for subspecialty care. The racial and ethnic breakdown of Hamilton County’s general population is 68% White, 27% Black/African American, 3% Asian, and 4% Hispanic or Latino (adds to more than 100% due to rounding).¹² The CCHMC Institutional Review Board approved this study.

Study Population

We identified eligible patients through a clinical registry within the EMR, which includes all T1D patients seen at CCHMC within the preceding 24 months. Patients are added to this registry upon initial T1D diagnosis by a research nurse who evaluates autoantibody status and clinical presentation. All patients ≤18 years old who had T1D, were active on the registry, and had an address within Hamilton County as of December 31, 2017, were included. Those with type 2 diabetes mellitus or secondary causes of diabetes were excluded. We captured patients on the registry as of December 31, 2017, and then examined whether they had been hospitalized from January 1, 2011, to December 31, 2017.

Covariates

We extracted demographic and clinical data for patients within the T1D registry as of December 31, 2017, and, for those hospitalized, at the time of their admission. Specifically, we extracted date of birth (to calculate age at encounter), sex, and ethnicity. Age was treated as a continuous variable. We defined ethnicity as Latinx (inclusive of Hispanic) or non-Latinx. We calculated duration of T1D diagnosis for patients within the registry by taking the difference of the diabetes onset date and the end of the study period. We extracted whether registry patients had an insulin pump or a continuous glucose monitor (CGM). Hospitalized patients were identified as new-onset or established T1D by comparing their hospitalization with the T1D diagnosis onset date.

For the entire T1D registry, we also assessed a key clinical characteristic: a patient’s median HbA1c. We enumerated the median HbA1c of all patients’ medians within the calendar year 2017, consistent with the methodology of a large international T1D registry.¹³ Among the subset of patients who were hospitalized, we obtained the HbA1c value during or prior to hospitalization.

Exposures

We captured exposure variables from EMR documentation of patients’ address, race, and insurance status. Given the small regional population of Latinx patients, we did not include ethnicity as an exposure. We used the address and insurance status at the beginning of the study period for established T1D patients, or at the time of diagnosis for new-onset patients, for analyses related to the likelihood of DKA admission. We used the address at the time of hospitalization for analyses of DKA presentation severity. We geocoded home addresses using ArcGIS software (Version 10.5.1; Esri), successfully geocoding >95% to the street level and assigning patients to a corresponding census tract—a census-defined geography consisting of approximately 4000 people.¹⁴ Census tracts, when drawn after the decennial (every 10 years) census, are designed to be sociodemographically homogenous¹⁵ and have long been used in medical and public health research.¹⁶ We connected geocoded addresses to the census tract poverty rate, defined as the percentage of individuals within a tract living below the federal poverty level. For our analyses, we primarily treated census tract poverty as a continuous variable ranging from 0% to 100%. We visualized the distribution of census tract poverty across Hamilton County using a choropleth map (Figure, part A). We also grouped the registry population into quartiles, ordered by census tract poverty levels, to ease the graphical display for Figure, part B. We defined race as White, Black (inclusive of African American), or other. We categorized insurance as Medicaid/public or Non-Medicaid/private.

Outcomes

The primary outcome was a hospitalization for DKA during the study period. We identified eligible events by admission diagnosis of DKA (International Classification of Disease, 10th Revision, Clinical Modification, E13.10 or E10.10). Secondarily, we assessed admission severity for each patient’s first DKA hospitalization using initial pH from a venous blood gas and initial bicarbonate from the basic metabolic panel. These measures were chosen in accordance with the American Diabetes Association’s published classification of DKA severity, which uses pH, bicarbonate, and mental status.¹⁷ We defined the time for correction as the minutes from admission time until the anion gap was ≤12 or the bicarbonate was ≥18. We calculated total inpatient bed-days and pediatric intensive care unit (PICU) bed-days, and identified whether a patient was readmitted for DKA during the study period.

Statistical Analyses

We used descriptive statistics to characterize demographic and clinical information for patients within the T1D registry and for the hospitalized subset. We examined DKA severity data for all admitted patients by diagnosis status (new-onset T1D vs established). Associations between categorical outcomes were assessed using the Chi-square or Fisher exact test, and continuous outcomes were evaluated using the Mann-Whitney Rank Sum test.

We individually examined the effects of census tract poverty, race, and insurance on the odds of hospitalization through logistic regression. After confirming overlap and variability in census tract poverty rates by race and admission status (Appendix Table 1), we included these exposures in a multivariable model to capture their independent effects. Because the impact of census tract poverty, race, and insurance on hospitalization odds could vary depending on T1D diagnosis status, we conducted a sensitivity analysis by repeating the multivariable model with only established T1D patients.

Given that median HbA1c levels differed for T1D registry patients based on census tract poverty, race, and insurance, we used linear regression in a post-hoc analysis to evaluate associations between those exposures and median HbA1c. Age, gender, ethnicity, and duration of diabetes were not included in the models given that these variables did not differ by outcome variables. Diabetes technology—CGM and insulin pumps—was not included as a confounder because we hypothesized it mediated the effect of area-level poverty. For analyses, SigmaPlot (Version 14.0; SPSS Inc.) and R statistical software were used (The R Project for Statistical Computing).

A total of 439 patients were in the T1D registry and lived in Hamilton County as of December 31, 2017. Their demographic characteristics are listed in Table 1. Their median age was 14 years; 48% were female. The median poverty rate of the census tracts in which these youth lived was 11%. A total of 24.6% of patients identified as Black, 73.1% as White, and 2.3% as Hispanic; 35.8% were publicly insured. Patients had a median duration of T1D of 61 months (5 years, 1 month), with a median HbA1c (of all patients’ medians) of 8.7%. A total of 58.1% had a CGM, and 56% used an insulin pump.

Hospitalization Characteristics of Those in the T1D Registry

Approximately one-third of registry patients (n = 152) experienced ≥1 hospitalization for DKA during the study period (inclusive of new-onset and established diagnosis). Age, gender, ethnicity, and duration of T1D diagnosis were similar between those who experienced a hospitalization and those who did not. The median census tract poverty rate was higher among those who were hospitalized compared to those who were not (15% vs 8%, P < .01). Among hospitalized patients, 41% identified as Black; among all who were not hospitalized, 16% identified as Black (P < .01). Among all hospitalized patients, 56.6% were covered by public insurance, while 24.7% of non-hospitalized patients were covered by public insurance (P < .01). Hospitalized patients were more likely to have a higher median HbA1c (9.5% vs 8.4%, P < .01) and were significantly less likely to have a CGM or insulin pump (both P < .01). These associations were magnified among the roughly 10% of the T1D registry population (n = 42) who experienced ≥2 hospitalizations.

Figure, part A is a choropleth map depicting the distribution of poverty by census tract across Hamilton County. The map is complemented by part B of the Figure, which displays the T1D registry population split into poverty quartiles and each quartile’s DKA admission rate per T1D registry population. As census tract poverty increases, so too does the rate of DKA hospitalizations. There is an almost threefold difference between the hospitalization rate of the highest poverty quartile and that of the lowest.

Likelihood of Hospitalization

We examined the exposures of interest—census tract poverty, race, and insurance—and odds of hospitalization at a patient level using logistic regression models (Table 2). In the multivariable model that included these exposures, we found that for every 10% increase in the census tract poverty rate, the odds of being hospitalized for DKA increased by 22% (95% CI, 1.03-1.47). Race was not significantly associated with odds of DKA hospitalization in the multivariable model. Public insurance was associated with a 2.71-times higher odds of hospitalization compared to those with private insurance (95% CI, 1.62-4.55). Our sensitivity analysis of only those with established T1D diagnoses showed similar results to that of the entire T1D registry population.

Severity of DKA Presentation

A total of 152 registry patients were admitted; 89 patients had new-onset T1D, and 63 had an established T1D diagnosis. We examined presenting factors (initial pH, initial bicarbonate, HbA1c) and clinical characteristics (time to correction, inpatient bed-days, PICU bed-days) by census tract poverty, race, and insurance for all hospitalized patients. There were no significant differences in presenting factors or outcomes by census tract poverty or insurance. However, we did note certain differences by race. Indeed, though initial pH, bicarbonate, and time to correction did not meaningfully differ by race, Black patients had a longer length of stay than their White peers (Appendix Table 2; 3.06 vs 2.16 days, P < .01). We further examined this by diagnosis status, as newly diagnosed patients often stay longer for diabetes education.¹⁸ Black patients had a longer length of stay than White patients, regardless of whether they had new-onset (3.85 vs 2.99 days, P < .01) or established T1D (1.83 vs 0.97 days, P < .01). Additionally, we found that the HbA1c was significantly higher in Black versus White patients (12.4% vs 10.8%, P = .01). HbA1c was significantly different between established T1D patients identifying as Black and those identifying as White (11.7% vs 9.5%, P < .01); however, HbA1c did not differ by race for newly diagnosed patients (12.5% vs 13.0%, P = .63).

Median HbA1c of All T1D Registry Patients

Given the suggestion of HbA1c variability by exposure variables, we performed a post-hoc linear regression analysis. Table 3 displays the linear regression models assessing associations of census tract poverty, race, and insurance on median HbA1c for all T1D registry patients. In the multivariable model, census tract poverty was no longer associated with higher HbA1c levels (0.13, 95% CI, –0.01 to 0.26). However, Black patients continued to have significantly higher HbA1c levels than White patients (1.09% higher; 95% CI, 0.59-1.59). Also, those with public insurance had significantly higher HbA1c levels than those with private insurance (0.93% higher; 95% CI, 0.51-1.35).

Living in high poverty areas, identifying as Black, and having public insurance were each associated with a higher likelihood of hospitalization for DKA in an unadjusted model. When we examined these exposures together in a multivariable model, census tract poverty and insurance remained associated with a higher likelihood of hospitalization for DKA. Race—with a wide confidence interval—was no longer associated with a higher likelihood of hospitalization. The increased likelihood of hospitalization initially seen for Black patients may be because they are more likely to reside in high poverty areas and be on public insurance—inequities associated with structural racism.

These results mirror those of other studies that have found area-level poverty to be associated with higher rates of intensive care admissions and hospitalizations for conditions like asthma.^10,19 Our findings are also similar to those of a previous study that found that area-level deprivation was associated with a greater likelihood of DKA readmissions in pediatric and adult patients, as well as an adult study that showed increased neighborhood deprivation was associated with higher odds of DKA hospitalization.^11,20 Further examination of the mechanisms underlying DKA hospitalization is clearly warranted and would be informed by a deeper awareness of the social determinants of health—such as community context and the neighborhood/built environment²¹—and their links to racial inequities.

Neighborhood socioeconomic status, here measured as census tract poverty, is likely linked to hospitalization for DKA through contextual factors like food insecurity, limited access to supportive care (eg, presence of a school nurse who can co-manage T1D with the patient/family and medical team), and adverse environmental exposures.^22-24 To improve the health of patients, prevent hospitalizations, and achieve health equity, we suggest that both individual factors and these contextual factors need to be evaluated and addressed through community-connected interventions.^25,26 For example, interventions that deploy community health workers to navigate the complexities of health systems have been effective in reducing HbA1c levels in adults with type 2 diabetes mellitus²⁷ and are beginning to be studied in pediatrics.²⁸

Black patients had a higher median HbA1c compared to their White counterparts in the T1D registry. One study has previously shown that hemoglobin glycation biologically differed by around 0.5% in Black patients compared to White patients.²⁹ Despite this purported biological difference, a greater difference in HbA1c—nearly 2 points—was seen in the univariate model, illustrating that contextual factors such as socioeconomic status and treatment patterns, like the use of diabetes technology, exceed this biological difference.^30-32 Our multivariable results support the importance of these contextual factors, as the HbA1c difference decreased from 1.9 to 1.09 after census tract poverty and insurance were included.

Such differences by race also emerge from experiences of discrimination and bias. Thus, to improve outcomes and equity, we must understand and address racism and its many ill effects. Different levels of racism, such as structural or personally mediated racism, interact with social determinants of health in various ways.^33,34 For example, redlining—racially discriminatory grading of neighborhoods’ creditworthiness in the 1930s—is an example of structural racism that continues to affect the health of patients today.^35,36 Black families are more likely to live in areas with higher poverty, even after controlling for household income level.³⁷ The interaction between race and poverty is complex given these historical underpinnings of structural racism, as evidenced by race no longer being associated with an increased likelihood of DKA hospitalization in our multivariable analysis. The longer length of stay for Black patients despite similar presenting severity also exemplifies this complexity. For example, longer stays could result from explicit or implicit biases that lead providers to keep patients longer due to preconceived beliefs that the caregiver does not understand diabetes management.³⁸ However, it also could be that the caregiver takes longer to complete diabetes education due to systemic barriers, such as work-related obligations during business hours when diabetes educators are present, or transportation and childcare needs.³⁹ These barriers are often tied to socioeconomic status, making it more difficult for those in poverty to complete required education. We cannot conclude from our study the reasons for the difference in length of stay, just that differences exist. Moving forward, we suggest that design of interventions aimed at achieving health equity consider the complex interplay between health, healthcare systems, racism, and neighborhood context.

There were limitations to our study. First, given that this was an observational study, unmeasured confounders like patient or parental education and personal income could affect the strength of associations seen in our models. Second, although the CCHMC’s Diabetes Center treats nearly all children with T1D in Hamilton County, some patients, or hospitalizations at other centers, could have been missed. Third, the demographics of Hamilton County may not generalize to other diabetes centers. Fourth, although more granular data—such as census block groups or blocks—are available from the US Census Bureau, we chose census tracts as our unit of analysis to reduce sampling error that emerges from the use of such smaller geographies. Last, we measured census tract poverty rather than other contextual markers of one’s neighborhood, like vacant housing or the proportion of those receiving public benefits. However, these measures are often highly correlated with census tract poverty.⁴⁰

Type 1 diabetes mellitus (T1D) is a common chronic condition of childhood. Its incidence has risen steadily over the past two decades.¹ Treatment requires complex daily tasks, including blood glucose monitoring and insulin administration. The potential long-term complications of T1D—kidney failure, retinopathy, cardiovascular disease, and death—are grave but can be attenuated with effective glycemic management.² Still, less than 25% of youths aged 13 to 19 years achieve target hemoglobin A1c (HbA1c) levels.³

Diabetic ketoacidosis (DKA) is an acute, life-threatening complication of T1D associated with suboptimal glycemic management. In the United States, DKA hospitalizations increased during 2009-2014.⁴ One study found the average standardized cost of a DKA-related pediatric hospitalization exceeded $7000,⁵ an amount not including missed school days, parent/caregiver workdays, and social and family disruptions. Costs extend to psychological well-being, with patients reporting that they “keep thinking [DKA] may happen again.”⁶

The burden of T1D-related morbidity is not equitably distributed. Recent evidence suggests that racial/ethnic minorities and those with public insurance experience disproportionately high rates of DKA.^7,8 Neighborhood socioeconomic measures, like poverty, add an important dimension illustrative of ecological conditions in which patients reside. These measures have been studied in relation to hospitalization rates for chronic conditions like asthma and heart failure.^9,10 For example, one recent study found links between area-level poverty in the United States and the likelihood of readmissions for pediatric patients following an admission for DKA.¹¹ We sought to build on this finding by investigating whether area-level poverty, which we measured as the census tract poverty rate, patient race, and insurance status were associated with the likelihood of initial DKA hospitalization and severity of DKA presentation.

Design

We conducted a single-center, retrospective cohort study that examined data on youth with T1D extracted from the Cincinnati Children’s Hospital Medical Center (CCHMC) electronic medical record (EMR). CCHMC is an urban, tertiary-care, freestanding pediatric hospital, with near-complete market penetration in Hamilton County, Ohio, particularly for subspecialty care. The racial and ethnic breakdown of Hamilton County’s general population is 68% White, 27% Black/African American, 3% Asian, and 4% Hispanic or Latino (adds to more than 100% due to rounding).¹² The CCHMC Institutional Review Board approved this study.

Study Population

We identified eligible patients through a clinical registry within the EMR, which includes all T1D patients seen at CCHMC within the preceding 24 months. Patients are added to this registry upon initial T1D diagnosis by a research nurse who evaluates autoantibody status and clinical presentation. All patients ≤18 years old who had T1D, were active on the registry, and had an address within Hamilton County as of December 31, 2017, were included. Those with type 2 diabetes mellitus or secondary causes of diabetes were excluded. We captured patients on the registry as of December 31, 2017, and then examined whether they had been hospitalized from January 1, 2011, to December 31, 2017.

Covariates

We extracted demographic and clinical data for patients within the T1D registry as of December 31, 2017, and, for those hospitalized, at the time of their admission. Specifically, we extracted date of birth (to calculate age at encounter), sex, and ethnicity. Age was treated as a continuous variable. We defined ethnicity as Latinx (inclusive of Hispanic) or non-Latinx. We calculated duration of T1D diagnosis for patients within the registry by taking the difference of the diabetes onset date and the end of the study period. We extracted whether registry patients had an insulin pump or a continuous glucose monitor (CGM). Hospitalized patients were identified as new-onset or established T1D by comparing their hospitalization with the T1D diagnosis onset date.

For the entire T1D registry, we also assessed a key clinical characteristic: a patient’s median HbA1c. We enumerated the median HbA1c of all patients’ medians within the calendar year 2017, consistent with the methodology of a large international T1D registry.¹³ Among the subset of patients who were hospitalized, we obtained the HbA1c value during or prior to hospitalization.

Exposures

We captured exposure variables from EMR documentation of patients’ address, race, and insurance status. Given the small regional population of Latinx patients, we did not include ethnicity as an exposure. We used the address and insurance status at the beginning of the study period for established T1D patients, or at the time of diagnosis for new-onset patients, for analyses related to the likelihood of DKA admission. We used the address at the time of hospitalization for analyses of DKA presentation severity. We geocoded home addresses using ArcGIS software (Version 10.5.1; Esri), successfully geocoding >95% to the street level and assigning patients to a corresponding census tract—a census-defined geography consisting of approximately 4000 people.¹⁴ Census tracts, when drawn after the decennial (every 10 years) census, are designed to be sociodemographically homogenous¹⁵ and have long been used in medical and public health research.¹⁶ We connected geocoded addresses to the census tract poverty rate, defined as the percentage of individuals within a tract living below the federal poverty level. For our analyses, we primarily treated census tract poverty as a continuous variable ranging from 0% to 100%. We visualized the distribution of census tract poverty across Hamilton County using a choropleth map (Figure, part A). We also grouped the registry population into quartiles, ordered by census tract poverty levels, to ease the graphical display for Figure, part B. We defined race as White, Black (inclusive of African American), or other. We categorized insurance as Medicaid/public or Non-Medicaid/private.

Outcomes

The primary outcome was a hospitalization for DKA during the study period. We identified eligible events by admission diagnosis of DKA (International Classification of Disease, 10th Revision, Clinical Modification, E13.10 or E10.10). Secondarily, we assessed admission severity for each patient’s first DKA hospitalization using initial pH from a venous blood gas and initial bicarbonate from the basic metabolic panel. These measures were chosen in accordance with the American Diabetes Association’s published classification of DKA severity, which uses pH, bicarbonate, and mental status.¹⁷ We defined the time for correction as the minutes from admission time until the anion gap was ≤12 or the bicarbonate was ≥18. We calculated total inpatient bed-days and pediatric intensive care unit (PICU) bed-days, and identified whether a patient was readmitted for DKA during the study period.

Statistical Analyses

We used descriptive statistics to characterize demographic and clinical information for patients within the T1D registry and for the hospitalized subset. We examined DKA severity data for all admitted patients by diagnosis status (new-onset T1D vs established). Associations between categorical outcomes were assessed using the Chi-square or Fisher exact test, and continuous outcomes were evaluated using the Mann-Whitney Rank Sum test.

We individually examined the effects of census tract poverty, race, and insurance on the odds of hospitalization through logistic regression. After confirming overlap and variability in census tract poverty rates by race and admission status (Appendix Table 1), we included these exposures in a multivariable model to capture their independent effects. Because the impact of census tract poverty, race, and insurance on hospitalization odds could vary depending on T1D diagnosis status, we conducted a sensitivity analysis by repeating the multivariable model with only established T1D patients.

Given that median HbA1c levels differed for T1D registry patients based on census tract poverty, race, and insurance, we used linear regression in a post-hoc analysis to evaluate associations between those exposures and median HbA1c. Age, gender, ethnicity, and duration of diabetes were not included in the models given that these variables did not differ by outcome variables. Diabetes technology—CGM and insulin pumps—was not included as a confounder because we hypothesized it mediated the effect of area-level poverty. For analyses, SigmaPlot (Version 14.0; SPSS Inc.) and R statistical software were used (The R Project for Statistical Computing).

A total of 439 patients were in the T1D registry and lived in Hamilton County as of December 31, 2017. Their demographic characteristics are listed in Table 1. Their median age was 14 years; 48% were female. The median poverty rate of the census tracts in which these youth lived was 11%. A total of 24.6% of patients identified as Black, 73.1% as White, and 2.3% as Hispanic; 35.8% were publicly insured. Patients had a median duration of T1D of 61 months (5 years, 1 month), with a median HbA1c (of all patients’ medians) of 8.7%. A total of 58.1% had a CGM, and 56% used an insulin pump.

Hospitalization Characteristics of Those in the T1D Registry

Approximately one-third of registry patients (n = 152) experienced ≥1 hospitalization for DKA during the study period (inclusive of new-onset and established diagnosis). Age, gender, ethnicity, and duration of T1D diagnosis were similar between those who experienced a hospitalization and those who did not. The median census tract poverty rate was higher among those who were hospitalized compared to those who were not (15% vs 8%, P < .01). Among hospitalized patients, 41% identified as Black; among all who were not hospitalized, 16% identified as Black (P < .01). Among all hospitalized patients, 56.6% were covered by public insurance, while 24.7% of non-hospitalized patients were covered by public insurance (P < .01). Hospitalized patients were more likely to have a higher median HbA1c (9.5% vs 8.4%, P < .01) and were significantly less likely to have a CGM or insulin pump (both P < .01). These associations were magnified among the roughly 10% of the T1D registry population (n = 42) who experienced ≥2 hospitalizations.

Figure, part A is a choropleth map depicting the distribution of poverty by census tract across Hamilton County. The map is complemented by part B of the Figure, which displays the T1D registry population split into poverty quartiles and each quartile’s DKA admission rate per T1D registry population. As census tract poverty increases, so too does the rate of DKA hospitalizations. There is an almost threefold difference between the hospitalization rate of the highest poverty quartile and that of the lowest.

Likelihood of Hospitalization

We examined the exposures of interest—census tract poverty, race, and insurance—and odds of hospitalization at a patient level using logistic regression models (Table 2). In the multivariable model that included these exposures, we found that for every 10% increase in the census tract poverty rate, the odds of being hospitalized for DKA increased by 22% (95% CI, 1.03-1.47). Race was not significantly associated with odds of DKA hospitalization in the multivariable model. Public insurance was associated with a 2.71-times higher odds of hospitalization compared to those with private insurance (95% CI, 1.62-4.55). Our sensitivity analysis of only those with established T1D diagnoses showed similar results to that of the entire T1D registry population.

Severity of DKA Presentation

A total of 152 registry patients were admitted; 89 patients had new-onset T1D, and 63 had an established T1D diagnosis. We examined presenting factors (initial pH, initial bicarbonate, HbA1c) and clinical characteristics (time to correction, inpatient bed-days, PICU bed-days) by census tract poverty, race, and insurance for all hospitalized patients. There were no significant differences in presenting factors or outcomes by census tract poverty or insurance. However, we did note certain differences by race. Indeed, though initial pH, bicarbonate, and time to correction did not meaningfully differ by race, Black patients had a longer length of stay than their White peers (Appendix Table 2; 3.06 vs 2.16 days, P < .01). We further examined this by diagnosis status, as newly diagnosed patients often stay longer for diabetes education.¹⁸ Black patients had a longer length of stay than White patients, regardless of whether they had new-onset (3.85 vs 2.99 days, P < .01) or established T1D (1.83 vs 0.97 days, P < .01). Additionally, we found that the HbA1c was significantly higher in Black versus White patients (12.4% vs 10.8%, P = .01). HbA1c was significantly different between established T1D patients identifying as Black and those identifying as White (11.7% vs 9.5%, P < .01); however, HbA1c did not differ by race for newly diagnosed patients (12.5% vs 13.0%, P = .63).

Median HbA1c of All T1D Registry Patients

Given the suggestion of HbA1c variability by exposure variables, we performed a post-hoc linear regression analysis. Table 3 displays the linear regression models assessing associations of census tract poverty, race, and insurance on median HbA1c for all T1D registry patients. In the multivariable model, census tract poverty was no longer associated with higher HbA1c levels (0.13, 95% CI, –0.01 to 0.26). However, Black patients continued to have significantly higher HbA1c levels than White patients (1.09% higher; 95% CI, 0.59-1.59). Also, those with public insurance had significantly higher HbA1c levels than those with private insurance (0.93% higher; 95% CI, 0.51-1.35).

Living in high poverty areas, identifying as Black, and having public insurance were each associated with a higher likelihood of hospitalization for DKA in an unadjusted model. When we examined these exposures together in a multivariable model, census tract poverty and insurance remained associated with a higher likelihood of hospitalization for DKA. Race—with a wide confidence interval—was no longer associated with a higher likelihood of hospitalization. The increased likelihood of hospitalization initially seen for Black patients may be because they are more likely to reside in high poverty areas and be on public insurance—inequities associated with structural racism.

These results mirror those of other studies that have found area-level poverty to be associated with higher rates of intensive care admissions and hospitalizations for conditions like asthma.^10,19 Our findings are also similar to those of a previous study that found that area-level deprivation was associated with a greater likelihood of DKA readmissions in pediatric and adult patients, as well as an adult study that showed increased neighborhood deprivation was associated with higher odds of DKA hospitalization.^11,20 Further examination of the mechanisms underlying DKA hospitalization is clearly warranted and would be informed by a deeper awareness of the social determinants of health—such as community context and the neighborhood/built environment²¹—and their links to racial inequities.

Neighborhood socioeconomic status, here measured as census tract poverty, is likely linked to hospitalization for DKA through contextual factors like food insecurity, limited access to supportive care (eg, presence of a school nurse who can co-manage T1D with the patient/family and medical team), and adverse environmental exposures.^22-24 To improve the health of patients, prevent hospitalizations, and achieve health equity, we suggest that both individual factors and these contextual factors need to be evaluated and addressed through community-connected interventions.^25,26 For example, interventions that deploy community health workers to navigate the complexities of health systems have been effective in reducing HbA1c levels in adults with type 2 diabetes mellitus²⁷ and are beginning to be studied in pediatrics.²⁸

Black patients had a higher median HbA1c compared to their White counterparts in the T1D registry. One study has previously shown that hemoglobin glycation biologically differed by around 0.5% in Black patients compared to White patients.²⁹ Despite this purported biological difference, a greater difference in HbA1c—nearly 2 points—was seen in the univariate model, illustrating that contextual factors such as socioeconomic status and treatment patterns, like the use of diabetes technology, exceed this biological difference.^30-32 Our multivariable results support the importance of these contextual factors, as the HbA1c difference decreased from 1.9 to 1.09 after census tract poverty and insurance were included.

Such differences by race also emerge from experiences of discrimination and bias. Thus, to improve outcomes and equity, we must understand and address racism and its many ill effects. Different levels of racism, such as structural or personally mediated racism, interact with social determinants of health in various ways.^33,34 For example, redlining—racially discriminatory grading of neighborhoods’ creditworthiness in the 1930s—is an example of structural racism that continues to affect the health of patients today.^35,36 Black families are more likely to live in areas with higher poverty, even after controlling for household income level.³⁷ The interaction between race and poverty is complex given these historical underpinnings of structural racism, as evidenced by race no longer being associated with an increased likelihood of DKA hospitalization in our multivariable analysis. The longer length of stay for Black patients despite similar presenting severity also exemplifies this complexity. For example, longer stays could result from explicit or implicit biases that lead providers to keep patients longer due to preconceived beliefs that the caregiver does not understand diabetes management.³⁸ However, it also could be that the caregiver takes longer to complete diabetes education due to systemic barriers, such as work-related obligations during business hours when diabetes educators are present, or transportation and childcare needs.³⁹ These barriers are often tied to socioeconomic status, making it more difficult for those in poverty to complete required education. We cannot conclude from our study the reasons for the difference in length of stay, just that differences exist. Moving forward, we suggest that design of interventions aimed at achieving health equity consider the complex interplay between health, healthcare systems, racism, and neighborhood context.

There were limitations to our study. First, given that this was an observational study, unmeasured confounders like patient or parental education and personal income could affect the strength of associations seen in our models. Second, although the CCHMC’s Diabetes Center treats nearly all children with T1D in Hamilton County, some patients, or hospitalizations at other centers, could have been missed. Third, the demographics of Hamilton County may not generalize to other diabetes centers. Fourth, although more granular data—such as census block groups or blocks—are available from the US Census Bureau, we chose census tracts as our unit of analysis to reduce sampling error that emerges from the use of such smaller geographies. Last, we measured census tract poverty rather than other contextual markers of one’s neighborhood, like vacant housing or the proportion of those receiving public benefits. However, these measures are often highly correlated with census tract poverty.⁴⁰

For many patients hospitalized with COVID-19, the impact of the illness continues well beyond hospital discharge.¹ Heavy burdens of persistent symptoms have been reported, albeit often from regional and single-hospital samples.^2-7 Critically, not all initial reports capture information on pre-COVID-19 symptom burden, so it is unclear whether these highly prevalent problems are truly new; an alternative explanation might be that patients already with symptoms were more likely to be infected with or seek care for SARS-CoV-2.⁸

Fewer data are available about patients’ abilities to go about the activities of their lives, nor is as much known about the relationships between new symptoms and other impacts. Most of the available information is from health systems during the initial surge of COVID-19 in early 2020—when testing for SARS-CoV-2 was limited even in the inpatient setting; when hospitals’ postdischarge care systems may have been heavily disrupted; and when clinicians were often reasonably focused primarily on reducing mortality in their first cases of COVID-19 rather than promoting recovery from an often-survivable illness. Increasing evidence shows that the inpatient case-fatality rate of COVID-19 is improving over time^9,10; this makes unclear the generalizability of outcomes data from early COVID-19 patients to more recent patients.¹¹

Therefore, we report multicenter measurements of incident levels of persistent cardiopulmonary symptoms, disability, return to baseline, and impact on employment among a recent cohort of COVID-19 patients hospitalized around the United States during the “third wave” of COVID-19—fall and winter 2020-2021. We focus on the 1-month time point after hospital discharge, as this time point is still in the early vulnerable period during which hospital transition-of-care programs are understood to have responsibility.

The first 253 patients who completed 1-month postdischarge telephone follow-up surveys from the ongoing nationwide BLUE CORAL study were included. BLUE CORAL will enroll up to 1,500 hospitalized COVID-19 patients at 36 US centers (the identities of which are reported in Appendix 1) as a part of the National Heart, Lung, and Blood Institute’s Prevention and Early Treatment of Acute Lung Injury (PETAL) Network. We report here on survey questions that allowed for a clear comparison to be made between 1-month follow-up responses and pre-COVID baseline variables; these comparisons were based on (1) previous in-hospital assessment; (2) explicitly asking patients to compare to pre-COVID-19 levels; or (3) explicitly asking patients for changes in relation to their COVID-19 hospitalization. Items were chosen for inclusion in this report without looking at their association with other variables.

This research was approved by the Vanderbilt Institutional Review Board (IRB), serving as central IRB for the PETAL Network; patients or their surrogates provided informed consent.

Participants

Patients with COVID-19 were identified during hospitalization and within 14 days of a positive molecular test for SARS-CoV-2. Eligible patients presented with fever and/or respiratory signs/symptoms, such as hypoxemia, shortness of breath, or infiltrates on chest imaging. Patients were enrolled within the first 72 hours of hospitalization (in order to avoid oversampling patients with relatively longer stays, and to study the biology of early COVID-19), and excluded if they had comfort-care orders (because of their limited likelihood of surviving to follow-up), or were incarcerated (because of difficulties in obtaining truly open informed consent and likely difficulties in follow-up). Pertinently, patients were not required to be in the intensive care unit.

Surviving patients who spoke English or Spanish, were not homeless on hospital admission, and were neither significantly disabled nor significantly cognitively impaired were eligible for follow-up. “Not significantly disabled” was defined as having limitations due to health on no more than three activities of daily living before their COVID-19 hospitalization, as assessed at BLUE CORAL enrollment; this was chosen because of the potentially limited sensitivity of many of our questionnaires to detect an impact of COVID-19 in patients with greater than this level of disability. We included patients who were able to consent for themselves in the study, or for whom the legally appointed representative consenting on their behalf in the hospital reported no evidence of cognitive impairment, defined as no more than four of the problems on the eight-item Alzheimer’s Dementia (AD8) scale.^12-14

Data Collection

One-month surveys were administered to patients or, when necessary, their proxies; the complete English- and Spanish-language instruments are presented in Appendix 2. Study data were collected and managed using REDCap electronic data capture tools hosted at the University of Michigan.^15,16

Patients were contacted via phone by trained interviewers beginning 21 days after hospital discharge; interviews were completed a median of 47 days after discharge (interquartile range [IQR], 26-61). Efforts prioritized former patients completing surveys themselves by phone, but a well-informed proxy was approached if needed. Proxies, who included spouses, adult children, or other relatives, family friends, or primary caregivers, were in regular contact with the patient and understood the patient’s health status. If necessary, the survey could be completed over multiple phone calls, and a written, mail-back option was available. Other best practices in accurate survey data collection and cohort retention were used.^17-19 Participants were given a $10 gift card.

New cardiopulmonary symptoms were queried with symptom-targeted questions informed by the Airways Questionnaire 20,²⁰ the Kansas City Cardiomyopathy Questionnaire,^21,22 and the Seattle Angina Questionnaire.²³ Whenever a respondent reported a given symptom, they were asked, “Compared to 1 month before your COVID-19 hospitalization, is this better, worse, or about the same?” We counted the number of symptoms which the patient reported as worse.

Using wording from the Health and Retirement Study,²⁴ disability was assessed based on a self-report of any of 14 health-related limitations in activities of daily living or instrumental activities of daily living, as in past studies²⁵: dressing, walking across a room, bathing, eating, getting out of bed, using a toilet, using a map, preparing a hot meal, shopping for groceries, making a phone call, taking medications, paying bills, carrying 10 lb (eg, a heavy bag of groceries), and, as a combined single item, stooping, kneeling, or crouching. Well-chosen proxy reports appear reliable for these items.²⁶ We counted the number of activities for which the patient reported a limitation, comparing those reported at 1 month to those reported during the in-hospital survey assessing pre-illness functioning.

The financial consequences of the COVID-19 hospitalization were assessed in two ways. First, we used a modified version of a World Health Organization Disability Assessment Schedule (WHODAS) 2.0 question²⁷: “Since your COVID-19 hospitalization, how much has your health been a drain on the financial resources of you or your family?” Second, we used the financial toxicity items developed with the Mi-COVID19 study³ based on extensive qualitative interviews with respiratory failure survivors²⁸; these questions were anchored explicitly on “the financial cost of dealing with your COVID-19 hospitalization and related care.”

Data Analysis

There were few missing data, and almost all were on outcome variables. Where present, the degree of missingness is reported and casewise deletion used. Because this was a planned early look at responses to an ongoing survey, with analysis based on the number of accrued responses, the ultimate denominator for response rate calculation is unknown. Therefore, two bounds are presented—the minimum, on the assumption that all remaining uncompleted surveys will be missed; and the maximum, as if the uncompleted surveys were not yet in the eligible denominator.

Variables were summarized with medians and IQRs. Multilevel logistic regression was used to test for differences across demographic characteristics in the rates of development of any new symptom or disability; site-level differences were modeled using a random effect. Gender, race/ethnicity, and age were included in all regressions unless noted otherwise; age was included with both linear and quadratic terms when used as a control variable. For the degree of return to baseline and for the number of new limitations in activities of daily living, we explored associations as dichotomized variables (any/none, using multilevel logistic regression) and as continuous variables (using multilevel linear regression). Percent of variance explained was calculated using the R² in unadjusted linear regression, and Spearman rank correlations were used to allow nonlinearities in comparisons across outcomes. All adjusted models are presented in Appendix Table 1. Analyses were conducted in Stata 16.1 (StataCorp, 2020); analytic code is presented in Appendix 3, and a log file of all analyses is in Appendix 4.

The 250th 1-month follow-up was completed on February 26, 2021. One month prior, 647 patients had been recruited at 26 centers in the inpatient phase of the study. Patient demographics for the 253 patients surveyed through that date are shown in Appendix Table 2. On the day of the early look at the data, 460 patients had become eligible for 1-month follow-up and 64 patients had been missed for 1-month follow-up (maximum response rate of 79.8%, minimum possible final response rate of 55.0%) (Figure 1). Seven surveys were completed by proxies. Respondents’ median age was 60 years (IQR, 45-68), and 111 (43.4%) were female. Their median hospital length of stay was 5 days(IQR, 3-8) . A total of 236 (93.3%) patients were discharged home, including 197 (77.9%) without home care services and 39 (15.4%) with home care services.

One hundred and thirty-nine patients (56.5%; 95% CI, 50.1%-62.8%) reported at least one new or worsened cardiopulmonary symptom after their COVID-19 hospitalization (Table; seven patients did not respond to these questions). Most patients with new symptoms had one (48 [19.5%]; 95% CI, 14.8%-25.0%) or two (32 [13%]; 95% CI, 9.7%-17.7%) of the new symptoms queried. The most common new cardiopulmonary symptom was cough, reported by 57 (23.2%; 95% CI, 18.0%-29.0%) patients. New oxygen use was reported by 28 (11.4%; 95% CI, 7.7%-16.0%) patients, with another 11 (4.5%; 95% CI, 2.3%-7.9%) reporting increased oxygen requirements. Women were twice as likely as men to report any new cardiopulmonary symptom (adjusted odds ratio [aOR], 2.24; 95% CI, 1.29-3.90) and non-Hispanic Black and Hispanic patients were less likely than White patients to report new symptoms (aOR, 0.31; 95% CI, 0.12-0.83; and aOR, 0.38; 95% CI, 0.21-0.71, respectively). Longer lengths of hospital stay were associated with greater 1-month cardiopulmonary symptoms (aOR, 1.82 per additional week in the hospital; 95% CI, 1.11-2.98), but discharge destination was not (aOR, 0.92; 95% CI, 0.39-1.71).

New limitations in activities of daily living or instrumental activities of daily living were present in 130 (52.8%; 95% CI, 46.4%-59.5%) patients (seven not responding), all of whom had 0 to 3 limitations before their COVID-19 hospitalization. Indeed, 62 (25.2%; 95% CI, 19.9%-31.1%) reported 3 or more new health-related limitations in activities of daily living or instrumental activities of daily living compared to their pre-COVID-19 baseline, as assessed separately during their hospitalization (Figure 2; rates of limitations in individual activities are shown in Appendix Table 3). Older patients were more likely to report a new health-related limitation, and Hispanic patients were less likely to have a new limitation. New limitations were common among patients discharged home without home health services. The number of new cardiopulmonary symptoms explained 11.2% of the variance in the number of new limitations in activities of daily living, a Spearman rank correlation of 0.30 (P < .0001; see Appendix Table 4). More than three in four COVID-19 patients reported new or worsened cardiopulmonary symptoms or new health-related limitations in activities of daily living at 1 month—only 62 (24.5%) patients reported neither.

At 1 month after hospital discharge, 213 (84.2%) patients reported that they were not fully back to their pre-COVID-19 level of functioning (3 declined to answer the question). When asked, “On a scale of 1 to 100, with 100 being all the way back to what you could do before COVID, how close to being back are you?” the median response was 80, with an IQR of 64-95 (Figure 3). Forty-two (16.8%; 95% CI, 12.4%-22.0%) patients reported a level of 50 or below. Women and older patients reported lower levels of return of functioning, as did those with longer hospital stays and new or worsened cardiopulmonary symptoms. Each additional week in hospital length of stay was associated with a 7.5-point lower response to the question (95% CI, –11.2 to –3.8), but discharge destination was not associated with the answer after adjusting for demographics. Patients with and without new limitations in activities of daily living and with and without new cardiopulmonary symptoms were found across the range of self-reported degree of recovery, although patients without a new problem in one of those domains were rarer among those reporting recovery of less than 70. The number of new cardiopulmonary symptoms explained 19.7% of the variance in the response to this question, a Spearman rank correlation of 0.47 (P < .0001).

More than half of respondents (115 [55.0%]; 95% CI, 48.0%-61.9%; 44 not responding) stated that their COVID-19 hospitalization had been a drain on the finances of their family; 53 (25.4%; 95% CI, 19.6%-31.8%; 44 not responding) rated that drain as moderate, severe, or extreme within the first month after hospital discharge. Forty-nine patients (19.8%; 95% CI, 15.1%-25.4%; 6 not responding) reported that they had to change their work because of their COVID-19 hospitalization, and 93 patients (37.8%; 95% CI, 31.7%-44.2%; 7 not responding) reported that a loved one had taken time off work to care for them. Altogether, one in five COVID-19 patients reported that, within the first month after hospital discharge, they used all or most of their savings because of their COVID-19 illness or hospitalization (58 [23.2%]; 95% CI, 18.1%-29.9%; 3 not responding). There were no demographic differences in the likelihood of losing a job or having a loved one take time off for caregiving, but non-Hispanic Black and Hispanic patients were much more likely to report having used all or most of their savings (aOR, 2.96; 95% CI, 1.09-8.04; and aOR, 2.68; 95% CI, 1.35-5.31, respectively) than White patients. Hospital length of stay and discharge destination were not consistently associated with these financial toxicities. The development of new or worsened cardiopulmonary symptoms was not associated with job change or having a caregiver take time off but was associated with increased likelihood of having used all or most savings (aOR, 2.30; 95% CI, 1.12-4.37).

In a geographically and demographically diverse national US cohort, we found that a decline in perceived health, new or worsened cardiopulmonary symptoms, new limitations in activities of daily living, and new financial stresses were common among patients hospitalized during the US third wave of COVID-19 at 1 month after hospital discharge. The new cardiopulmonary symptoms were significantly associated with the self-report of incomplete recovery and financial stress, but less closely associated with incident disability, inability to work, and caregiving receipt. There were not consistent differences between any demographic groups on these outcomes. Patients with longer lengths of stay generally reported more problems. New problems were very common among patients discharged directly home without home health services.

These data suggest a broad range of new problems among survivors of COVID-19 hospitalization. Moreover, these problems are not well-correlated with each other. This raises the possibility that there may be multiple phenotypes of post-acute sequelae after COVID-19 hospitalization. It is not clear to what extent these differences are mediated by differences in tissue damage from or immunologic response to SARS-CoV-2, distinct from or interacting with other elements of treatment, hospitalization, or the illness experience. The degree of financial stress, savings loss, and job dislocation reported here suggests these patients will face substantial challenges in guiding their own recovery in the absence of a dedicated set of services.^28,29The persistent symptoms faced by these COVID-19 patients can be considered in the context of post-acute sequelae among survivors of community-acquired pneumonia in previous studies, as summarized in a recent systematic review.³⁰ For example, only 35% of a large cohort of adults with community-acquired pneumonia who were evaluated in the emergency department were completely free of pneumonia-related symptoms 6 weeks after antibiotic therapy.^31,32 Limitations in activities of daily living have been reported at 1 month after community-acquired pneumonia³³; rehospitalization and early post-discharge mortality rates may also be similar.^34,35 These findings suggest that the persistent problems of both COVID-19 and other pneumonia patients may highlight important opportunities for improvements in healthcare systems,³⁶ and that burdensome postacute sequelae of COVID-19 may not be attributable solely to distinctive features of the SARS-CoV-2 virus.

A majority of patients discharged home without home health services reported new difficulties in their activities of daily living; 77% of patients with new disability at 1 month had been discharged without home services. These data, however, do not show to what extent this lack of home health services resulted from lack of referral for services, home health provider unavailability, or patient refusal of recommended services. Nonetheless, this nonreceipt of home health services may have been consequential. Among hospitalized patients recovering from pneumonia pre-COVID, the use of post-hospital physical and occupational therapy was associated with reduced risk of readmissions and death.³⁷ This association was greater among patients with lower baseline mobility scores and in patients discharged to home directly. Further, the risk of poor outcomes decreased in a dose-response fashion with increased post-hospital therapy delivery. Failure to provide services for postdischarge disability was previously identified as a potential vulnerability of patients during COVID-19.³⁸

This study adds to the literature. The focus on sequelae perceived by the patient to be incident, as distinguished from symptoms and disability existing before COVID-19, increases the likelihood that these data reflect the influence of the COVID-19 hospitalization. These data emphasize that, despite relatively brief hospitalizations, diverse problems are quite common and consequential for patients’ ability to return to their pre-COVID-19 roles. They further add to the literature by demonstrating the relatively loose coupling between various ways in which postacute sequelae of COVID-19 might be defined: the cardiopulmonary symptoms examined here, the patient’s reported completeness of recovery, the financial stresses the hospitalization placed on the patient and their family, or the development of new limitations in activities of daily living.

Our findings highlight a potential second public health crisis from COVID, related to post-COVID recovery, resulting from the incident disability and economic loss among COVID survivors. While much attention is paid to deaths from COVID, there is less (albeit growing) recognition of the long-term consequences in survivors of COVID-19.³⁹ The downstream economic impacts from job loss and financial insolvency for COVID-19 survivors have ramifications for caregivers, family units that include dependents, and the broader US economy—and may do so for generations if uncorrected, as has been suggested after the 1918 influenza pandemic.⁴⁰ These data may, indeed, look worse at later follow-up given the delay in hospital billing and new expenses in the wake of illness and hospitalization.^28,36,41 It is important that the healthcare system and policymakers consider early investments in post-hospital rehabilitation and adaptive services to allow workers to return to the workforce as soon as possible, and prepare for an increased need for financial support for recovering COVID patients.⁴²

Importantly, these data cannot distinguish between the impact of SARS-CoV-2 infection itself from the treatment received for COVID-19 or other non-COVID-19-specific aspects of hospital care. COVID-19 inpatient case fatality rates and management have changed over time, and so generalizability to future cohorts is unknown.^9-11 This cohort was recruited in the inpatient setting at largely teaching hospitals; therefore, these patients’ experience may be not be representative of all hospitalized COVID-19 patients during this time period. The generalizability of hospital-based studies to patients not hospitalized for COVID-19 remains a subject of active inquiry. We only interviewed patients who were not homeless (excluding 7 of 588 eligible, 1.2%) and who spoke English or Spanish (excluding 4 of 588 eligible, 0.7%); these and other inclusion/exclusion criteria should be considered when evaluating the generalizability of these findings to other patients. We did not prospectively collect measures of fatigue to examine this important and complex symptom, nor did we evaluate outpatient therapy. Finally, self-report was used, rather than using objective measurements of what the patient did or did not do in their home environment. This is consistent with clinical practice that emphasizes patients as primary reporters of their present state, but may introduce measurement error compared to more invasive strategies if those are considered the gold standard.

Patients who survived hospitalization from COVID-19 during the period of August 2020 to January 2021 continued to face significant burdens of new cardiopulmonary symptoms, incomplete recovery, disability, and financial toxicity, all of which extend to patients discharged directly home without services. The correlations between these potential symptoms are no more than partial, and an exclusive focus on one area may neglect other areas of patient need.

The authors thank the patients and families of the Biology and Longitudinal Epidemiology: COVID-19 Observational (BLUE CORAL) study for their generous sharing of their time with us. We acknowledge Hallie C Prescott (University of Michigan and VA Ann Arbor) for her assistance in developing the financial toxicity questions.

For many patients hospitalized with COVID-19, the impact of the illness continues well beyond hospital discharge.¹ Heavy burdens of persistent symptoms have been reported, albeit often from regional and single-hospital samples.^2-7 Critically, not all initial reports capture information on pre-COVID-19 symptom burden, so it is unclear whether these highly prevalent problems are truly new; an alternative explanation might be that patients already with symptoms were more likely to be infected with or seek care for SARS-CoV-2.⁸

Fewer data are available about patients’ abilities to go about the activities of their lives, nor is as much known about the relationships between new symptoms and other impacts. Most of the available information is from health systems during the initial surge of COVID-19 in early 2020—when testing for SARS-CoV-2 was limited even in the inpatient setting; when hospitals’ postdischarge care systems may have been heavily disrupted; and when clinicians were often reasonably focused primarily on reducing mortality in their first cases of COVID-19 rather than promoting recovery from an often-survivable illness. Increasing evidence shows that the inpatient case-fatality rate of COVID-19 is improving over time^9,10; this makes unclear the generalizability of outcomes data from early COVID-19 patients to more recent patients.¹¹

Therefore, we report multicenter measurements of incident levels of persistent cardiopulmonary symptoms, disability, return to baseline, and impact on employment among a recent cohort of COVID-19 patients hospitalized around the United States during the “third wave” of COVID-19—fall and winter 2020-2021. We focus on the 1-month time point after hospital discharge, as this time point is still in the early vulnerable period during which hospital transition-of-care programs are understood to have responsibility.

The first 253 patients who completed 1-month postdischarge telephone follow-up surveys from the ongoing nationwide BLUE CORAL study were included. BLUE CORAL will enroll up to 1,500 hospitalized COVID-19 patients at 36 US centers (the identities of which are reported in Appendix 1) as a part of the National Heart, Lung, and Blood Institute’s Prevention and Early Treatment of Acute Lung Injury (PETAL) Network. We report here on survey questions that allowed for a clear comparison to be made between 1-month follow-up responses and pre-COVID baseline variables; these comparisons were based on (1) previous in-hospital assessment; (2) explicitly asking patients to compare to pre-COVID-19 levels; or (3) explicitly asking patients for changes in relation to their COVID-19 hospitalization. Items were chosen for inclusion in this report without looking at their association with other variables.

This research was approved by the Vanderbilt Institutional Review Board (IRB), serving as central IRB for the PETAL Network; patients or their surrogates provided informed consent.

Participants

Patients with COVID-19 were identified during hospitalization and within 14 days of a positive molecular test for SARS-CoV-2. Eligible patients presented with fever and/or respiratory signs/symptoms, such as hypoxemia, shortness of breath, or infiltrates on chest imaging. Patients were enrolled within the first 72 hours of hospitalization (in order to avoid oversampling patients with relatively longer stays, and to study the biology of early COVID-19), and excluded if they had comfort-care orders (because of their limited likelihood of surviving to follow-up), or were incarcerated (because of difficulties in obtaining truly open informed consent and likely difficulties in follow-up). Pertinently, patients were not required to be in the intensive care unit.

Surviving patients who spoke English or Spanish, were not homeless on hospital admission, and were neither significantly disabled nor significantly cognitively impaired were eligible for follow-up. “Not significantly disabled” was defined as having limitations due to health on no more than three activities of daily living before their COVID-19 hospitalization, as assessed at BLUE CORAL enrollment; this was chosen because of the potentially limited sensitivity of many of our questionnaires to detect an impact of COVID-19 in patients with greater than this level of disability. We included patients who were able to consent for themselves in the study, or for whom the legally appointed representative consenting on their behalf in the hospital reported no evidence of cognitive impairment, defined as no more than four of the problems on the eight-item Alzheimer’s Dementia (AD8) scale.^12-14

Data Collection

One-month surveys were administered to patients or, when necessary, their proxies; the complete English- and Spanish-language instruments are presented in Appendix 2. Study data were collected and managed using REDCap electronic data capture tools hosted at the University of Michigan.^15,16

Patients were contacted via phone by trained interviewers beginning 21 days after hospital discharge; interviews were completed a median of 47 days after discharge (interquartile range [IQR], 26-61). Efforts prioritized former patients completing surveys themselves by phone, but a well-informed proxy was approached if needed. Proxies, who included spouses, adult children, or other relatives, family friends, or primary caregivers, were in regular contact with the patient and understood the patient’s health status. If necessary, the survey could be completed over multiple phone calls, and a written, mail-back option was available. Other best practices in accurate survey data collection and cohort retention were used.^17-19 Participants were given a $10 gift card.

New cardiopulmonary symptoms were queried with symptom-targeted questions informed by the Airways Questionnaire 20,²⁰ the Kansas City Cardiomyopathy Questionnaire,^21,22 and the Seattle Angina Questionnaire.²³ Whenever a respondent reported a given symptom, they were asked, “Compared to 1 month before your COVID-19 hospitalization, is this better, worse, or about the same?” We counted the number of symptoms which the patient reported as worse.

Using wording from the Health and Retirement Study,²⁴ disability was assessed based on a self-report of any of 14 health-related limitations in activities of daily living or instrumental activities of daily living, as in past studies²⁵: dressing, walking across a room, bathing, eating, getting out of bed, using a toilet, using a map, preparing a hot meal, shopping for groceries, making a phone call, taking medications, paying bills, carrying 10 lb (eg, a heavy bag of groceries), and, as a combined single item, stooping, kneeling, or crouching. Well-chosen proxy reports appear reliable for these items.²⁶ We counted the number of activities for which the patient reported a limitation, comparing those reported at 1 month to those reported during the in-hospital survey assessing pre-illness functioning.

The financial consequences of the COVID-19 hospitalization were assessed in two ways. First, we used a modified version of a World Health Organization Disability Assessment Schedule (WHODAS) 2.0 question²⁷: “Since your COVID-19 hospitalization, how much has your health been a drain on the financial resources of you or your family?” Second, we used the financial toxicity items developed with the Mi-COVID19 study³ based on extensive qualitative interviews with respiratory failure survivors²⁸; these questions were anchored explicitly on “the financial cost of dealing with your COVID-19 hospitalization and related care.”

Data Analysis

There were few missing data, and almost all were on outcome variables. Where present, the degree of missingness is reported and casewise deletion used. Because this was a planned early look at responses to an ongoing survey, with analysis based on the number of accrued responses, the ultimate denominator for response rate calculation is unknown. Therefore, two bounds are presented—the minimum, on the assumption that all remaining uncompleted surveys will be missed; and the maximum, as if the uncompleted surveys were not yet in the eligible denominator.

Variables were summarized with medians and IQRs. Multilevel logistic regression was used to test for differences across demographic characteristics in the rates of development of any new symptom or disability; site-level differences were modeled using a random effect. Gender, race/ethnicity, and age were included in all regressions unless noted otherwise; age was included with both linear and quadratic terms when used as a control variable. For the degree of return to baseline and for the number of new limitations in activities of daily living, we explored associations as dichotomized variables (any/none, using multilevel logistic regression) and as continuous variables (using multilevel linear regression). Percent of variance explained was calculated using the R² in unadjusted linear regression, and Spearman rank correlations were used to allow nonlinearities in comparisons across outcomes. All adjusted models are presented in Appendix Table 1. Analyses were conducted in Stata 16.1 (StataCorp, 2020); analytic code is presented in Appendix 3, and a log file of all analyses is in Appendix 4.

The 250th 1-month follow-up was completed on February 26, 2021. One month prior, 647 patients had been recruited at 26 centers in the inpatient phase of the study. Patient demographics for the 253 patients surveyed through that date are shown in Appendix Table 2. On the day of the early look at the data, 460 patients had become eligible for 1-month follow-up and 64 patients had been missed for 1-month follow-up (maximum response rate of 79.8%, minimum possible final response rate of 55.0%) (Figure 1). Seven surveys were completed by proxies. Respondents’ median age was 60 years (IQR, 45-68), and 111 (43.4%) were female. Their median hospital length of stay was 5 days(IQR, 3-8) . A total of 236 (93.3%) patients were discharged home, including 197 (77.9%) without home care services and 39 (15.4%) with home care services.

One hundred and thirty-nine patients (56.5%; 95% CI, 50.1%-62.8%) reported at least one new or worsened cardiopulmonary symptom after their COVID-19 hospitalization (Table; seven patients did not respond to these questions). Most patients with new symptoms had one (48 [19.5%]; 95% CI, 14.8%-25.0%) or two (32 [13%]; 95% CI, 9.7%-17.7%) of the new symptoms queried. The most common new cardiopulmonary symptom was cough, reported by 57 (23.2%; 95% CI, 18.0%-29.0%) patients. New oxygen use was reported by 28 (11.4%; 95% CI, 7.7%-16.0%) patients, with another 11 (4.5%; 95% CI, 2.3%-7.9%) reporting increased oxygen requirements. Women were twice as likely as men to report any new cardiopulmonary symptom (adjusted odds ratio [aOR], 2.24; 95% CI, 1.29-3.90) and non-Hispanic Black and Hispanic patients were less likely than White patients to report new symptoms (aOR, 0.31; 95% CI, 0.12-0.83; and aOR, 0.38; 95% CI, 0.21-0.71, respectively). Longer lengths of hospital stay were associated with greater 1-month cardiopulmonary symptoms (aOR, 1.82 per additional week in the hospital; 95% CI, 1.11-2.98), but discharge destination was not (aOR, 0.92; 95% CI, 0.39-1.71).

New limitations in activities of daily living or instrumental activities of daily living were present in 130 (52.8%; 95% CI, 46.4%-59.5%) patients (seven not responding), all of whom had 0 to 3 limitations before their COVID-19 hospitalization. Indeed, 62 (25.2%; 95% CI, 19.9%-31.1%) reported 3 or more new health-related limitations in activities of daily living or instrumental activities of daily living compared to their pre-COVID-19 baseline, as assessed separately during their hospitalization (Figure 2; rates of limitations in individual activities are shown in Appendix Table 3). Older patients were more likely to report a new health-related limitation, and Hispanic patients were less likely to have a new limitation. New limitations were common among patients discharged home without home health services. The number of new cardiopulmonary symptoms explained 11.2% of the variance in the number of new limitations in activities of daily living, a Spearman rank correlation of 0.30 (P < .0001; see Appendix Table 4). More than three in four COVID-19 patients reported new or worsened cardiopulmonary symptoms or new health-related limitations in activities of daily living at 1 month—only 62 (24.5%) patients reported neither.

At 1 month after hospital discharge, 213 (84.2%) patients reported that they were not fully back to their pre-COVID-19 level of functioning (3 declined to answer the question). When asked, “On a scale of 1 to 100, with 100 being all the way back to what you could do before COVID, how close to being back are you?” the median response was 80, with an IQR of 64-95 (Figure 3). Forty-two (16.8%; 95% CI, 12.4%-22.0%) patients reported a level of 50 or below. Women and older patients reported lower levels of return of functioning, as did those with longer hospital stays and new or worsened cardiopulmonary symptoms. Each additional week in hospital length of stay was associated with a 7.5-point lower response to the question (95% CI, –11.2 to –3.8), but discharge destination was not associated with the answer after adjusting for demographics. Patients with and without new limitations in activities of daily living and with and without new cardiopulmonary symptoms were found across the range of self-reported degree of recovery, although patients without a new problem in one of those domains were rarer among those reporting recovery of less than 70. The number of new cardiopulmonary symptoms explained 19.7% of the variance in the response to this question, a Spearman rank correlation of 0.47 (P < .0001).

More than half of respondents (115 [55.0%]; 95% CI, 48.0%-61.9%; 44 not responding) stated that their COVID-19 hospitalization had been a drain on the finances of their family; 53 (25.4%; 95% CI, 19.6%-31.8%; 44 not responding) rated that drain as moderate, severe, or extreme within the first month after hospital discharge. Forty-nine patients (19.8%; 95% CI, 15.1%-25.4%; 6 not responding) reported that they had to change their work because of their COVID-19 hospitalization, and 93 patients (37.8%; 95% CI, 31.7%-44.2%; 7 not responding) reported that a loved one had taken time off work to care for them. Altogether, one in five COVID-19 patients reported that, within the first month after hospital discharge, they used all or most of their savings because of their COVID-19 illness or hospitalization (58 [23.2%]; 95% CI, 18.1%-29.9%; 3 not responding). There were no demographic differences in the likelihood of losing a job or having a loved one take time off for caregiving, but non-Hispanic Black and Hispanic patients were much more likely to report having used all or most of their savings (aOR, 2.96; 95% CI, 1.09-8.04; and aOR, 2.68; 95% CI, 1.35-5.31, respectively) than White patients. Hospital length of stay and discharge destination were not consistently associated with these financial toxicities. The development of new or worsened cardiopulmonary symptoms was not associated with job change or having a caregiver take time off but was associated with increased likelihood of having used all or most savings (aOR, 2.30; 95% CI, 1.12-4.37).

In a geographically and demographically diverse national US cohort, we found that a decline in perceived health, new or worsened cardiopulmonary symptoms, new limitations in activities of daily living, and new financial stresses were common among patients hospitalized during the US third wave of COVID-19 at 1 month after hospital discharge. The new cardiopulmonary symptoms were significantly associated with the self-report of incomplete recovery and financial stress, but less closely associated with incident disability, inability to work, and caregiving receipt. There were not consistent differences between any demographic groups on these outcomes. Patients with longer lengths of stay generally reported more problems. New problems were very common among patients discharged directly home without home health services.

These data suggest a broad range of new problems among survivors of COVID-19 hospitalization. Moreover, these problems are not well-correlated with each other. This raises the possibility that there may be multiple phenotypes of post-acute sequelae after COVID-19 hospitalization. It is not clear to what extent these differences are mediated by differences in tissue damage from or immunologic response to SARS-CoV-2, distinct from or interacting with other elements of treatment, hospitalization, or the illness experience. The degree of financial stress, savings loss, and job dislocation reported here suggests these patients will face substantial challenges in guiding their own recovery in the absence of a dedicated set of services.^28,29The persistent symptoms faced by these COVID-19 patients can be considered in the context of post-acute sequelae among survivors of community-acquired pneumonia in previous studies, as summarized in a recent systematic review.³⁰ For example, only 35% of a large cohort of adults with community-acquired pneumonia who were evaluated in the emergency department were completely free of pneumonia-related symptoms 6 weeks after antibiotic therapy.^31,32 Limitations in activities of daily living have been reported at 1 month after community-acquired pneumonia³³; rehospitalization and early post-discharge mortality rates may also be similar.^34,35 These findings suggest that the persistent problems of both COVID-19 and other pneumonia patients may highlight important opportunities for improvements in healthcare systems,³⁶ and that burdensome postacute sequelae of COVID-19 may not be attributable solely to distinctive features of the SARS-CoV-2 virus.

A majority of patients discharged home without home health services reported new difficulties in their activities of daily living; 77% of patients with new disability at 1 month had been discharged without home services. These data, however, do not show to what extent this lack of home health services resulted from lack of referral for services, home health provider unavailability, or patient refusal of recommended services. Nonetheless, this nonreceipt of home health services may have been consequential. Among hospitalized patients recovering from pneumonia pre-COVID, the use of post-hospital physical and occupational therapy was associated with reduced risk of readmissions and death.³⁷ This association was greater among patients with lower baseline mobility scores and in patients discharged to home directly. Further, the risk of poor outcomes decreased in a dose-response fashion with increased post-hospital therapy delivery. Failure to provide services for postdischarge disability was previously identified as a potential vulnerability of patients during COVID-19.³⁸

This study adds to the literature. The focus on sequelae perceived by the patient to be incident, as distinguished from symptoms and disability existing before COVID-19, increases the likelihood that these data reflect the influence of the COVID-19 hospitalization. These data emphasize that, despite relatively brief hospitalizations, diverse problems are quite common and consequential for patients’ ability to return to their pre-COVID-19 roles. They further add to the literature by demonstrating the relatively loose coupling between various ways in which postacute sequelae of COVID-19 might be defined: the cardiopulmonary symptoms examined here, the patient’s reported completeness of recovery, the financial stresses the hospitalization placed on the patient and their family, or the development of new limitations in activities of daily living.

Our findings highlight a potential second public health crisis from COVID, related to post-COVID recovery, resulting from the incident disability and economic loss among COVID survivors. While much attention is paid to deaths from COVID, there is less (albeit growing) recognition of the long-term consequences in survivors of COVID-19.³⁹ The downstream economic impacts from job loss and financial insolvency for COVID-19 survivors have ramifications for caregivers, family units that include dependents, and the broader US economy—and may do so for generations if uncorrected, as has been suggested after the 1918 influenza pandemic.⁴⁰ These data may, indeed, look worse at later follow-up given the delay in hospital billing and new expenses in the wake of illness and hospitalization.^28,36,41 It is important that the healthcare system and policymakers consider early investments in post-hospital rehabilitation and adaptive services to allow workers to return to the workforce as soon as possible, and prepare for an increased need for financial support for recovering COVID patients.⁴²

Importantly, these data cannot distinguish between the impact of SARS-CoV-2 infection itself from the treatment received for COVID-19 or other non-COVID-19-specific aspects of hospital care. COVID-19 inpatient case fatality rates and management have changed over time, and so generalizability to future cohorts is unknown.^9-11 This cohort was recruited in the inpatient setting at largely teaching hospitals; therefore, these patients’ experience may be not be representative of all hospitalized COVID-19 patients during this time period. The generalizability of hospital-based studies to patients not hospitalized for COVID-19 remains a subject of active inquiry. We only interviewed patients who were not homeless (excluding 7 of 588 eligible, 1.2%) and who spoke English or Spanish (excluding 4 of 588 eligible, 0.7%); these and other inclusion/exclusion criteria should be considered when evaluating the generalizability of these findings to other patients. We did not prospectively collect measures of fatigue to examine this important and complex symptom, nor did we evaluate outpatient therapy. Finally, self-report was used, rather than using objective measurements of what the patient did or did not do in their home environment. This is consistent with clinical practice that emphasizes patients as primary reporters of their present state, but may introduce measurement error compared to more invasive strategies if those are considered the gold standard.

Patients who survived hospitalization from COVID-19 during the period of August 2020 to January 2021 continued to face significant burdens of new cardiopulmonary symptoms, incomplete recovery, disability, and financial toxicity, all of which extend to patients discharged directly home without services. The correlations between these potential symptoms are no more than partial, and an exclusive focus on one area may neglect other areas of patient need.

The authors thank the patients and families of the Biology and Longitudinal Epidemiology: COVID-19 Observational (BLUE CORAL) study for their generous sharing of their time with us. We acknowledge Hallie C Prescott (University of Michigan and VA Ann Arbor) for her assistance in developing the financial toxicity questions.

For many patients hospitalized with COVID-19, the impact of the illness continues well beyond hospital discharge.¹ Heavy burdens of persistent symptoms have been reported, albeit often from regional and single-hospital samples.^2-7 Critically, not all initial reports capture information on pre-COVID-19 symptom burden, so it is unclear whether these highly prevalent problems are truly new; an alternative explanation might be that patients already with symptoms were more likely to be infected with or seek care for SARS-CoV-2.⁸

Fewer data are available about patients’ abilities to go about the activities of their lives, nor is as much known about the relationships between new symptoms and other impacts. Most of the available information is from health systems during the initial surge of COVID-19 in early 2020—when testing for SARS-CoV-2 was limited even in the inpatient setting; when hospitals’ postdischarge care systems may have been heavily disrupted; and when clinicians were often reasonably focused primarily on reducing mortality in their first cases of COVID-19 rather than promoting recovery from an often-survivable illness. Increasing evidence shows that the inpatient case-fatality rate of COVID-19 is improving over time^9,10; this makes unclear the generalizability of outcomes data from early COVID-19 patients to more recent patients.¹¹

Therefore, we report multicenter measurements of incident levels of persistent cardiopulmonary symptoms, disability, return to baseline, and impact on employment among a recent cohort of COVID-19 patients hospitalized around the United States during the “third wave” of COVID-19—fall and winter 2020-2021. We focus on the 1-month time point after hospital discharge, as this time point is still in the early vulnerable period during which hospital transition-of-care programs are understood to have responsibility.

The first 253 patients who completed 1-month postdischarge telephone follow-up surveys from the ongoing nationwide BLUE CORAL study were included. BLUE CORAL will enroll up to 1,500 hospitalized COVID-19 patients at 36 US centers (the identities of which are reported in Appendix 1) as a part of the National Heart, Lung, and Blood Institute’s Prevention and Early Treatment of Acute Lung Injury (PETAL) Network. We report here on survey questions that allowed for a clear comparison to be made between 1-month follow-up responses and pre-COVID baseline variables; these comparisons were based on (1) previous in-hospital assessment; (2) explicitly asking patients to compare to pre-COVID-19 levels; or (3) explicitly asking patients for changes in relation to their COVID-19 hospitalization. Items were chosen for inclusion in this report without looking at their association with other variables.

This research was approved by the Vanderbilt Institutional Review Board (IRB), serving as central IRB for the PETAL Network; patients or their surrogates provided informed consent.

Participants

Patients with COVID-19 were identified during hospitalization and within 14 days of a positive molecular test for SARS-CoV-2. Eligible patients presented with fever and/or respiratory signs/symptoms, such as hypoxemia, shortness of breath, or infiltrates on chest imaging. Patients were enrolled within the first 72 hours of hospitalization (in order to avoid oversampling patients with relatively longer stays, and to study the biology of early COVID-19), and excluded if they had comfort-care orders (because of their limited likelihood of surviving to follow-up), or were incarcerated (because of difficulties in obtaining truly open informed consent and likely difficulties in follow-up). Pertinently, patients were not required to be in the intensive care unit.

Surviving patients who spoke English or Spanish, were not homeless on hospital admission, and were neither significantly disabled nor significantly cognitively impaired were eligible for follow-up. “Not significantly disabled” was defined as having limitations due to health on no more than three activities of daily living before their COVID-19 hospitalization, as assessed at BLUE CORAL enrollment; this was chosen because of the potentially limited sensitivity of many of our questionnaires to detect an impact of COVID-19 in patients with greater than this level of disability. We included patients who were able to consent for themselves in the study, or for whom the legally appointed representative consenting on their behalf in the hospital reported no evidence of cognitive impairment, defined as no more than four of the problems on the eight-item Alzheimer’s Dementia (AD8) scale.^12-14

Data Collection

One-month surveys were administered to patients or, when necessary, their proxies; the complete English- and Spanish-language instruments are presented in Appendix 2. Study data were collected and managed using REDCap electronic data capture tools hosted at the University of Michigan.^15,16

Patients were contacted via phone by trained interviewers beginning 21 days after hospital discharge; interviews were completed a median of 47 days after discharge (interquartile range [IQR], 26-61). Efforts prioritized former patients completing surveys themselves by phone, but a well-informed proxy was approached if needed. Proxies, who included spouses, adult children, or other relatives, family friends, or primary caregivers, were in regular contact with the patient and understood the patient’s health status. If necessary, the survey could be completed over multiple phone calls, and a written, mail-back option was available. Other best practices in accurate survey data collection and cohort retention were used.^17-19 Participants were given a $10 gift card.

New cardiopulmonary symptoms were queried with symptom-targeted questions informed by the Airways Questionnaire 20,²⁰ the Kansas City Cardiomyopathy Questionnaire,^21,22 and the Seattle Angina Questionnaire.²³ Whenever a respondent reported a given symptom, they were asked, “Compared to 1 month before your COVID-19 hospitalization, is this better, worse, or about the same?” We counted the number of symptoms which the patient reported as worse.

Using wording from the Health and Retirement Study,²⁴ disability was assessed based on a self-report of any of 14 health-related limitations in activities of daily living or instrumental activities of daily living, as in past studies²⁵: dressing, walking across a room, bathing, eating, getting out of bed, using a toilet, using a map, preparing a hot meal, shopping for groceries, making a phone call, taking medications, paying bills, carrying 10 lb (eg, a heavy bag of groceries), and, as a combined single item, stooping, kneeling, or crouching. Well-chosen proxy reports appear reliable for these items.²⁶ We counted the number of activities for which the patient reported a limitation, comparing those reported at 1 month to those reported during the in-hospital survey assessing pre-illness functioning.

The financial consequences of the COVID-19 hospitalization were assessed in two ways. First, we used a modified version of a World Health Organization Disability Assessment Schedule (WHODAS) 2.0 question²⁷: “Since your COVID-19 hospitalization, how much has your health been a drain on the financial resources of you or your family?” Second, we used the financial toxicity items developed with the Mi-COVID19 study³ based on extensive qualitative interviews with respiratory failure survivors²⁸; these questions were anchored explicitly on “the financial cost of dealing with your COVID-19 hospitalization and related care.”

Data Analysis

There were few missing data, and almost all were on outcome variables. Where present, the degree of missingness is reported and casewise deletion used. Because this was a planned early look at responses to an ongoing survey, with analysis based on the number of accrued responses, the ultimate denominator for response rate calculation is unknown. Therefore, two bounds are presented—the minimum, on the assumption that all remaining uncompleted surveys will be missed; and the maximum, as if the uncompleted surveys were not yet in the eligible denominator.

Variables were summarized with medians and IQRs. Multilevel logistic regression was used to test for differences across demographic characteristics in the rates of development of any new symptom or disability; site-level differences were modeled using a random effect. Gender, race/ethnicity, and age were included in all regressions unless noted otherwise; age was included with both linear and quadratic terms when used as a control variable. For the degree of return to baseline and for the number of new limitations in activities of daily living, we explored associations as dichotomized variables (any/none, using multilevel logistic regression) and as continuous variables (using multilevel linear regression). Percent of variance explained was calculated using the R² in unadjusted linear regression, and Spearman rank correlations were used to allow nonlinearities in comparisons across outcomes. All adjusted models are presented in Appendix Table 1. Analyses were conducted in Stata 16.1 (StataCorp, 2020); analytic code is presented in Appendix 3, and a log file of all analyses is in Appendix 4.

The 250th 1-month follow-up was completed on February 26, 2021. One month prior, 647 patients had been recruited at 26 centers in the inpatient phase of the study. Patient demographics for the 253 patients surveyed through that date are shown in Appendix Table 2. On the day of the early look at the data, 460 patients had become eligible for 1-month follow-up and 64 patients had been missed for 1-month follow-up (maximum response rate of 79.8%, minimum possible final response rate of 55.0%) (Figure 1). Seven surveys were completed by proxies. Respondents’ median age was 60 years (IQR, 45-68), and 111 (43.4%) were female. Their median hospital length of stay was 5 days(IQR, 3-8) . A total of 236 (93.3%) patients were discharged home, including 197 (77.9%) without home care services and 39 (15.4%) with home care services.

One hundred and thirty-nine patients (56.5%; 95% CI, 50.1%-62.8%) reported at least one new or worsened cardiopulmonary symptom after their COVID-19 hospitalization (Table; seven patients did not respond to these questions). Most patients with new symptoms had one (48 [19.5%]; 95% CI, 14.8%-25.0%) or two (32 [13%]; 95% CI, 9.7%-17.7%) of the new symptoms queried. The most common new cardiopulmonary symptom was cough, reported by 57 (23.2%; 95% CI, 18.0%-29.0%) patients. New oxygen use was reported by 28 (11.4%; 95% CI, 7.7%-16.0%) patients, with another 11 (4.5%; 95% CI, 2.3%-7.9%) reporting increased oxygen requirements. Women were twice as likely as men to report any new cardiopulmonary symptom (adjusted odds ratio [aOR], 2.24; 95% CI, 1.29-3.90) and non-Hispanic Black and Hispanic patients were less likely than White patients to report new symptoms (aOR, 0.31; 95% CI, 0.12-0.83; and aOR, 0.38; 95% CI, 0.21-0.71, respectively). Longer lengths of hospital stay were associated with greater 1-month cardiopulmonary symptoms (aOR, 1.82 per additional week in the hospital; 95% CI, 1.11-2.98), but discharge destination was not (aOR, 0.92; 95% CI, 0.39-1.71).

New limitations in activities of daily living or instrumental activities of daily living were present in 130 (52.8%; 95% CI, 46.4%-59.5%) patients (seven not responding), all of whom had 0 to 3 limitations before their COVID-19 hospitalization. Indeed, 62 (25.2%; 95% CI, 19.9%-31.1%) reported 3 or more new health-related limitations in activities of daily living or instrumental activities of daily living compared to their pre-COVID-19 baseline, as assessed separately during their hospitalization (Figure 2; rates of limitations in individual activities are shown in Appendix Table 3). Older patients were more likely to report a new health-related limitation, and Hispanic patients were less likely to have a new limitation. New limitations were common among patients discharged home without home health services. The number of new cardiopulmonary symptoms explained 11.2% of the variance in the number of new limitations in activities of daily living, a Spearman rank correlation of 0.30 (P < .0001; see Appendix Table 4). More than three in four COVID-19 patients reported new or worsened cardiopulmonary symptoms or new health-related limitations in activities of daily living at 1 month—only 62 (24.5%) patients reported neither.

At 1 month after hospital discharge, 213 (84.2%) patients reported that they were not fully back to their pre-COVID-19 level of functioning (3 declined to answer the question). When asked, “On a scale of 1 to 100, with 100 being all the way back to what you could do before COVID, how close to being back are you?” the median response was 80, with an IQR of 64-95 (Figure 3). Forty-two (16.8%; 95% CI, 12.4%-22.0%) patients reported a level of 50 or below. Women and older patients reported lower levels of return of functioning, as did those with longer hospital stays and new or worsened cardiopulmonary symptoms. Each additional week in hospital length of stay was associated with a 7.5-point lower response to the question (95% CI, –11.2 to –3.8), but discharge destination was not associated with the answer after adjusting for demographics. Patients with and without new limitations in activities of daily living and with and without new cardiopulmonary symptoms were found across the range of self-reported degree of recovery, although patients without a new problem in one of those domains were rarer among those reporting recovery of less than 70. The number of new cardiopulmonary symptoms explained 19.7% of the variance in the response to this question, a Spearman rank correlation of 0.47 (P < .0001).

More than half of respondents (115 [55.0%]; 95% CI, 48.0%-61.9%; 44 not responding) stated that their COVID-19 hospitalization had been a drain on the finances of their family; 53 (25.4%; 95% CI, 19.6%-31.8%; 44 not responding) rated that drain as moderate, severe, or extreme within the first month after hospital discharge. Forty-nine patients (19.8%; 95% CI, 15.1%-25.4%; 6 not responding) reported that they had to change their work because of their COVID-19 hospitalization, and 93 patients (37.8%; 95% CI, 31.7%-44.2%; 7 not responding) reported that a loved one had taken time off work to care for them. Altogether, one in five COVID-19 patients reported that, within the first month after hospital discharge, they used all or most of their savings because of their COVID-19 illness or hospitalization (58 [23.2%]; 95% CI, 18.1%-29.9%; 3 not responding). There were no demographic differences in the likelihood of losing a job or having a loved one take time off for caregiving, but non-Hispanic Black and Hispanic patients were much more likely to report having used all or most of their savings (aOR, 2.96; 95% CI, 1.09-8.04; and aOR, 2.68; 95% CI, 1.35-5.31, respectively) than White patients. Hospital length of stay and discharge destination were not consistently associated with these financial toxicities. The development of new or worsened cardiopulmonary symptoms was not associated with job change or having a caregiver take time off but was associated with increased likelihood of having used all or most savings (aOR, 2.30; 95% CI, 1.12-4.37).

In a geographically and demographically diverse national US cohort, we found that a decline in perceived health, new or worsened cardiopulmonary symptoms, new limitations in activities of daily living, and new financial stresses were common among patients hospitalized during the US third wave of COVID-19 at 1 month after hospital discharge. The new cardiopulmonary symptoms were significantly associated with the self-report of incomplete recovery and financial stress, but less closely associated with incident disability, inability to work, and caregiving receipt. There were not consistent differences between any demographic groups on these outcomes. Patients with longer lengths of stay generally reported more problems. New problems were very common among patients discharged directly home without home health services.

These data suggest a broad range of new problems among survivors of COVID-19 hospitalization. Moreover, these problems are not well-correlated with each other. This raises the possibility that there may be multiple phenotypes of post-acute sequelae after COVID-19 hospitalization. It is not clear to what extent these differences are mediated by differences in tissue damage from or immunologic response to SARS-CoV-2, distinct from or interacting with other elements of treatment, hospitalization, or the illness experience. The degree of financial stress, savings loss, and job dislocation reported here suggests these patients will face substantial challenges in guiding their own recovery in the absence of a dedicated set of services.^28,29The persistent symptoms faced by these COVID-19 patients can be considered in the context of post-acute sequelae among survivors of community-acquired pneumonia in previous studies, as summarized in a recent systematic review.³⁰ For example, only 35% of a large cohort of adults with community-acquired pneumonia who were evaluated in the emergency department were completely free of pneumonia-related symptoms 6 weeks after antibiotic therapy.^31,32 Limitations in activities of daily living have been reported at 1 month after community-acquired pneumonia³³; rehospitalization and early post-discharge mortality rates may also be similar.^34,35 These findings suggest that the persistent problems of both COVID-19 and other pneumonia patients may highlight important opportunities for improvements in healthcare systems,³⁶ and that burdensome postacute sequelae of COVID-19 may not be attributable solely to distinctive features of the SARS-CoV-2 virus.

A majority of patients discharged home without home health services reported new difficulties in their activities of daily living; 77% of patients with new disability at 1 month had been discharged without home services. These data, however, do not show to what extent this lack of home health services resulted from lack of referral for services, home health provider unavailability, or patient refusal of recommended services. Nonetheless, this nonreceipt of home health services may have been consequential. Among hospitalized patients recovering from pneumonia pre-COVID, the use of post-hospital physical and occupational therapy was associated with reduced risk of readmissions and death.³⁷ This association was greater among patients with lower baseline mobility scores and in patients discharged to home directly. Further, the risk of poor outcomes decreased in a dose-response fashion with increased post-hospital therapy delivery. Failure to provide services for postdischarge disability was previously identified as a potential vulnerability of patients during COVID-19.³⁸

This study adds to the literature. The focus on sequelae perceived by the patient to be incident, as distinguished from symptoms and disability existing before COVID-19, increases the likelihood that these data reflect the influence of the COVID-19 hospitalization. These data emphasize that, despite relatively brief hospitalizations, diverse problems are quite common and consequential for patients’ ability to return to their pre-COVID-19 roles. They further add to the literature by demonstrating the relatively loose coupling between various ways in which postacute sequelae of COVID-19 might be defined: the cardiopulmonary symptoms examined here, the patient’s reported completeness of recovery, the financial stresses the hospitalization placed on the patient and their family, or the development of new limitations in activities of daily living.

Our findings highlight a potential second public health crisis from COVID, related to post-COVID recovery, resulting from the incident disability and economic loss among COVID survivors. While much attention is paid to deaths from COVID, there is less (albeit growing) recognition of the long-term consequences in survivors of COVID-19.³⁹ The downstream economic impacts from job loss and financial insolvency for COVID-19 survivors have ramifications for caregivers, family units that include dependents, and the broader US economy—and may do so for generations if uncorrected, as has been suggested after the 1918 influenza pandemic.⁴⁰ These data may, indeed, look worse at later follow-up given the delay in hospital billing and new expenses in the wake of illness and hospitalization.^28,36,41 It is important that the healthcare system and policymakers consider early investments in post-hospital rehabilitation and adaptive services to allow workers to return to the workforce as soon as possible, and prepare for an increased need for financial support for recovering COVID patients.⁴²

Importantly, these data cannot distinguish between the impact of SARS-CoV-2 infection itself from the treatment received for COVID-19 or other non-COVID-19-specific aspects of hospital care. COVID-19 inpatient case fatality rates and management have changed over time, and so generalizability to future cohorts is unknown.^9-11 This cohort was recruited in the inpatient setting at largely teaching hospitals; therefore, these patients’ experience may be not be representative of all hospitalized COVID-19 patients during this time period. The generalizability of hospital-based studies to patients not hospitalized for COVID-19 remains a subject of active inquiry. We only interviewed patients who were not homeless (excluding 7 of 588 eligible, 1.2%) and who spoke English or Spanish (excluding 4 of 588 eligible, 0.7%); these and other inclusion/exclusion criteria should be considered when evaluating the generalizability of these findings to other patients. We did not prospectively collect measures of fatigue to examine this important and complex symptom, nor did we evaluate outpatient therapy. Finally, self-report was used, rather than using objective measurements of what the patient did or did not do in their home environment. This is consistent with clinical practice that emphasizes patients as primary reporters of their present state, but may introduce measurement error compared to more invasive strategies if those are considered the gold standard.

Patients who survived hospitalization from COVID-19 during the period of August 2020 to January 2021 continued to face significant burdens of new cardiopulmonary symptoms, incomplete recovery, disability, and financial toxicity, all of which extend to patients discharged directly home without services. The correlations between these potential symptoms are no more than partial, and an exclusive focus on one area may neglect other areas of patient need.

The authors thank the patients and families of the Biology and Longitudinal Epidemiology: COVID-19 Observational (BLUE CORAL) study for their generous sharing of their time with us. We acknowledge Hallie C Prescott (University of Michigan and VA Ann Arbor) for her assistance in developing the financial toxicity questions.

The American Board of Medical Specialties approved subspecialty designation for the field of pediatric hospital medicine (PHM) in 2016.¹ For those who started independent practice prior to July 2019, there were two options for board eligibility: the “practice pathway” or completion of a PHM fellowship. The practice pathway allows for pediatric and combined internal medicine–pediatric (med-peds) providers who graduated by July 2019 to sit for the PHM board-certification examination if they meet specific criteria in their pediatric practice.² For pediatric and med-peds residents who graduated after July 2019, PHM board eligibility is available only through completion of a PHM fellowship.

PHM subspecialty designation with fellowship training requirements may pose unique challenges to med-peds residents interested in practicing both pediatric and adult hospital medicine (HM).^3,4 Each year, an estimated 25% of med-peds residency graduates go on to practice HM.⁵ The majority (62%-83%) of currently practicing med-peds–trained hospitalists care for both adults and children.^5,6 Further, med-peds–trained hospitalists comprise at least 10% of the PHM workforce⁵ and play an important role in caring for adult survivors of childhood diseases.³

Limited existing data suggest that the future practice patterns of med-peds residents may be affected by PHM fellowship requirements. One previous survey study indicated that, although med-peds residents see value in additional training opportunities offered by fellowship, the majority are less likely to pursue PHM as a result of the new requirements.⁴ Prominent factors dissuading residents from pursuing PHM fellowship included forfeited earnings during fellowship, student loan obligations, family obligations, and the perception that training received during residency was sufficient. Although these data provide important insights into potential changes in practice patterns, they do not explore qualities of PHM fellowship that may make additional training more appealing to med-peds residents and promote retention of med-peds–trained providers in the PHM workforce.

Further, there is no existing literature exploring if and how PHM fellowship programs are equipped to support the needs of med-peds–trained fellows. Other subspecialties have supported med-peds trainees in combined fellowship training programs, including rheumatology, neurology, pediatric emergency medicine, allergy/immunology, physical medicine and rehabilitation, and psychiatry.^7,8 However, the extent to which PHM fellowships follow a similar model to accommodate the career goals of med-peds participants is unclear.

Given the large numbers of med-peds residents who go on to practice combined PHM and adult HM, it is crucial to understand the training needs of this group within the context of PHM fellowship and board certification. The primary objectives of this study were to understand (1) the perceived PHM fellowship needs of med-peds residents interested in HM, and (2) how the current PHM fellowship training environment can meet those needs. Understanding that additional training requirements to practice PHM may affect the career trajectory of residents interested in HM, secondary objectives included describing perceptions of med-peds residents on PHM specialty designation and whether designation affected their career plans.

Study Design

This cross-sectional study took place over a 3-month period from May to July 2019 and included two surveys of different populations to develop a comprehensive understanding of stakeholder perceptions of PHM fellowship. The first survey (resident survey) invited med-peds residents who were members of the National Med-Peds Residents’ Association (NMPRA)⁹ in 2019 and who were interested in HM. The second survey (fellowship director [FD] survey) included PHM FDs. The study was determined to be exempt by the University of Pittsburgh Institutional Review Board.

Study Population and Recruitment

Resident Survey

Two attempts were made to elicit participation via the NMPRA electronic mailing list. The NMPRA membership includes med-peds residents and chief residents from US med-peds residency programs. As of May 2019, 77 med-peds residency programs and their residents were members of NMPRA, which encompassed all med-peds programs in the United States and its territories. NMPRA maintains a listserv for all members, and all existing US/territory programs were members at the time of the survey. Med-peds interns, residents, and chief residents interested in HM were invited to participate in this study.

FD Survey

Forty-eight FDs, representing member institutions of the PHM Fellowship Directors’ Council, were surveyed via the PHM Fellowship Directors listserv.

Survey Instruments

We constructed two de novo surveys consisting of multiple-choice and short-answer questions (Appendix 1 and Appendix 2). To enhance the validity of survey responses, questions were designed and tested using an iterative consensus process among authors and additional participants, including current med-peds PHM fellows, PHM fellowship program directors, med-peds residency program directors, and current med-peds residents. These revisions were repeated for a total of four cycles. Items were created to increase knowledge on the following key areas: resident-perceived needs in fellowship training, impact of PHM subspecialty designation on career choices related to HM, health system structure of fellowship programs, and ability to accommodate med-peds clinical training within a PHM fellowship. A combined med-peds fellowship, as defined in the survey and referenced in this study, is a “combined internal medicine–pediatrics hospital medicine fellowship whereby you would remain eligible for PHM board certification.” To ensure a broad and inclusive view of potential needs of med-peds trainees considering fellowship, all respondents were asked to complete questions pertaining to anticipated fellowship needs regardless of their indicated interest in fellowship.

Data Collection

Survey completion was voluntary. Email identifiers were not linked to completed surveys. Study data were collected and managed by using Qualtrics XM. Only completed survey entries were included in analysis.

Statistical Methods and Data Analysis

R software version 4.0.2 (R Foundation for Statistical Computing) was used for statistical analysis. Demographic data were summarized using frequency distributions. The intent of the free-text questions for both surveys was qualitative explanatory thematic analysis. Authors EB, HL, and AJ used a deductive approach to identify common themes that elucidated med-peds resident–anticipated needs in fellowship and PHM program strategies and barriers to accommodate these needs. Preliminary themes and action items were reviewed and discussed among the full authorship team until consensus was reached.

Demographic Data

Resident Survey

A total of 466 med-peds residents completed the resident survey. There are approximately 1300 med-peds residents annually, creating an estimated response rate of 35.8% of all US med-peds residents. The majority (n = 380, 81.5%) of respondents were med-peds postgraduate years 1 through 3 and thus only eligible for PHM board certification via the PHM fellowship pathway (Table 1). Most (n = 446, 95.7%) respondents had considered a career in adult, pediatric, or combined HM at some point. Of those med-peds residents who considered a career in HM (Appendix Table 1), 92.8% (n = 414) would prefer to practice combined adult HM and PHM.

FD Survey

Twenty-eight FDs completed the FD survey, representing 58.3% of 2019 PHM fellowship programs. Of the responding programs, 23 (82.1%) were associated with a freestanding children’s hospital, and 24 (85.7%) were integrated or affiliated with a health system that provides adult inpatient care (Table 2). Sixteen (57.1%) programs had a med-peds residency program at their institution.

Med-Peds Resident Perceptions of PHM Fellowship

In considering the importance of PHM board certification for physicians practicing PHM, 59.0% (n= 275) of respondents rated board certification as “not at all important” (Appendix Table 2). Most (n = 420, 90.1%) med-peds trainees responded that PHM subspecialty designation “decreased” or “significantly decreased” their desire to pursue a career that includes PHM. Of the respondents who reported no interest in hospital medicine, eight (40%) reported that PHM subspecialty status dissuaded them from a career in HM at least a moderate amount (Appendix Table 3). Roughly one third (n=158, 33.9%) of respondents reported that PHM subspecialty designation increased or significantly increased their desire to pursue a career that includes adult HM (Appenidx Table 2). Finally, although the majority (n = 275, 59%) of respondents said they had no interest in a HM fellowship, 114 (24.5%) indicated interest in a combined med-peds HM fellowship (Appendix Table 1). Short-answer questions revealed that commitment to additional training on top of a 4-year residency program was a possible deterring factor, particularly in light of student loan debt and family obligations. Respondents reported adequate clinical training during residency as another deterring factor.

Med-Peds Resident–Perceived Needs in PHM Fellowship

Regardless of interest in completing a PHM fellowship, all resident survey respondents were asked how their ideal PHM fellowship should be structured. Almost all (n = 456, 97.9%) respondents indicated that they would prefer to complete a combined med-peds HM fellowship (Table 3), and most preferred to complete a fellowship in 2 years. Only 10 (2.1%) respondents preferred to complete a PHM fellowship alone in 2 or 3 years. More than half (n=253, 54.3%) of respondents indicated that it would be ideal to obtain a master’s degree as part of fellowship.

Three quarters (n = 355, 75.8%) of med-peds residents reported that they would want 41% or more of clinical time in an ideal fellowship dedicated to adult HM. Importantly, most (n = 322, 69.1%) of the med-peds residents did not consider moonlighting alone in either PHM or adult HM to be enough to maintain training. In addition, many (n = 366, 78.5%) respondents felt that it was important or very important for scholarly work during fellowship to bridge pediatrics and internal medicine.

Short-answer questions indicated that the ability to practice both internal medicine and pediatrics during fellowship emerged as an important deciding factor, with emphasis on adequate opportunities to maintain internal medicine knowledge base (Figure). Similarly, access to med-peds mentorship was an important component of the decision. Compensation both during fellowship and potential future earnings was also a prominent consideration.

Capacity of PHM Programs to Support Med-Peds Fellows

Fifteen (53.6%) FDs reported that their programs were able to accommodate both PHM and adult HM clinical time during fellowship, 11 (39.3%) were unsure, and 2 (7.1%) were unable to accommodate both (Table 2).

The options for adult HM clinical time varied by institution and included precepted time on adult HM, full attending privileges on adult HM, and adult HM time through moonlighting only. Short-answer responses from FDs with experience training med-peds fellows cited using PHM elective time for adult HM and offering moonlighting in adult HM as ways to address career goals of med-peds trainees. Scholarship time for fellows was preserved by decreasing required time on pediatric intensive care unit and complex care services.

Accessibility of Med-Peds Mentorship

As noted above, med-peds residents identified mentorship as an important factor in consideration of PHM fellowship. A total of 23 (82.1%) FDs reported their programs had med-peds faculty members within their PHM team (Table 2). The majority (n = 21, 91.3%) of those med-peds faculty had both PHM and adult HM clinical time.

This study characterized the ideal PHM fellowship structure from the perspective of med-peds residents and described the current ability of PHM fellowships to support med-peds residents. The majority of residents stated that they had no interest in an HM fellowship. However, for med-peds residents who considered a career in HM, 88.8% preferred to complete a combined internal medicine and pediatrics HM fellowship with close to half of clinical time dedicated to adult HM. Just over half (53.6%) of programs reported that they could currently accommodate both PHM and adult clinical time during fellowship, and all but two programs reported that they could accommodate both PHM and HM time in the future.

PHM subspecialty designation with associated fellowship training requirements decreased desire to practice HM among med-peds residents who responded to our survey. This reflects findings from a recently published study that evaluated whether PHM fellowship requirements for board certification influenced pediatric and med-peds residents’ decision to pursue PHM in 2018.⁴ Additionally, Chandrasekar et al⁴ found that 87% of respondents indicated that sufficient residency training was an important factor in discouraging them from pursuing PHM fellowship. We noted similar findings in our open-ended survey responses, which indicate that med-peds respondents perceived that the intended purpose of PHM fellowship was to provide additional clinical training, and that served as a deterrent for fellowship. However, the survey by Chandrasekar et al⁴ assessed only four factors for understanding what was important in encouraging pursuit of a PHM fellowship: opportunity to gain new skills, potential increase in salary, opportunity for a master’s degree, and increased prestige. Our survey expands on med-peds residents’ needs, indicating that med-peds residents want a combined med-peds/HM fellowship that allows them to meet PHM board-eligibility requirements while also continuing to develop their adult HM clinical practice and other nonclinical training objectives in a way that combines both adult HM and PHM. Both surveys demonstrate the role that residency program directors and other resident mentors can have in counseling trainees on the nonclinical training objectives of PHM fellowship, including research, quality improvement, medical education, and leadership and clinical operations. Additional emphasis can be placed on opportunities for an individualized curriculum to address the specific career aims of each resident.

In this study, med-peds trainees viewed distribution of clinical time during fellowship as an important factor in pursuing PHM fellowship. The perceived importance of balancing clinical time is not surprising considering that most survey respondents interested in HM ultimately intend to practice both PHM and adult HM. This finding corresponds with current practice patterns of med-peds hospitalists, the majority of whom care for both children and adults.^4,5 Moonlighting in adult medicine was not considered sufficient, suggesting desire for mentorship and training integration on the internal medicine side. Opportunities for trainees to maintain and expand their internal medicine knowledge base and clinical decision-making outside of moonlighting will be key to meeting the needs of med-peds residents in PHM fellowship.

Fortunately, more than half of responding programs reported that they could allow for adult HM practice during PHM fellowship. Twelve programs were unsure if they could accommodate adult HM clinical time, and only two programs reported they could not. We suspect that the ability to support this training with clinical time in both adult HM and PHM is more likely available at programs with established internal medicine relationships, often in the form of med-peds residency programs and med-peds faculty. Further, these established relationships may be more common at pediatric health systems that are integrated or affiliated with an adult health system. Most PHM fellowships surveyed indicated that their pediatric institution had an affiliation with an adult facility, and most had med-peds HM faculty.

Precedent for supporting med-peds fellows is somewhat limited given that only five of the responding PHM fellowship programs reported having fellows with med-peds residency training. However, discrepancies between the expressed needs of med-peds residents and the current Accreditation Council for Graduate Medical Education (ACGME)–accredited PHM fellowship structure highlight opportunities to tailor fellowship training to support the career goals of med-peds residents. The current PHM fellowship structure consists of 26 educational units, with each unit representing 4 calendar weeks. A minimum of eight units are spent on each of the following: core clinical rotations, systems and scholarship, and individualized curriculum.^10,11 The Society of Hospital Medicine has published core competencies for both PHM and adult HM, which highlight significant overlap in each field’s skill competency, particularly in areas such as quality improvement, legal issues and risk management, and handoffs and transitions of care.^12,13 We contend that competencies addressed within PHM fellowship core clinical rotations may overlap with adult HM. Training in adult HM could be completed as part of the individualized curriculum with the ACGME, allowing adult HM practice to count toward this requirement. This would offer med-peds fellows the option to maintain their adult HM knowledge base without eliminating all elective time. Ultimately, it will be important to be creative in how training is accomplished and skills are acquired during both core clinical and individualized training blocks for med-peds trainees completing PHM fellowship.

In order to meet the expressed needs of med-peds residents interested in incorporating both adult HM and PHM into their future careers through PHM fellowship, we offer key recommendations for consideration by the ACGME, PHM FDs, and med-peds program directors (Figure). We encourage current PHM fellowship programs to establish relationships with adult HM programs to develop structured clinical opportunities that will allow fellows to gain the additional clinical training desired.

There were important limitations in this study. First, our estimated response rate for the resident survey was 35.8% of all med-peds residents in 2019, which may be interpreted as low. However, it is important to note that the survey was targeted to residents interested in HM. More than 25% of med-peds residents pursue a career in HM,⁵ suggesting our response rate may be attributed to residents who did not complete the survey because they were interested in other fields. The program director survey response rate was higher at 58.3%, though it is possible that response bias resulted in a higher response rate from programs with the ability to support med-peds trainees. Regardless, data from programs with the ability to support med-peds trainees are highly valuable in describing how PHM fellowship can be inclusive of med-peds–trained physicians interested in pursuing HM.

Both surveys were completed in 2019, prior to the ACGME accreditation of PHM fellowship, which likely presents new, unique challenges to fellowship programs trying to support the needs of med-peds fellows. However, insights noted above from programs with experience training med-peds fellows are still applicable within the constraints of ACGME requirements.

Many med-peds residents express strong interest in practicing HM and including PHM as part of their future hospitalist practice. With the introduction of PHM subspecialty board certification through the American Board of Pediatrics, med-peds residents face new considerations when choosing a career path after residency. The majority of resident respondents express the desire to spend a substantial portion of their clinical practice and/or fellowship practicing adult HM. A majority of PHM fellowships can or are willing to explore how to provide both pediatric and adult hospitalist training to med-peds residency–trained fellows. Understanding the facilitators and barriers to recruiting med-peds trainees for PHM fellowship ultimately has significant implications for the future of the PHM workforce. Incorporating the recommendations noted in this study may increase retention of med-peds providers in PHM by enabling fellowship training and ultimately board certification. Collaboration among the ACGME, PHM program directors, and med-peds residency program directors could help to develop PHM fellowship training programs that will meet the needs of med-peds residents interested in practicing PHM while still meeting ACGME requirements for PHM board eligibility.

The authors thank Dr Anoop Agrawal of National Med-Peds Residents’ Association (NMPRA).

The American Board of Medical Specialties approved subspecialty designation for the field of pediatric hospital medicine (PHM) in 2016.¹ For those who started independent practice prior to July 2019, there were two options for board eligibility: the “practice pathway” or completion of a PHM fellowship. The practice pathway allows for pediatric and combined internal medicine–pediatric (med-peds) providers who graduated by July 2019 to sit for the PHM board-certification examination if they meet specific criteria in their pediatric practice.² For pediatric and med-peds residents who graduated after July 2019, PHM board eligibility is available only through completion of a PHM fellowship.

PHM subspecialty designation with fellowship training requirements may pose unique challenges to med-peds residents interested in practicing both pediatric and adult hospital medicine (HM).^3,4 Each year, an estimated 25% of med-peds residency graduates go on to practice HM.⁵ The majority (62%-83%) of currently practicing med-peds–trained hospitalists care for both adults and children.^5,6 Further, med-peds–trained hospitalists comprise at least 10% of the PHM workforce⁵ and play an important role in caring for adult survivors of childhood diseases.³

Limited existing data suggest that the future practice patterns of med-peds residents may be affected by PHM fellowship requirements. One previous survey study indicated that, although med-peds residents see value in additional training opportunities offered by fellowship, the majority are less likely to pursue PHM as a result of the new requirements.⁴ Prominent factors dissuading residents from pursuing PHM fellowship included forfeited earnings during fellowship, student loan obligations, family obligations, and the perception that training received during residency was sufficient. Although these data provide important insights into potential changes in practice patterns, they do not explore qualities of PHM fellowship that may make additional training more appealing to med-peds residents and promote retention of med-peds–trained providers in the PHM workforce.

Further, there is no existing literature exploring if and how PHM fellowship programs are equipped to support the needs of med-peds–trained fellows. Other subspecialties have supported med-peds trainees in combined fellowship training programs, including rheumatology, neurology, pediatric emergency medicine, allergy/immunology, physical medicine and rehabilitation, and psychiatry.^7,8 However, the extent to which PHM fellowships follow a similar model to accommodate the career goals of med-peds participants is unclear.

Given the large numbers of med-peds residents who go on to practice combined PHM and adult HM, it is crucial to understand the training needs of this group within the context of PHM fellowship and board certification. The primary objectives of this study were to understand (1) the perceived PHM fellowship needs of med-peds residents interested in HM, and (2) how the current PHM fellowship training environment can meet those needs. Understanding that additional training requirements to practice PHM may affect the career trajectory of residents interested in HM, secondary objectives included describing perceptions of med-peds residents on PHM specialty designation and whether designation affected their career plans.

Study Design

This cross-sectional study took place over a 3-month period from May to July 2019 and included two surveys of different populations to develop a comprehensive understanding of stakeholder perceptions of PHM fellowship. The first survey (resident survey) invited med-peds residents who were members of the National Med-Peds Residents’ Association (NMPRA)⁹ in 2019 and who were interested in HM. The second survey (fellowship director [FD] survey) included PHM FDs. The study was determined to be exempt by the University of Pittsburgh Institutional Review Board.

Study Population and Recruitment

Resident Survey

Two attempts were made to elicit participation via the NMPRA electronic mailing list. The NMPRA membership includes med-peds residents and chief residents from US med-peds residency programs. As of May 2019, 77 med-peds residency programs and their residents were members of NMPRA, which encompassed all med-peds programs in the United States and its territories. NMPRA maintains a listserv for all members, and all existing US/territory programs were members at the time of the survey. Med-peds interns, residents, and chief residents interested in HM were invited to participate in this study.

FD Survey

Forty-eight FDs, representing member institutions of the PHM Fellowship Directors’ Council, were surveyed via the PHM Fellowship Directors listserv.

Survey Instruments

We constructed two de novo surveys consisting of multiple-choice and short-answer questions (Appendix 1 and Appendix 2). To enhance the validity of survey responses, questions were designed and tested using an iterative consensus process among authors and additional participants, including current med-peds PHM fellows, PHM fellowship program directors, med-peds residency program directors, and current med-peds residents. These revisions were repeated for a total of four cycles. Items were created to increase knowledge on the following key areas: resident-perceived needs in fellowship training, impact of PHM subspecialty designation on career choices related to HM, health system structure of fellowship programs, and ability to accommodate med-peds clinical training within a PHM fellowship. A combined med-peds fellowship, as defined in the survey and referenced in this study, is a “combined internal medicine–pediatrics hospital medicine fellowship whereby you would remain eligible for PHM board certification.” To ensure a broad and inclusive view of potential needs of med-peds trainees considering fellowship, all respondents were asked to complete questions pertaining to anticipated fellowship needs regardless of their indicated interest in fellowship.

Data Collection

Survey completion was voluntary. Email identifiers were not linked to completed surveys. Study data were collected and managed by using Qualtrics XM. Only completed survey entries were included in analysis.

Statistical Methods and Data Analysis

R software version 4.0.2 (R Foundation for Statistical Computing) was used for statistical analysis. Demographic data were summarized using frequency distributions. The intent of the free-text questions for both surveys was qualitative explanatory thematic analysis. Authors EB, HL, and AJ used a deductive approach to identify common themes that elucidated med-peds resident–anticipated needs in fellowship and PHM program strategies and barriers to accommodate these needs. Preliminary themes and action items were reviewed and discussed among the full authorship team until consensus was reached.

Demographic Data

Resident Survey

A total of 466 med-peds residents completed the resident survey. There are approximately 1300 med-peds residents annually, creating an estimated response rate of 35.8% of all US med-peds residents. The majority (n = 380, 81.5%) of respondents were med-peds postgraduate years 1 through 3 and thus only eligible for PHM board certification via the PHM fellowship pathway (Table 1). Most (n = 446, 95.7%) respondents had considered a career in adult, pediatric, or combined HM at some point. Of those med-peds residents who considered a career in HM (Appendix Table 1), 92.8% (n = 414) would prefer to practice combined adult HM and PHM.

FD Survey

Twenty-eight FDs completed the FD survey, representing 58.3% of 2019 PHM fellowship programs. Of the responding programs, 23 (82.1%) were associated with a freestanding children’s hospital, and 24 (85.7%) were integrated or affiliated with a health system that provides adult inpatient care (Table 2). Sixteen (57.1%) programs had a med-peds residency program at their institution.

Med-Peds Resident Perceptions of PHM Fellowship

In considering the importance of PHM board certification for physicians practicing PHM, 59.0% (n= 275) of respondents rated board certification as “not at all important” (Appendix Table 2). Most (n = 420, 90.1%) med-peds trainees responded that PHM subspecialty designation “decreased” or “significantly decreased” their desire to pursue a career that includes PHM. Of the respondents who reported no interest in hospital medicine, eight (40%) reported that PHM subspecialty status dissuaded them from a career in HM at least a moderate amount (Appendix Table 3). Roughly one third (n=158, 33.9%) of respondents reported that PHM subspecialty designation increased or significantly increased their desire to pursue a career that includes adult HM (Appenidx Table 2). Finally, although the majority (n = 275, 59%) of respondents said they had no interest in a HM fellowship, 114 (24.5%) indicated interest in a combined med-peds HM fellowship (Appendix Table 1). Short-answer questions revealed that commitment to additional training on top of a 4-year residency program was a possible deterring factor, particularly in light of student loan debt and family obligations. Respondents reported adequate clinical training during residency as another deterring factor.

Med-Peds Resident–Perceived Needs in PHM Fellowship

Regardless of interest in completing a PHM fellowship, all resident survey respondents were asked how their ideal PHM fellowship should be structured. Almost all (n = 456, 97.9%) respondents indicated that they would prefer to complete a combined med-peds HM fellowship (Table 3), and most preferred to complete a fellowship in 2 years. Only 10 (2.1%) respondents preferred to complete a PHM fellowship alone in 2 or 3 years. More than half (n=253, 54.3%) of respondents indicated that it would be ideal to obtain a master’s degree as part of fellowship.

Three quarters (n = 355, 75.8%) of med-peds residents reported that they would want 41% or more of clinical time in an ideal fellowship dedicated to adult HM. Importantly, most (n = 322, 69.1%) of the med-peds residents did not consider moonlighting alone in either PHM or adult HM to be enough to maintain training. In addition, many (n = 366, 78.5%) respondents felt that it was important or very important for scholarly work during fellowship to bridge pediatrics and internal medicine.

Short-answer questions indicated that the ability to practice both internal medicine and pediatrics during fellowship emerged as an important deciding factor, with emphasis on adequate opportunities to maintain internal medicine knowledge base (Figure). Similarly, access to med-peds mentorship was an important component of the decision. Compensation both during fellowship and potential future earnings was also a prominent consideration.

Capacity of PHM Programs to Support Med-Peds Fellows

Fifteen (53.6%) FDs reported that their programs were able to accommodate both PHM and adult HM clinical time during fellowship, 11 (39.3%) were unsure, and 2 (7.1%) were unable to accommodate both (Table 2).

The options for adult HM clinical time varied by institution and included precepted time on adult HM, full attending privileges on adult HM, and adult HM time through moonlighting only. Short-answer responses from FDs with experience training med-peds fellows cited using PHM elective time for adult HM and offering moonlighting in adult HM as ways to address career goals of med-peds trainees. Scholarship time for fellows was preserved by decreasing required time on pediatric intensive care unit and complex care services.

Accessibility of Med-Peds Mentorship

As noted above, med-peds residents identified mentorship as an important factor in consideration of PHM fellowship. A total of 23 (82.1%) FDs reported their programs had med-peds faculty members within their PHM team (Table 2). The majority (n = 21, 91.3%) of those med-peds faculty had both PHM and adult HM clinical time.

This study characterized the ideal PHM fellowship structure from the perspective of med-peds residents and described the current ability of PHM fellowships to support med-peds residents. The majority of residents stated that they had no interest in an HM fellowship. However, for med-peds residents who considered a career in HM, 88.8% preferred to complete a combined internal medicine and pediatrics HM fellowship with close to half of clinical time dedicated to adult HM. Just over half (53.6%) of programs reported that they could currently accommodate both PHM and adult clinical time during fellowship, and all but two programs reported that they could accommodate both PHM and HM time in the future.

PHM subspecialty designation with associated fellowship training requirements decreased desire to practice HM among med-peds residents who responded to our survey. This reflects findings from a recently published study that evaluated whether PHM fellowship requirements for board certification influenced pediatric and med-peds residents’ decision to pursue PHM in 2018.⁴ Additionally, Chandrasekar et al⁴ found that 87% of respondents indicated that sufficient residency training was an important factor in discouraging them from pursuing PHM fellowship. We noted similar findings in our open-ended survey responses, which indicate that med-peds respondents perceived that the intended purpose of PHM fellowship was to provide additional clinical training, and that served as a deterrent for fellowship. However, the survey by Chandrasekar et al⁴ assessed only four factors for understanding what was important in encouraging pursuit of a PHM fellowship: opportunity to gain new skills, potential increase in salary, opportunity for a master’s degree, and increased prestige. Our survey expands on med-peds residents’ needs, indicating that med-peds residents want a combined med-peds/HM fellowship that allows them to meet PHM board-eligibility requirements while also continuing to develop their adult HM clinical practice and other nonclinical training objectives in a way that combines both adult HM and PHM. Both surveys demonstrate the role that residency program directors and other resident mentors can have in counseling trainees on the nonclinical training objectives of PHM fellowship, including research, quality improvement, medical education, and leadership and clinical operations. Additional emphasis can be placed on opportunities for an individualized curriculum to address the specific career aims of each resident.

In this study, med-peds trainees viewed distribution of clinical time during fellowship as an important factor in pursuing PHM fellowship. The perceived importance of balancing clinical time is not surprising considering that most survey respondents interested in HM ultimately intend to practice both PHM and adult HM. This finding corresponds with current practice patterns of med-peds hospitalists, the majority of whom care for both children and adults.^4,5 Moonlighting in adult medicine was not considered sufficient, suggesting desire for mentorship and training integration on the internal medicine side. Opportunities for trainees to maintain and expand their internal medicine knowledge base and clinical decision-making outside of moonlighting will be key to meeting the needs of med-peds residents in PHM fellowship.

Fortunately, more than half of responding programs reported that they could allow for adult HM practice during PHM fellowship. Twelve programs were unsure if they could accommodate adult HM clinical time, and only two programs reported they could not. We suspect that the ability to support this training with clinical time in both adult HM and PHM is more likely available at programs with established internal medicine relationships, often in the form of med-peds residency programs and med-peds faculty. Further, these established relationships may be more common at pediatric health systems that are integrated or affiliated with an adult health system. Most PHM fellowships surveyed indicated that their pediatric institution had an affiliation with an adult facility, and most had med-peds HM faculty.

Precedent for supporting med-peds fellows is somewhat limited given that only five of the responding PHM fellowship programs reported having fellows with med-peds residency training. However, discrepancies between the expressed needs of med-peds residents and the current Accreditation Council for Graduate Medical Education (ACGME)–accredited PHM fellowship structure highlight opportunities to tailor fellowship training to support the career goals of med-peds residents. The current PHM fellowship structure consists of 26 educational units, with each unit representing 4 calendar weeks. A minimum of eight units are spent on each of the following: core clinical rotations, systems and scholarship, and individualized curriculum.^10,11 The Society of Hospital Medicine has published core competencies for both PHM and adult HM, which highlight significant overlap in each field’s skill competency, particularly in areas such as quality improvement, legal issues and risk management, and handoffs and transitions of care.^12,13 We contend that competencies addressed within PHM fellowship core clinical rotations may overlap with adult HM. Training in adult HM could be completed as part of the individualized curriculum with the ACGME, allowing adult HM practice to count toward this requirement. This would offer med-peds fellows the option to maintain their adult HM knowledge base without eliminating all elective time. Ultimately, it will be important to be creative in how training is accomplished and skills are acquired during both core clinical and individualized training blocks for med-peds trainees completing PHM fellowship.

In order to meet the expressed needs of med-peds residents interested in incorporating both adult HM and PHM into their future careers through PHM fellowship, we offer key recommendations for consideration by the ACGME, PHM FDs, and med-peds program directors (Figure). We encourage current PHM fellowship programs to establish relationships with adult HM programs to develop structured clinical opportunities that will allow fellows to gain the additional clinical training desired.

There were important limitations in this study. First, our estimated response rate for the resident survey was 35.8% of all med-peds residents in 2019, which may be interpreted as low. However, it is important to note that the survey was targeted to residents interested in HM. More than 25% of med-peds residents pursue a career in HM,⁵ suggesting our response rate may be attributed to residents who did not complete the survey because they were interested in other fields. The program director survey response rate was higher at 58.3%, though it is possible that response bias resulted in a higher response rate from programs with the ability to support med-peds trainees. Regardless, data from programs with the ability to support med-peds trainees are highly valuable in describing how PHM fellowship can be inclusive of med-peds–trained physicians interested in pursuing HM.

Both surveys were completed in 2019, prior to the ACGME accreditation of PHM fellowship, which likely presents new, unique challenges to fellowship programs trying to support the needs of med-peds fellows. However, insights noted above from programs with experience training med-peds fellows are still applicable within the constraints of ACGME requirements.

Many med-peds residents express strong interest in practicing HM and including PHM as part of their future hospitalist practice. With the introduction of PHM subspecialty board certification through the American Board of Pediatrics, med-peds residents face new considerations when choosing a career path after residency. The majority of resident respondents express the desire to spend a substantial portion of their clinical practice and/or fellowship practicing adult HM. A majority of PHM fellowships can or are willing to explore how to provide both pediatric and adult hospitalist training to med-peds residency–trained fellows. Understanding the facilitators and barriers to recruiting med-peds trainees for PHM fellowship ultimately has significant implications for the future of the PHM workforce. Incorporating the recommendations noted in this study may increase retention of med-peds providers in PHM by enabling fellowship training and ultimately board certification. Collaboration among the ACGME, PHM program directors, and med-peds residency program directors could help to develop PHM fellowship training programs that will meet the needs of med-peds residents interested in practicing PHM while still meeting ACGME requirements for PHM board eligibility.

The authors thank Dr Anoop Agrawal of National Med-Peds Residents’ Association (NMPRA).

The American Board of Medical Specialties approved subspecialty designation for the field of pediatric hospital medicine (PHM) in 2016.¹ For those who started independent practice prior to July 2019, there were two options for board eligibility: the “practice pathway” or completion of a PHM fellowship. The practice pathway allows for pediatric and combined internal medicine–pediatric (med-peds) providers who graduated by July 2019 to sit for the PHM board-certification examination if they meet specific criteria in their pediatric practice.² For pediatric and med-peds residents who graduated after July 2019, PHM board eligibility is available only through completion of a PHM fellowship.

PHM subspecialty designation with fellowship training requirements may pose unique challenges to med-peds residents interested in practicing both pediatric and adult hospital medicine (HM).^3,4 Each year, an estimated 25% of med-peds residency graduates go on to practice HM.⁵ The majority (62%-83%) of currently practicing med-peds–trained hospitalists care for both adults and children.^5,6 Further, med-peds–trained hospitalists comprise at least 10% of the PHM workforce⁵ and play an important role in caring for adult survivors of childhood diseases.³

Limited existing data suggest that the future practice patterns of med-peds residents may be affected by PHM fellowship requirements. One previous survey study indicated that, although med-peds residents see value in additional training opportunities offered by fellowship, the majority are less likely to pursue PHM as a result of the new requirements.⁴ Prominent factors dissuading residents from pursuing PHM fellowship included forfeited earnings during fellowship, student loan obligations, family obligations, and the perception that training received during residency was sufficient. Although these data provide important insights into potential changes in practice patterns, they do not explore qualities of PHM fellowship that may make additional training more appealing to med-peds residents and promote retention of med-peds–trained providers in the PHM workforce.

Further, there is no existing literature exploring if and how PHM fellowship programs are equipped to support the needs of med-peds–trained fellows. Other subspecialties have supported med-peds trainees in combined fellowship training programs, including rheumatology, neurology, pediatric emergency medicine, allergy/immunology, physical medicine and rehabilitation, and psychiatry.^7,8 However, the extent to which PHM fellowships follow a similar model to accommodate the career goals of med-peds participants is unclear.

Given the large numbers of med-peds residents who go on to practice combined PHM and adult HM, it is crucial to understand the training needs of this group within the context of PHM fellowship and board certification. The primary objectives of this study were to understand (1) the perceived PHM fellowship needs of med-peds residents interested in HM, and (2) how the current PHM fellowship training environment can meet those needs. Understanding that additional training requirements to practice PHM may affect the career trajectory of residents interested in HM, secondary objectives included describing perceptions of med-peds residents on PHM specialty designation and whether designation affected their career plans.

Study Design

This cross-sectional study took place over a 3-month period from May to July 2019 and included two surveys of different populations to develop a comprehensive understanding of stakeholder perceptions of PHM fellowship. The first survey (resident survey) invited med-peds residents who were members of the National Med-Peds Residents’ Association (NMPRA)⁹ in 2019 and who were interested in HM. The second survey (fellowship director [FD] survey) included PHM FDs. The study was determined to be exempt by the University of Pittsburgh Institutional Review Board.

Study Population and Recruitment

Resident Survey

Two attempts were made to elicit participation via the NMPRA electronic mailing list. The NMPRA membership includes med-peds residents and chief residents from US med-peds residency programs. As of May 2019, 77 med-peds residency programs and their residents were members of NMPRA, which encompassed all med-peds programs in the United States and its territories. NMPRA maintains a listserv for all members, and all existing US/territory programs were members at the time of the survey. Med-peds interns, residents, and chief residents interested in HM were invited to participate in this study.

FD Survey

Forty-eight FDs, representing member institutions of the PHM Fellowship Directors’ Council, were surveyed via the PHM Fellowship Directors listserv.

Survey Instruments

We constructed two de novo surveys consisting of multiple-choice and short-answer questions (Appendix 1 and Appendix 2). To enhance the validity of survey responses, questions were designed and tested using an iterative consensus process among authors and additional participants, including current med-peds PHM fellows, PHM fellowship program directors, med-peds residency program directors, and current med-peds residents. These revisions were repeated for a total of four cycles. Items were created to increase knowledge on the following key areas: resident-perceived needs in fellowship training, impact of PHM subspecialty designation on career choices related to HM, health system structure of fellowship programs, and ability to accommodate med-peds clinical training within a PHM fellowship. A combined med-peds fellowship, as defined in the survey and referenced in this study, is a “combined internal medicine–pediatrics hospital medicine fellowship whereby you would remain eligible for PHM board certification.” To ensure a broad and inclusive view of potential needs of med-peds trainees considering fellowship, all respondents were asked to complete questions pertaining to anticipated fellowship needs regardless of their indicated interest in fellowship.

Data Collection

Survey completion was voluntary. Email identifiers were not linked to completed surveys. Study data were collected and managed by using Qualtrics XM. Only completed survey entries were included in analysis.

Statistical Methods and Data Analysis

R software version 4.0.2 (R Foundation for Statistical Computing) was used for statistical analysis. Demographic data were summarized using frequency distributions. The intent of the free-text questions for both surveys was qualitative explanatory thematic analysis. Authors EB, HL, and AJ used a deductive approach to identify common themes that elucidated med-peds resident–anticipated needs in fellowship and PHM program strategies and barriers to accommodate these needs. Preliminary themes and action items were reviewed and discussed among the full authorship team until consensus was reached.

Demographic Data

Resident Survey

A total of 466 med-peds residents completed the resident survey. There are approximately 1300 med-peds residents annually, creating an estimated response rate of 35.8% of all US med-peds residents. The majority (n = 380, 81.5%) of respondents were med-peds postgraduate years 1 through 3 and thus only eligible for PHM board certification via the PHM fellowship pathway (Table 1). Most (n = 446, 95.7%) respondents had considered a career in adult, pediatric, or combined HM at some point. Of those med-peds residents who considered a career in HM (Appendix Table 1), 92.8% (n = 414) would prefer to practice combined adult HM and PHM.

FD Survey

Twenty-eight FDs completed the FD survey, representing 58.3% of 2019 PHM fellowship programs. Of the responding programs, 23 (82.1%) were associated with a freestanding children’s hospital, and 24 (85.7%) were integrated or affiliated with a health system that provides adult inpatient care (Table 2). Sixteen (57.1%) programs had a med-peds residency program at their institution.

Med-Peds Resident Perceptions of PHM Fellowship

In considering the importance of PHM board certification for physicians practicing PHM, 59.0% (n= 275) of respondents rated board certification as “not at all important” (Appendix Table 2). Most (n = 420, 90.1%) med-peds trainees responded that PHM subspecialty designation “decreased” or “significantly decreased” their desire to pursue a career that includes PHM. Of the respondents who reported no interest in hospital medicine, eight (40%) reported that PHM subspecialty status dissuaded them from a career in HM at least a moderate amount (Appendix Table 3). Roughly one third (n=158, 33.9%) of respondents reported that PHM subspecialty designation increased or significantly increased their desire to pursue a career that includes adult HM (Appenidx Table 2). Finally, although the majority (n = 275, 59%) of respondents said they had no interest in a HM fellowship, 114 (24.5%) indicated interest in a combined med-peds HM fellowship (Appendix Table 1). Short-answer questions revealed that commitment to additional training on top of a 4-year residency program was a possible deterring factor, particularly in light of student loan debt and family obligations. Respondents reported adequate clinical training during residency as another deterring factor.

Med-Peds Resident–Perceived Needs in PHM Fellowship

Regardless of interest in completing a PHM fellowship, all resident survey respondents were asked how their ideal PHM fellowship should be structured. Almost all (n = 456, 97.9%) respondents indicated that they would prefer to complete a combined med-peds HM fellowship (Table 3), and most preferred to complete a fellowship in 2 years. Only 10 (2.1%) respondents preferred to complete a PHM fellowship alone in 2 or 3 years. More than half (n=253, 54.3%) of respondents indicated that it would be ideal to obtain a master’s degree as part of fellowship.

Three quarters (n = 355, 75.8%) of med-peds residents reported that they would want 41% or more of clinical time in an ideal fellowship dedicated to adult HM. Importantly, most (n = 322, 69.1%) of the med-peds residents did not consider moonlighting alone in either PHM or adult HM to be enough to maintain training. In addition, many (n = 366, 78.5%) respondents felt that it was important or very important for scholarly work during fellowship to bridge pediatrics and internal medicine.

Short-answer questions indicated that the ability to practice both internal medicine and pediatrics during fellowship emerged as an important deciding factor, with emphasis on adequate opportunities to maintain internal medicine knowledge base (Figure). Similarly, access to med-peds mentorship was an important component of the decision. Compensation both during fellowship and potential future earnings was also a prominent consideration.

Capacity of PHM Programs to Support Med-Peds Fellows

Fifteen (53.6%) FDs reported that their programs were able to accommodate both PHM and adult HM clinical time during fellowship, 11 (39.3%) were unsure, and 2 (7.1%) were unable to accommodate both (Table 2).

The options for adult HM clinical time varied by institution and included precepted time on adult HM, full attending privileges on adult HM, and adult HM time through moonlighting only. Short-answer responses from FDs with experience training med-peds fellows cited using PHM elective time for adult HM and offering moonlighting in adult HM as ways to address career goals of med-peds trainees. Scholarship time for fellows was preserved by decreasing required time on pediatric intensive care unit and complex care services.

Accessibility of Med-Peds Mentorship

As noted above, med-peds residents identified mentorship as an important factor in consideration of PHM fellowship. A total of 23 (82.1%) FDs reported their programs had med-peds faculty members within their PHM team (Table 2). The majority (n = 21, 91.3%) of those med-peds faculty had both PHM and adult HM clinical time.

This study characterized the ideal PHM fellowship structure from the perspective of med-peds residents and described the current ability of PHM fellowships to support med-peds residents. The majority of residents stated that they had no interest in an HM fellowship. However, for med-peds residents who considered a career in HM, 88.8% preferred to complete a combined internal medicine and pediatrics HM fellowship with close to half of clinical time dedicated to adult HM. Just over half (53.6%) of programs reported that they could currently accommodate both PHM and adult clinical time during fellowship, and all but two programs reported that they could accommodate both PHM and HM time in the future.

PHM subspecialty designation with associated fellowship training requirements decreased desire to practice HM among med-peds residents who responded to our survey. This reflects findings from a recently published study that evaluated whether PHM fellowship requirements for board certification influenced pediatric and med-peds residents’ decision to pursue PHM in 2018.⁴ Additionally, Chandrasekar et al⁴ found that 87% of respondents indicated that sufficient residency training was an important factor in discouraging them from pursuing PHM fellowship. We noted similar findings in our open-ended survey responses, which indicate that med-peds respondents perceived that the intended purpose of PHM fellowship was to provide additional clinical training, and that served as a deterrent for fellowship. However, the survey by Chandrasekar et al⁴ assessed only four factors for understanding what was important in encouraging pursuit of a PHM fellowship: opportunity to gain new skills, potential increase in salary, opportunity for a master’s degree, and increased prestige. Our survey expands on med-peds residents’ needs, indicating that med-peds residents want a combined med-peds/HM fellowship that allows them to meet PHM board-eligibility requirements while also continuing to develop their adult HM clinical practice and other nonclinical training objectives in a way that combines both adult HM and PHM. Both surveys demonstrate the role that residency program directors and other resident mentors can have in counseling trainees on the nonclinical training objectives of PHM fellowship, including research, quality improvement, medical education, and leadership and clinical operations. Additional emphasis can be placed on opportunities for an individualized curriculum to address the specific career aims of each resident.

In this study, med-peds trainees viewed distribution of clinical time during fellowship as an important factor in pursuing PHM fellowship. The perceived importance of balancing clinical time is not surprising considering that most survey respondents interested in HM ultimately intend to practice both PHM and adult HM. This finding corresponds with current practice patterns of med-peds hospitalists, the majority of whom care for both children and adults.^4,5 Moonlighting in adult medicine was not considered sufficient, suggesting desire for mentorship and training integration on the internal medicine side. Opportunities for trainees to maintain and expand their internal medicine knowledge base and clinical decision-making outside of moonlighting will be key to meeting the needs of med-peds residents in PHM fellowship.

Fortunately, more than half of responding programs reported that they could allow for adult HM practice during PHM fellowship. Twelve programs were unsure if they could accommodate adult HM clinical time, and only two programs reported they could not. We suspect that the ability to support this training with clinical time in both adult HM and PHM is more likely available at programs with established internal medicine relationships, often in the form of med-peds residency programs and med-peds faculty. Further, these established relationships may be more common at pediatric health systems that are integrated or affiliated with an adult health system. Most PHM fellowships surveyed indicated that their pediatric institution had an affiliation with an adult facility, and most had med-peds HM faculty.

Precedent for supporting med-peds fellows is somewhat limited given that only five of the responding PHM fellowship programs reported having fellows with med-peds residency training. However, discrepancies between the expressed needs of med-peds residents and the current Accreditation Council for Graduate Medical Education (ACGME)–accredited PHM fellowship structure highlight opportunities to tailor fellowship training to support the career goals of med-peds residents. The current PHM fellowship structure consists of 26 educational units, with each unit representing 4 calendar weeks. A minimum of eight units are spent on each of the following: core clinical rotations, systems and scholarship, and individualized curriculum.^10,11 The Society of Hospital Medicine has published core competencies for both PHM and adult HM, which highlight significant overlap in each field’s skill competency, particularly in areas such as quality improvement, legal issues and risk management, and handoffs and transitions of care.^12,13 We contend that competencies addressed within PHM fellowship core clinical rotations may overlap with adult HM. Training in adult HM could be completed as part of the individualized curriculum with the ACGME, allowing adult HM practice to count toward this requirement. This would offer med-peds fellows the option to maintain their adult HM knowledge base without eliminating all elective time. Ultimately, it will be important to be creative in how training is accomplished and skills are acquired during both core clinical and individualized training blocks for med-peds trainees completing PHM fellowship.

In order to meet the expressed needs of med-peds residents interested in incorporating both adult HM and PHM into their future careers through PHM fellowship, we offer key recommendations for consideration by the ACGME, PHM FDs, and med-peds program directors (Figure). We encourage current PHM fellowship programs to establish relationships with adult HM programs to develop structured clinical opportunities that will allow fellows to gain the additional clinical training desired.

There were important limitations in this study. First, our estimated response rate for the resident survey was 35.8% of all med-peds residents in 2019, which may be interpreted as low. However, it is important to note that the survey was targeted to residents interested in HM. More than 25% of med-peds residents pursue a career in HM,⁵ suggesting our response rate may be attributed to residents who did not complete the survey because they were interested in other fields. The program director survey response rate was higher at 58.3%, though it is possible that response bias resulted in a higher response rate from programs with the ability to support med-peds trainees. Regardless, data from programs with the ability to support med-peds trainees are highly valuable in describing how PHM fellowship can be inclusive of med-peds–trained physicians interested in pursuing HM.

Both surveys were completed in 2019, prior to the ACGME accreditation of PHM fellowship, which likely presents new, unique challenges to fellowship programs trying to support the needs of med-peds fellows. However, insights noted above from programs with experience training med-peds fellows are still applicable within the constraints of ACGME requirements.

Many med-peds residents express strong interest in practicing HM and including PHM as part of their future hospitalist practice. With the introduction of PHM subspecialty board certification through the American Board of Pediatrics, med-peds residents face new considerations when choosing a career path after residency. The majority of resident respondents express the desire to spend a substantial portion of their clinical practice and/or fellowship practicing adult HM. A majority of PHM fellowships can or are willing to explore how to provide both pediatric and adult hospitalist training to med-peds residency–trained fellows. Understanding the facilitators and barriers to recruiting med-peds trainees for PHM fellowship ultimately has significant implications for the future of the PHM workforce. Incorporating the recommendations noted in this study may increase retention of med-peds providers in PHM by enabling fellowship training and ultimately board certification. Collaboration among the ACGME, PHM program directors, and med-peds residency program directors could help to develop PHM fellowship training programs that will meet the needs of med-peds residents interested in practicing PHM while still meeting ACGME requirements for PHM board eligibility.

The authors thank Dr Anoop Agrawal of National Med-Peds Residents’ Association (NMPRA).

Mobile applications (apps) have been a growing part of medicine for the last decade. In 2020, more than 15.5 million apps were available for download,¹ and more than 325,000 apps were health related.² Much of the peer-reviewed literature on health-related apps has focused on apps that target patients. Therefore, we studied apps for health care providers, specifically dermatology residents of different sexes throughout residency. We investigated the role of apps in their training, including how often residents consult apps, which apps they utilize, and why.

Methods

An original online survey regarding mobile apps was emailed to all 1587 dermatology residents in America by the American Academy of Dermatology from summer 2019 to summer 2020. Responses were anonymous, voluntary, unincentivized, and collected over 17 days. To protect respondent privacy, minimal data were collected regarding training programs; geography served as a proxy for how resource rich or resource poor those programs may be. Categorization of urban vs rural was based on the 2010 Census classification, such that Arizona; California; Colorado; Connecticut; Florida; Illinois; Maryland; Massachusetts; New Jersey; New York; Oregon; Puerto Rico; Rhode Island; Texas; Utah; and Washington, DC, were urban, and the remaining states were rural.³

We hypothesized that VisualDx would be 1 of 3 most prevalent apps; “diagnosis and workup” and “self-education” would be top reasons for using apps; “up-to-date and accurate information” would be a top 3 consideration when choosing apps; the most consulted resources for clinical experiences would be providers, followed by websites, apps, and lastly printed text; and the percentage of clinical experiences for which a provider was consulted would be higher for first-year residents than other years and for female residents than male residents.

Fisher exact 2-tailed and Kruskal-Wallis (KW) pairwise tests were used to compare groups. Statistical significance was set at P<.05.

Results

Respondents
The response rate was 16.6% (n=263), which is similar to prior response rates for American Academy of Dermatology surveys. Table 1 contains respondent demographics. The mean age of respondents was 31 years. Sixty percent of respondents were female; 62% of respondents were training in urban states or territories. Regarding the dermatology residency year, 34% of respondents were in their first year, 32% were in their second, and 34% were in their third. Eighty-seven percent of respondents used Apple iOS. Every respondent used at least 1 dermatology-related app (mean, 5; range, 1–11)(Table 2).

Top Dermatology-Related Apps
The 10 most prevalent apps are listed in Table 2. The 3 most prevalent apps were VisualDx (84%, majority of respondents used daily), UpToDate (67%, majority of respondents used daily), and Mohs Surgery Appropriate Use Criteria (63%, majority of respondents used weekly). A higher percentage of third-year residents used GoodRx compared to first- and second-year residents (Fisher exact test: P=.014 and P=.041, respectively). A lower percentage of female respondents used GoodRx compared to male residents (Fisher exact test: P=.003). None of the apps were app versions of printed text, including textbooks or journals.

Reasons for Using Apps
The 10 primary reasons for using apps are listed in Table 2. The top 3 reasons were diagnosis and workup (83%), medication dosage (72%), and self-education (69%). Medication dosage and saving time were both selected by a higher percentage of third-year residents than first-year residents (Fisher exact test: P=.041 and P=.024, respectively). Self-education was selected by a lower percentage of third-year residents than second-year residents (Fisher exact test: P=.025). 

Considerations When Choosing Apps
The 10 primary considerations when choosing apps are listed in Table 2. The top 3 considerations were up-to-date and accurate information (81%), no/low cost (80%), and user-friendly design (74%). Up-to-date and accurate information was selected by a lower percentage of third-year residents than first- and second-year residents (Fisher exact test: P=.02 and P=.03, respectively).

Consulted Resources
Apps were the second most consulted resource (26%) during clinical work, behind human guidance (73%). Female respondents consulted both resources more than male respondents (KW: P≤.005 and P≤.003, respectively). First-year residents consulted humans more than second-year and third-year residents (KW: P<.0001).

There were no significant differences by geography or mobile operating system.

Comment

The response rate and demographic results suggest that our study sample is representative of the target population of dermatology residents in America. Overall, the survey results support our hypotheses.

A survey conducted in 2008 before apps were readily available found that dermatology residents felt they learned more successfully when engaging in hands-on, direct experience; talking with experts/consultants; and studying printed materials than when using multimedia programs.⁴ Our study suggests that the usage of and preference for multimedia programs, including apps, in dermatology resident training has risen substantially, despite the continued availability of guidance from attendings and senior residents.

As residents progress through training, they increasingly turn to virtual resources. According to our survey, junior residents are more likely than third-year residents to use apps for self-education, and up-to-date and accurate information was a more important consideration when choosing apps. Third-year residents are more likely than junior residents to use apps for medication dosage and saving time. Perhaps related, GoodRx, an app that provides prescription discounts, was more prevalent among third-year residents. It is notable that most of the reported apps, including those used for diagnosis and treatment, did not need premarket government approval to ensure patient safety, are not required to contain up-to-date information, and do not reference primary sources. Additionally, only UpToDate has been shown in peer-reviewed literature to improve clinical outcomes.⁵

Our survey also revealed a few differences by sex. Female respondents consulted resources during clinical work more often than male residents. This finding is similar to the limited existing research on dermatologists’ utilization of information showing higher dermoscopy use among female attendings.⁶ Use of GoodRx was less prevalent among female vs male respondents. Perhaps related, a 2011 study found that female primary care physicians are less likely to prescribe medications than their male counterparts.⁷

Our study had several limitations. There may have been selection bias such that the residents who chose to participate were relatively more interested in mobile health. Certain demographic data, such as race, were not captured because prior studies do not suggest disparity by those demographics for mobile health utilization among residents, but those data could be incorporated into future studies. Our survey was intentionally limited in scope. For example, it did not capture the amount of time spent on each consult resource or the motivations for consulting an app instead of a provider.

Conclusion

A main objective of residency is to train new physicians to provide excellent patient care. Our survey highlights the increasing role of apps in dermatology residency, different priorities among years of residency, and different information utilization between sexes. This knowledge should encourage and help guide standardization and quality assurance of virtual residency education and integration of virtual resources into formal curricula. Residency administrators and residents should be aware of the apps used to learn and deliver care, consider the evidence for and regulation of those apps, and evaluate the accessibility and approachability of attendings to residents. Future research should examine the educational and clinical outcomes of app utilization among residents and the impact of residency programs’ unspoken cultures and expectations on relationships among residents of different demographics and their attendings.

Mobile applications (apps) have been a growing part of medicine for the last decade. In 2020, more than 15.5 million apps were available for download,¹ and more than 325,000 apps were health related.² Much of the peer-reviewed literature on health-related apps has focused on apps that target patients. Therefore, we studied apps for health care providers, specifically dermatology residents of different sexes throughout residency. We investigated the role of apps in their training, including how often residents consult apps, which apps they utilize, and why.

Methods

An original online survey regarding mobile apps was emailed to all 1587 dermatology residents in America by the American Academy of Dermatology from summer 2019 to summer 2020. Responses were anonymous, voluntary, unincentivized, and collected over 17 days. To protect respondent privacy, minimal data were collected regarding training programs; geography served as a proxy for how resource rich or resource poor those programs may be. Categorization of urban vs rural was based on the 2010 Census classification, such that Arizona; California; Colorado; Connecticut; Florida; Illinois; Maryland; Massachusetts; New Jersey; New York; Oregon; Puerto Rico; Rhode Island; Texas; Utah; and Washington, DC, were urban, and the remaining states were rural.³

We hypothesized that VisualDx would be 1 of 3 most prevalent apps; “diagnosis and workup” and “self-education” would be top reasons for using apps; “up-to-date and accurate information” would be a top 3 consideration when choosing apps; the most consulted resources for clinical experiences would be providers, followed by websites, apps, and lastly printed text; and the percentage of clinical experiences for which a provider was consulted would be higher for first-year residents than other years and for female residents than male residents.

Fisher exact 2-tailed and Kruskal-Wallis (KW) pairwise tests were used to compare groups. Statistical significance was set at P<.05.

Results

Respondents
The response rate was 16.6% (n=263), which is similar to prior response rates for American Academy of Dermatology surveys. Table 1 contains respondent demographics. The mean age of respondents was 31 years. Sixty percent of respondents were female; 62% of respondents were training in urban states or territories. Regarding the dermatology residency year, 34% of respondents were in their first year, 32% were in their second, and 34% were in their third. Eighty-seven percent of respondents used Apple iOS. Every respondent used at least 1 dermatology-related app (mean, 5; range, 1–11)(Table 2).

Top Dermatology-Related Apps
The 10 most prevalent apps are listed in Table 2. The 3 most prevalent apps were VisualDx (84%, majority of respondents used daily), UpToDate (67%, majority of respondents used daily), and Mohs Surgery Appropriate Use Criteria (63%, majority of respondents used weekly). A higher percentage of third-year residents used GoodRx compared to first- and second-year residents (Fisher exact test: P=.014 and P=.041, respectively). A lower percentage of female respondents used GoodRx compared to male residents (Fisher exact test: P=.003). None of the apps were app versions of printed text, including textbooks or journals.

Reasons for Using Apps
The 10 primary reasons for using apps are listed in Table 2. The top 3 reasons were diagnosis and workup (83%), medication dosage (72%), and self-education (69%). Medication dosage and saving time were both selected by a higher percentage of third-year residents than first-year residents (Fisher exact test: P=.041 and P=.024, respectively). Self-education was selected by a lower percentage of third-year residents than second-year residents (Fisher exact test: P=.025). 

Considerations When Choosing Apps
The 10 primary considerations when choosing apps are listed in Table 2. The top 3 considerations were up-to-date and accurate information (81%), no/low cost (80%), and user-friendly design (74%). Up-to-date and accurate information was selected by a lower percentage of third-year residents than first- and second-year residents (Fisher exact test: P=.02 and P=.03, respectively).

Consulted Resources
Apps were the second most consulted resource (26%) during clinical work, behind human guidance (73%). Female respondents consulted both resources more than male respondents (KW: P≤.005 and P≤.003, respectively). First-year residents consulted humans more than second-year and third-year residents (KW: P<.0001).

There were no significant differences by geography or mobile operating system.

Comment

The response rate and demographic results suggest that our study sample is representative of the target population of dermatology residents in America. Overall, the survey results support our hypotheses.

A survey conducted in 2008 before apps were readily available found that dermatology residents felt they learned more successfully when engaging in hands-on, direct experience; talking with experts/consultants; and studying printed materials than when using multimedia programs.⁴ Our study suggests that the usage of and preference for multimedia programs, including apps, in dermatology resident training has risen substantially, despite the continued availability of guidance from attendings and senior residents.

As residents progress through training, they increasingly turn to virtual resources. According to our survey, junior residents are more likely than third-year residents to use apps for self-education, and up-to-date and accurate information was a more important consideration when choosing apps. Third-year residents are more likely than junior residents to use apps for medication dosage and saving time. Perhaps related, GoodRx, an app that provides prescription discounts, was more prevalent among third-year residents. It is notable that most of the reported apps, including those used for diagnosis and treatment, did not need premarket government approval to ensure patient safety, are not required to contain up-to-date information, and do not reference primary sources. Additionally, only UpToDate has been shown in peer-reviewed literature to improve clinical outcomes.⁵

Our survey also revealed a few differences by sex. Female respondents consulted resources during clinical work more often than male residents. This finding is similar to the limited existing research on dermatologists’ utilization of information showing higher dermoscopy use among female attendings.⁶ Use of GoodRx was less prevalent among female vs male respondents. Perhaps related, a 2011 study found that female primary care physicians are less likely to prescribe medications than their male counterparts.⁷

Our study had several limitations. There may have been selection bias such that the residents who chose to participate were relatively more interested in mobile health. Certain demographic data, such as race, were not captured because prior studies do not suggest disparity by those demographics for mobile health utilization among residents, but those data could be incorporated into future studies. Our survey was intentionally limited in scope. For example, it did not capture the amount of time spent on each consult resource or the motivations for consulting an app instead of a provider.

Conclusion

A main objective of residency is to train new physicians to provide excellent patient care. Our survey highlights the increasing role of apps in dermatology residency, different priorities among years of residency, and different information utilization between sexes. This knowledge should encourage and help guide standardization and quality assurance of virtual residency education and integration of virtual resources into formal curricula. Residency administrators and residents should be aware of the apps used to learn and deliver care, consider the evidence for and regulation of those apps, and evaluate the accessibility and approachability of attendings to residents. Future research should examine the educational and clinical outcomes of app utilization among residents and the impact of residency programs’ unspoken cultures and expectations on relationships among residents of different demographics and their attendings.

Mobile applications (apps) have been a growing part of medicine for the last decade. In 2020, more than 15.5 million apps were available for download,¹ and more than 325,000 apps were health related.² Much of the peer-reviewed literature on health-related apps has focused on apps that target patients. Therefore, we studied apps for health care providers, specifically dermatology residents of different sexes throughout residency. We investigated the role of apps in their training, including how often residents consult apps, which apps they utilize, and why.

Methods

An original online survey regarding mobile apps was emailed to all 1587 dermatology residents in America by the American Academy of Dermatology from summer 2019 to summer 2020. Responses were anonymous, voluntary, unincentivized, and collected over 17 days. To protect respondent privacy, minimal data were collected regarding training programs; geography served as a proxy for how resource rich or resource poor those programs may be. Categorization of urban vs rural was based on the 2010 Census classification, such that Arizona; California; Colorado; Connecticut; Florida; Illinois; Maryland; Massachusetts; New Jersey; New York; Oregon; Puerto Rico; Rhode Island; Texas; Utah; and Washington, DC, were urban, and the remaining states were rural.³

We hypothesized that VisualDx would be 1 of 3 most prevalent apps; “diagnosis and workup” and “self-education” would be top reasons for using apps; “up-to-date and accurate information” would be a top 3 consideration when choosing apps; the most consulted resources for clinical experiences would be providers, followed by websites, apps, and lastly printed text; and the percentage of clinical experiences for which a provider was consulted would be higher for first-year residents than other years and for female residents than male residents.

Fisher exact 2-tailed and Kruskal-Wallis (KW) pairwise tests were used to compare groups. Statistical significance was set at P<.05.

Results

Respondents
The response rate was 16.6% (n=263), which is similar to prior response rates for American Academy of Dermatology surveys. Table 1 contains respondent demographics. The mean age of respondents was 31 years. Sixty percent of respondents were female; 62% of respondents were training in urban states or territories. Regarding the dermatology residency year, 34% of respondents were in their first year, 32% were in their second, and 34% were in their third. Eighty-seven percent of respondents used Apple iOS. Every respondent used at least 1 dermatology-related app (mean, 5; range, 1–11)(Table 2).

Top Dermatology-Related Apps
The 10 most prevalent apps are listed in Table 2. The 3 most prevalent apps were VisualDx (84%, majority of respondents used daily), UpToDate (67%, majority of respondents used daily), and Mohs Surgery Appropriate Use Criteria (63%, majority of respondents used weekly). A higher percentage of third-year residents used GoodRx compared to first- and second-year residents (Fisher exact test: P=.014 and P=.041, respectively). A lower percentage of female respondents used GoodRx compared to male residents (Fisher exact test: P=.003). None of the apps were app versions of printed text, including textbooks or journals.

Reasons for Using Apps
The 10 primary reasons for using apps are listed in Table 2. The top 3 reasons were diagnosis and workup (83%), medication dosage (72%), and self-education (69%). Medication dosage and saving time were both selected by a higher percentage of third-year residents than first-year residents (Fisher exact test: P=.041 and P=.024, respectively). Self-education was selected by a lower percentage of third-year residents than second-year residents (Fisher exact test: P=.025). 

Considerations When Choosing Apps
The 10 primary considerations when choosing apps are listed in Table 2. The top 3 considerations were up-to-date and accurate information (81%), no/low cost (80%), and user-friendly design (74%). Up-to-date and accurate information was selected by a lower percentage of third-year residents than first- and second-year residents (Fisher exact test: P=.02 and P=.03, respectively).

Consulted Resources
Apps were the second most consulted resource (26%) during clinical work, behind human guidance (73%). Female respondents consulted both resources more than male respondents (KW: P≤.005 and P≤.003, respectively). First-year residents consulted humans more than second-year and third-year residents (KW: P<.0001).

There were no significant differences by geography or mobile operating system.

Comment

The response rate and demographic results suggest that our study sample is representative of the target population of dermatology residents in America. Overall, the survey results support our hypotheses.

A survey conducted in 2008 before apps were readily available found that dermatology residents felt they learned more successfully when engaging in hands-on, direct experience; talking with experts/consultants; and studying printed materials than when using multimedia programs.⁴ Our study suggests that the usage of and preference for multimedia programs, including apps, in dermatology resident training has risen substantially, despite the continued availability of guidance from attendings and senior residents.

As residents progress through training, they increasingly turn to virtual resources. According to our survey, junior residents are more likely than third-year residents to use apps for self-education, and up-to-date and accurate information was a more important consideration when choosing apps. Third-year residents are more likely than junior residents to use apps for medication dosage and saving time. Perhaps related, GoodRx, an app that provides prescription discounts, was more prevalent among third-year residents. It is notable that most of the reported apps, including those used for diagnosis and treatment, did not need premarket government approval to ensure patient safety, are not required to contain up-to-date information, and do not reference primary sources. Additionally, only UpToDate has been shown in peer-reviewed literature to improve clinical outcomes.⁵

Our survey also revealed a few differences by sex. Female respondents consulted resources during clinical work more often than male residents. This finding is similar to the limited existing research on dermatologists’ utilization of information showing higher dermoscopy use among female attendings.⁶ Use of GoodRx was less prevalent among female vs male respondents. Perhaps related, a 2011 study found that female primary care physicians are less likely to prescribe medications than their male counterparts.⁷

Our study had several limitations. There may have been selection bias such that the residents who chose to participate were relatively more interested in mobile health. Certain demographic data, such as race, were not captured because prior studies do not suggest disparity by those demographics for mobile health utilization among residents, but those data could be incorporated into future studies. Our survey was intentionally limited in scope. For example, it did not capture the amount of time spent on each consult resource or the motivations for consulting an app instead of a provider.

Conclusion

A main objective of residency is to train new physicians to provide excellent patient care. Our survey highlights the increasing role of apps in dermatology residency, different priorities among years of residency, and different information utilization between sexes. This knowledge should encourage and help guide standardization and quality assurance of virtual residency education and integration of virtual resources into formal curricula. Residency administrators and residents should be aware of the apps used to learn and deliver care, consider the evidence for and regulation of those apps, and evaluate the accessibility and approachability of attendings to residents. Future research should examine the educational and clinical outcomes of app utilization among residents and the impact of residency programs’ unspoken cultures and expectations on relationships among residents of different demographics and their attendings.

To the Editor:

Nail psoriasis is highly prevalent in patients with cutaneous psoriasis and also may present as an isolated finding. There is a strong association between nail psoriasis and development of psoriatic arthritis (PsA). However, publications on nail psoriasis are sparse compared with articles describing cutaneous psoriasis.¹ Our objectives were to analyze the nail psoriasis literature for content, citations, and media attention.

The Web of Science database was searched for the term nail psoriasis on April 27, 2020, and publications by year, subject, and article type were compiled. Total and average yearly citations were calculated to create a list of the top 100 most-cited articles (eTable). First and last authors, sex, and Altmetric Attention Scores were then recorded. The Wilcoxon rank sum test was calculated to compare the relationship of Altmetric scores between nail psoriasis–specific references and others on the list.

In our data set, the average total number of citations was 134.09 (range, 42–1617), with average yearly citations ranging from 2 to 108. Altmetric scores—measures of media attention of scholarly work—were available for 58 of 100 papers (58%), with an average score of 33.2 (range, 1–509).

Of the top 100 most-cited articles using the search term nail psoriasis, only 20% focused on nail psoriasis, with the remainder concentrating on psoriasis/PsA. Only 32% and 24% of first and last authors, respectively, were female. Fifty-two percent and 31% of the articles were published in dermatology and arthritis/rheumatology journals, respectively. There was no statistically significant difference in Altmetric scores between nail psoriasis–specific and other articles in our data set (P=.7551).

For the nail psoriasis–specific articles, all 20 highlighted a lack of nail clinical trials, a positive association with PsA, and a correlation of increased cutaneous psoriasis body surface area with increased onychodystrophy likelihood.² Three of 20 (15%) articles stated that nail psoriasis often is overlooked, despite the negative impact on quality of life,¹ and emphasized the importance of patient compliance owing to the chronic nature of the disease. Only 1 of 20 (5%) articles focused on nail psoriasis treatments.³ There was no overlap between the 100 most-cited psoriasis articles from 1970 to 2012 and our top 100 articles on nail psoriasis.⁴

Treatment recommendations for nail psoriasis by consensus were published by a nail expert group in 2019.⁵ For 3 or fewer nails involved, suggested first-line treatment is intralesional matrix injections with triamcinolone acetonide. For more than 3 affected nails, systemic treatment with oral or biologic therapy is recommended.⁵ Although this article is likely to change clinical practice, it did not qualify for our list because it did not garner sufficient citations in the brief period between its publication date and our search (July 2019–April 2020).

This study is subject to several limitations. Only the Web of Science database was utilized, and only the term nail psoriasis was searched, potentially excluding relevant articles. Using total citations biases toward older articles.

Our bibliometric analysis highlights a lack of publications on nail psoriasis, with most articles focusing on psoriasis and PsA. This deficiency in highly cited nail psoriasis references is likely to be a barrier to physicians in managing patients with nail disease. There is a need for controlled clinical trials and better mechanisms to disseminate information on management of nail psoriasis to practicing physicians.

The eTable is available in the PDF of this article

To the Editor:

Nail psoriasis is highly prevalent in patients with cutaneous psoriasis and also may present as an isolated finding. There is a strong association between nail psoriasis and development of psoriatic arthritis (PsA). However, publications on nail psoriasis are sparse compared with articles describing cutaneous psoriasis.¹ Our objectives were to analyze the nail psoriasis literature for content, citations, and media attention.

The Web of Science database was searched for the term nail psoriasis on April 27, 2020, and publications by year, subject, and article type were compiled. Total and average yearly citations were calculated to create a list of the top 100 most-cited articles (eTable). First and last authors, sex, and Altmetric Attention Scores were then recorded. The Wilcoxon rank sum test was calculated to compare the relationship of Altmetric scores between nail psoriasis–specific references and others on the list.

In our data set, the average total number of citations was 134.09 (range, 42–1617), with average yearly citations ranging from 2 to 108. Altmetric scores—measures of media attention of scholarly work—were available for 58 of 100 papers (58%), with an average score of 33.2 (range, 1–509).

Of the top 100 most-cited articles using the search term nail psoriasis, only 20% focused on nail psoriasis, with the remainder concentrating on psoriasis/PsA. Only 32% and 24% of first and last authors, respectively, were female. Fifty-two percent and 31% of the articles were published in dermatology and arthritis/rheumatology journals, respectively. There was no statistically significant difference in Altmetric scores between nail psoriasis–specific and other articles in our data set (P=.7551).

For the nail psoriasis–specific articles, all 20 highlighted a lack of nail clinical trials, a positive association with PsA, and a correlation of increased cutaneous psoriasis body surface area with increased onychodystrophy likelihood.² Three of 20 (15%) articles stated that nail psoriasis often is overlooked, despite the negative impact on quality of life,¹ and emphasized the importance of patient compliance owing to the chronic nature of the disease. Only 1 of 20 (5%) articles focused on nail psoriasis treatments.³ There was no overlap between the 100 most-cited psoriasis articles from 1970 to 2012 and our top 100 articles on nail psoriasis.⁴

Treatment recommendations for nail psoriasis by consensus were published by a nail expert group in 2019.⁵ For 3 or fewer nails involved, suggested first-line treatment is intralesional matrix injections with triamcinolone acetonide. For more than 3 affected nails, systemic treatment with oral or biologic therapy is recommended.⁵ Although this article is likely to change clinical practice, it did not qualify for our list because it did not garner sufficient citations in the brief period between its publication date and our search (July 2019–April 2020).

This study is subject to several limitations. Only the Web of Science database was utilized, and only the term nail psoriasis was searched, potentially excluding relevant articles. Using total citations biases toward older articles.

Our bibliometric analysis highlights a lack of publications on nail psoriasis, with most articles focusing on psoriasis and PsA. This deficiency in highly cited nail psoriasis references is likely to be a barrier to physicians in managing patients with nail disease. There is a need for controlled clinical trials and better mechanisms to disseminate information on management of nail psoriasis to practicing physicians.

The eTable is available in the PDF of this article

To the Editor:

Nail psoriasis is highly prevalent in patients with cutaneous psoriasis and also may present as an isolated finding. There is a strong association between nail psoriasis and development of psoriatic arthritis (PsA). However, publications on nail psoriasis are sparse compared with articles describing cutaneous psoriasis.¹ Our objectives were to analyze the nail psoriasis literature for content, citations, and media attention.

The Web of Science database was searched for the term nail psoriasis on April 27, 2020, and publications by year, subject, and article type were compiled. Total and average yearly citations were calculated to create a list of the top 100 most-cited articles (eTable). First and last authors, sex, and Altmetric Attention Scores were then recorded. The Wilcoxon rank sum test was calculated to compare the relationship of Altmetric scores between nail psoriasis–specific references and others on the list.

In our data set, the average total number of citations was 134.09 (range, 42–1617), with average yearly citations ranging from 2 to 108. Altmetric scores—measures of media attention of scholarly work—were available for 58 of 100 papers (58%), with an average score of 33.2 (range, 1–509).

Of the top 100 most-cited articles using the search term nail psoriasis, only 20% focused on nail psoriasis, with the remainder concentrating on psoriasis/PsA. Only 32% and 24% of first and last authors, respectively, were female. Fifty-two percent and 31% of the articles were published in dermatology and arthritis/rheumatology journals, respectively. There was no statistically significant difference in Altmetric scores between nail psoriasis–specific and other articles in our data set (P=.7551).

For the nail psoriasis–specific articles, all 20 highlighted a lack of nail clinical trials, a positive association with PsA, and a correlation of increased cutaneous psoriasis body surface area with increased onychodystrophy likelihood.² Three of 20 (15%) articles stated that nail psoriasis often is overlooked, despite the negative impact on quality of life,¹ and emphasized the importance of patient compliance owing to the chronic nature of the disease. Only 1 of 20 (5%) articles focused on nail psoriasis treatments.³ There was no overlap between the 100 most-cited psoriasis articles from 1970 to 2012 and our top 100 articles on nail psoriasis.⁴

Treatment recommendations for nail psoriasis by consensus were published by a nail expert group in 2019.⁵ For 3 or fewer nails involved, suggested first-line treatment is intralesional matrix injections with triamcinolone acetonide. For more than 3 affected nails, systemic treatment with oral or biologic therapy is recommended.⁵ Although this article is likely to change clinical practice, it did not qualify for our list because it did not garner sufficient citations in the brief period between its publication date and our search (July 2019–April 2020).

This study is subject to several limitations. Only the Web of Science database was utilized, and only the term nail psoriasis was searched, potentially excluding relevant articles. Using total citations biases toward older articles.

Our bibliometric analysis highlights a lack of publications on nail psoriasis, with most articles focusing on psoriasis and PsA. This deficiency in highly cited nail psoriasis references is likely to be a barrier to physicians in managing patients with nail disease. There is a need for controlled clinical trials and better mechanisms to disseminate information on management of nail psoriasis to practicing physicians.

The eTable is available in the PDF of this article

Irritable bowel syndrome (IBS) is one of the most common disorders managed by primary care physicians and gastroenterologists.¹ Characterized by abdominal pain coinciding with altered stool form and/or frequency as defined by the Rome IV diagnostic criteria,² symptoms range from mild to debilitating and may remarkably impair quality of life and work productivity.¹

The cause of IBS is poorly understood. Proposed pathophysiologic factors include impaired mucosal function, microbial imbalance, visceral hypersensitivity, psychologic dysfunction, genetic factors, neurotransmitter imbalance, postinfectious gastroenteritis, inflammation, and food intolerance, any or all of which may lead to the development and maintenance of IBS symptoms.³ More recent observations of inflammation in the intestinal lining^4,5 and proinflammatory peripherally circulating cytokines⁶ challenge its traditional classification as a functional disorder.

The cause of this inflammation is of intense interest, with speculation that the bacterial microbiota, bile acids, association with postinfectious gastroenteritis and inflammatory bowel disease cases, and/or foods may contribute. Although approximately 50% of individuals with IBS report that foods aggravate their symptoms,⁷ studies investigating type I antibody–mediated immediate hypersensitivity have largely failed to demonstrate a substantial link, prompting many authorities to regard these associations as food “intolerances” rather than true allergies. Based on this body of literature, a large 2010 consensus report on all aspects of food allergies advises against food allergy testing for IBS.⁸

In contrast, by utilizing type IV food allergen skin patch testing, 2 proof-of-concept studies^9,10 investigated a different allergic mechanism in IBS, namely cell-mediated delayed-type hypersensitivity. Because many foods and food additives are known to cause allergic contact dermatitis,¹¹ it was hypothesized that these foods may elicit a similar delayed-type hypersensitivity response in the intestinal lining in previously sensitized individuals. By following a patch test–guided food avoidance diet, a large subpopulation of patients with IBS experienced partial or complete IBS symptom relief.^9,10 Our study further investigates a role for food-related delayed-type hypersensitivities in the pathogenesis of IBS.

Methods

Patient Selection
This study was conducted in a secondary care community-based setting. All patients were self-referred over an 18-month period ending in October 2019, had physician-diagnosed IBS, and/or met the Rome IV criteria for IBS and presented expressly for the food patch testing on a fee-for-service basis. Subtype of IBS was determined on presentation by the self-reported historically predominant symptom. Duration of IBS symptoms was self-reported and was rounded to the nearest year for purposes of data collection.

Exclusion criteria included pregnancy, known allergy to adhesive tape or any of the food allergens used in the study, severe skin rash, symptoms that had a known cause other than IBS, or active treatment with systemic immunosuppressive medications.

Patch Testing
Skin patch testing was initiated using an extensive panel of 117 type IV food allergens (eTable)¹¹ identified in the literature,¹² most of which utilized standard compounded formulations¹³ or were available from reputable patch test manufacturers (Brial Allergen GmbH; Chemotechnique Diagnostics). This panel was not approved by the US Food and Drug Administration. The freeze-dried vegetable formulations were taken from the 2018 report.⁹ Standard skin patch test procedure protocols¹² were used, affixing the patches to the upper aspect of the back.

Following patch test application on day 1, two follow-up visits occurred on day 3 and either day 4 or day 5. On day 3, patches were removed, and the initial results were read by a board-certified dermatologist according to a standard grading system.¹⁴ Interpretation of patch tests included no reaction, questionable reaction consisting of macular erythema, weak reaction consisting of erythema and slight edema, or strong reaction consisting of erythema and marked edema. On day 4 or day 5, the final patch test reading was performed, and patients were informed of their results. Patients were advised to avoid ingestion of all foods that elicited a questionable or positive patch test response for at least 3 months, and information about the foods and their avoidance also was distributed and reviewed.

Food Avoidance Questionnaire
Patients with questionable or positive patch tests at 72 or 96 hours were advised of their eligibility to participate in an institutional review board–approved food avoidance questionnaire study investigating the utility of patch test–guided food avoidance on IBS symptoms. The questionnaire assessed the following: (1) baseline average abdominal pain prior to patch test–guided avoidance diet (0=no symptoms; 10=very severe); (2) average abdominal pain since initiation of patch test–guided avoidance diet (0=no symptoms; 10=very severe); (3) degree of improvement in overall IBS symptoms by the end of the food avoidance period (0=no improvement; 10=great improvement); (4) compliance with the avoidance diet for the duration of the avoidance period (completely, partially, not at all, or not sure).

Questionnaires and informed consent were mailed to patients via the US Postal Service 3 months after completing the patch testing. The questionnaire and consent were to be completed and returned after dietary avoidance of the identified allergens for at least 3 months. Patients were not compensated for participation in the study.

Statistical Analysis
Statistical analysis of data collected from study questionnaires was performed with Microsoft Excel. Mean abdominal pain and mean global improvement scores were reported along with 1 SD of the mean. For comparison of mean abdominal pain and improvement in global IBS symptoms from baseline to after 3 months of identified allergen avoidance, a Mann-Whitney U test was performed, with P<.05 being considered statistically significant.

Results

Thirty-seven consecutive patients underwent the testing and were eligible for the study. Nineteen patients were included in the study by virtue of completing and returning their posttest food avoidance questionnaire and informed consent. Eighteen patients were White and 1 was Asian. Subcategories of IBS were diarrhea predominant (9 [47.4%]), constipation predominant (3 [15.8%]), mixed type (5 [26.3%]), and undetermined type (2 [10.5%]). Questionnaire answers were reported after a mean (SD) duration of patch test–directed food avoidance of 4.5 (3.0) months (Table 1).

Overall Improvement
Fifteen (78.9%) patients reported at least slight to great improvement in their global IBS symptoms, and 4 (21.1%) reported no improvement (Table 2), with a mean (SD) improvement score of 5.1 (3.3)(P<.00001).

Comment

Despite intense research interest and a growing number of new medications for IBS approved by the US Food and Drug Administration, there remains a large void in the search for cost-effective and efficacious approaches for IBS evaluation and treatment. In addition to major disturbances in quality of life,^14,15 the cost to society in direct medical expenses and indirect costs associated with loss of productivity and work absenteeism is considerable; estimates range from $21 billion or more annually.¹⁶

Food Hypersensitivities Triggering IBS
This study further evaluated a role for skin patch testing to identify delayed-type (type IV) food hypersensitivities that trigger IBS symptoms and differed from the prior investigations^9,10 in that the symptoms used to define IBS were updated from the Rome III¹⁷ to the newer Rome IV² criteria. The data presented here show moderate to great improvement in global IBS symptoms in 58% (11/19) of patients, which is in line with a 2018 report of 40 study participants for whom follow-up at 3 or more months was available,⁹ providing additional support for a role for type IV food allergies in causing the same gastrointestinal tract symptoms that define IBS. The distinction between food-related studies, including this one, that implicate food allergies^9,10 and prior studies that did not support a role for food allergies in IBS pathogenesis⁸ can be accounted for by the type of allergy investigated. Conclusions that IBS flares after food ingestion were attributable to intolerance rather than true allergy were based on results investigating only the humoral arm and failed to consider the cell-mediated arm of the immune system. As such, foods that appear to trigger IBS symptoms on an allergic basis in our study are recognized in the literature¹² as type IV allergens that elicit cell-mediated immunologic responses rather than more widely recognized type I allergens, such as peanuts and shellfish, that elicit immediate-type hypersensitivity responses. Although any type IV food allergen(s) could be responsible, a pattern emerged in this study and the study published in 2018.⁹ Namely, some foods stood out as more frequently inducing patch test reactions, with the 3 most common being carmine, cinnamon bark oil, and sodium bisulfite (eTable). The sample size is relatively small, but the results raise the question of whether these foods are the most likely to trigger IBS symptoms in the general population. If so, is it the result of a higher innate sensitizing potential and/or a higher frequency of exposure in commonly eaten foods? Larger randomized clinical trials are needed.

Immune Response and IBS
There is mounting evidence that the immune system may play a role in the pathophysiology of IBS.¹⁸ Both lymphocyte infiltration of the myenteric plexus and an increase in intestinal mucosal T lymphocytes have been observed, and it is generally accepted that the mucosal immune system seems to be activated, at least in a subset of patients with IBS.¹⁹ Irritable bowel syndrome associations with quiescent inflammatory bowel disease or postinfectious gastroenteritis provide 2 potential causes for the inflammation, but most IBS patients have had neither.²⁰ The mucosal lining of the intestine and immune system have vast exposure to intraluminal allergens in transit, and it is hypothesized that the same delayed-type hypersensitivity response elicited in the skin by patch testing is elicited in the intestine, resulting in the inflammation that triggers IBS symptoms.¹⁰ The results here add to the growing body of evidence that ingestion of type IV food allergens by previously sensitized individuals could, in fact, be the primary source of the inflammation observed in a large subpopulation of individuals who carry a diagnosis of IBS.

Food Allergens in Patch Testing
Many of the food allergens used in this study are commonly found in various nonfood products that may contact the skin. For example, many flavorings are used as fragrances, and many preservatives, binders, thickeners, emulsifiers, and stabilizers serve the same role in moisturizers, cosmetics, and topical medications. Likewise, nickel sulfate hexahydrate, ubiquitous in foods that arise from the earth, often is found in metal in jewelry, clothing components, and cell phones. All are potential sensitizers. Thus, the question may arise whether the causal relationship between the food allergens identified by patch testing and IBS symptoms might be more of a systemic effect akin to systemic contact dermatitis as sometimes follows ingestion of an allergen to which an individual has been topically sensitized, rather than the proposed localized immunologic response in the intestinal lining. We were unaware of patient history of allergic contact dermatitis to any of the patch test allergens in this study, but the dermatologist author here (M.S.) has unpublished experience with 2 other patients with IBS who have benefited from low-nickel diets after having had positive patch tests to nickel sulfate hexahydrate and who, in retrospect, did report a history of earring dermatitis. Future investigations using pre– and post–food challenge histologic assessments of the intestinal mucosa in patients who benefit from patch test–guided food avoidance diets should help to better define the mechanism.

Because IBS has not been traditionally associated with structural or biochemical abnormalities detectable with current routine diagnostic tools, it has long been viewed as a functional disorder. The findings published more recently,^9,10 in addition to this study’s results, would negate this functional classification in the subset of patients with IBS symptoms who experience sustained relief of their symptoms by patch test–directed food avoidance. The underlying delayed-type hypersensitivity pathogenesis of the IBS-like symptoms in these individuals would mandate an organic classification, aptly named allergic contact enteritis.¹⁰

Follow-up Data
The mean (SD) follow-up duration for this study and the 2018 report⁹ was 4.5 (3.0) months and 7.6 (3.9) months, respectively. The placebo effect is a concern for disorders such as IBS in which primarily subjective outcome measures are available,²¹ and in a retrospective analysis of 25 randomized, placebo-controlled IBS clinical trials, Spiller²² concluded the optimum length of such trials to be more than 3 months, which these studies exceed. Although not blinded or placebo controlled, the length of follow-up in the 2018 report⁹ and here enhances the validity of the results.

Limitation
The retrospective manner in which the self-assessments were reported in this study introduces the potential for recall bias, a variable that could affect results. The presence and direction of bias by any given individual cannot be known, making it difficult to determine any effect it may have had. Further investigation should include daily assessments and refine the primary study end points to include both abdominal pain and the defecation considerations that define IBS.

Conclusion

Food patch testing has the potential to offer a safe, cost-effective approach to the evaluation and management of IBS symptoms. Randomized clinical trials are needed to further investigate the validity of the proof-of-concept results to date. For patients who benefit from a patch test–guided avoidance diet, invasive and costly endoscopic, radiologic, and laboratory testing and pharmacologic management could be averted. Symptomatic relief could be attained simply by avoiding the implicated foods, essentially doing more by doing less. 


Irritable bowel syndrome (IBS) is one of the most common disorders managed by primary care physicians and gastroenterologists.¹ Characterized by abdominal pain coinciding with altered stool form and/or frequency as defined by the Rome IV diagnostic criteria,² symptoms range from mild to debilitating and may remarkably impair quality of life and work productivity.¹

The cause of IBS is poorly understood. Proposed pathophysiologic factors include impaired mucosal function, microbial imbalance, visceral hypersensitivity, psychologic dysfunction, genetic factors, neurotransmitter imbalance, postinfectious gastroenteritis, inflammation, and food intolerance, any or all of which may lead to the development and maintenance of IBS symptoms.³ More recent observations of inflammation in the intestinal lining^4,5 and proinflammatory peripherally circulating cytokines⁶ challenge its traditional classification as a functional disorder.

The cause of this inflammation is of intense interest, with speculation that the bacterial microbiota, bile acids, association with postinfectious gastroenteritis and inflammatory bowel disease cases, and/or foods may contribute. Although approximately 50% of individuals with IBS report that foods aggravate their symptoms,⁷ studies investigating type I antibody–mediated immediate hypersensitivity have largely failed to demonstrate a substantial link, prompting many authorities to regard these associations as food “intolerances” rather than true allergies. Based on this body of literature, a large 2010 consensus report on all aspects of food allergies advises against food allergy testing for IBS.⁸

In contrast, by utilizing type IV food allergen skin patch testing, 2 proof-of-concept studies^9,10 investigated a different allergic mechanism in IBS, namely cell-mediated delayed-type hypersensitivity. Because many foods and food additives are known to cause allergic contact dermatitis,¹¹ it was hypothesized that these foods may elicit a similar delayed-type hypersensitivity response in the intestinal lining in previously sensitized individuals. By following a patch test–guided food avoidance diet, a large subpopulation of patients with IBS experienced partial or complete IBS symptom relief.^9,10 Our study further investigates a role for food-related delayed-type hypersensitivities in the pathogenesis of IBS.

Methods

Patient Selection
This study was conducted in a secondary care community-based setting. All patients were self-referred over an 18-month period ending in October 2019, had physician-diagnosed IBS, and/or met the Rome IV criteria for IBS and presented expressly for the food patch testing on a fee-for-service basis. Subtype of IBS was determined on presentation by the self-reported historically predominant symptom. Duration of IBS symptoms was self-reported and was rounded to the nearest year for purposes of data collection.

Exclusion criteria included pregnancy, known allergy to adhesive tape or any of the food allergens used in the study, severe skin rash, symptoms that had a known cause other than IBS, or active treatment with systemic immunosuppressive medications.

Patch Testing
Skin patch testing was initiated using an extensive panel of 117 type IV food allergens (eTable)¹¹ identified in the literature,¹² most of which utilized standard compounded formulations¹³ or were available from reputable patch test manufacturers (Brial Allergen GmbH; Chemotechnique Diagnostics). This panel was not approved by the US Food and Drug Administration. The freeze-dried vegetable formulations were taken from the 2018 report.⁹ Standard skin patch test procedure protocols¹² were used, affixing the patches to the upper aspect of the back.

Following patch test application on day 1, two follow-up visits occurred on day 3 and either day 4 or day 5. On day 3, patches were removed, and the initial results were read by a board-certified dermatologist according to a standard grading system.¹⁴ Interpretation of patch tests included no reaction, questionable reaction consisting of macular erythema, weak reaction consisting of erythema and slight edema, or strong reaction consisting of erythema and marked edema. On day 4 or day 5, the final patch test reading was performed, and patients were informed of their results. Patients were advised to avoid ingestion of all foods that elicited a questionable or positive patch test response for at least 3 months, and information about the foods and their avoidance also was distributed and reviewed.

Food Avoidance Questionnaire
Patients with questionable or positive patch tests at 72 or 96 hours were advised of their eligibility to participate in an institutional review board–approved food avoidance questionnaire study investigating the utility of patch test–guided food avoidance on IBS symptoms. The questionnaire assessed the following: (1) baseline average abdominal pain prior to patch test–guided avoidance diet (0=no symptoms; 10=very severe); (2) average abdominal pain since initiation of patch test–guided avoidance diet (0=no symptoms; 10=very severe); (3) degree of improvement in overall IBS symptoms by the end of the food avoidance period (0=no improvement; 10=great improvement); (4) compliance with the avoidance diet for the duration of the avoidance period (completely, partially, not at all, or not sure).

Questionnaires and informed consent were mailed to patients via the US Postal Service 3 months after completing the patch testing. The questionnaire and consent were to be completed and returned after dietary avoidance of the identified allergens for at least 3 months. Patients were not compensated for participation in the study.

Statistical Analysis
Statistical analysis of data collected from study questionnaires was performed with Microsoft Excel. Mean abdominal pain and mean global improvement scores were reported along with 1 SD of the mean. For comparison of mean abdominal pain and improvement in global IBS symptoms from baseline to after 3 months of identified allergen avoidance, a Mann-Whitney U test was performed, with P<.05 being considered statistically significant.

Results

Thirty-seven consecutive patients underwent the testing and were eligible for the study. Nineteen patients were included in the study by virtue of completing and returning their posttest food avoidance questionnaire and informed consent. Eighteen patients were White and 1 was Asian. Subcategories of IBS were diarrhea predominant (9 [47.4%]), constipation predominant (3 [15.8%]), mixed type (5 [26.3%]), and undetermined type (2 [10.5%]). Questionnaire answers were reported after a mean (SD) duration of patch test–directed food avoidance of 4.5 (3.0) months (Table 1).

Overall Improvement
Fifteen (78.9%) patients reported at least slight to great improvement in their global IBS symptoms, and 4 (21.1%) reported no improvement (Table 2), with a mean (SD) improvement score of 5.1 (3.3)(P<.00001).

Comment

Despite intense research interest and a growing number of new medications for IBS approved by the US Food and Drug Administration, there remains a large void in the search for cost-effective and efficacious approaches for IBS evaluation and treatment. In addition to major disturbances in quality of life,^14,15 the cost to society in direct medical expenses and indirect costs associated with loss of productivity and work absenteeism is considerable; estimates range from $21 billion or more annually.¹⁶

Food Hypersensitivities Triggering IBS
This study further evaluated a role for skin patch testing to identify delayed-type (type IV) food hypersensitivities that trigger IBS symptoms and differed from the prior investigations^9,10 in that the symptoms used to define IBS were updated from the Rome III¹⁷ to the newer Rome IV² criteria. The data presented here show moderate to great improvement in global IBS symptoms in 58% (11/19) of patients, which is in line with a 2018 report of 40 study participants for whom follow-up at 3 or more months was available,⁹ providing additional support for a role for type IV food allergies in causing the same gastrointestinal tract symptoms that define IBS. The distinction between food-related studies, including this one, that implicate food allergies^9,10 and prior studies that did not support a role for food allergies in IBS pathogenesis⁸ can be accounted for by the type of allergy investigated. Conclusions that IBS flares after food ingestion were attributable to intolerance rather than true allergy were based on results investigating only the humoral arm and failed to consider the cell-mediated arm of the immune system. As such, foods that appear to trigger IBS symptoms on an allergic basis in our study are recognized in the literature¹² as type IV allergens that elicit cell-mediated immunologic responses rather than more widely recognized type I allergens, such as peanuts and shellfish, that elicit immediate-type hypersensitivity responses. Although any type IV food allergen(s) could be responsible, a pattern emerged in this study and the study published in 2018.⁹ Namely, some foods stood out as more frequently inducing patch test reactions, with the 3 most common being carmine, cinnamon bark oil, and sodium bisulfite (eTable). The sample size is relatively small, but the results raise the question of whether these foods are the most likely to trigger IBS symptoms in the general population. If so, is it the result of a higher innate sensitizing potential and/or a higher frequency of exposure in commonly eaten foods? Larger randomized clinical trials are needed.

Immune Response and IBS
There is mounting evidence that the immune system may play a role in the pathophysiology of IBS.¹⁸ Both lymphocyte infiltration of the myenteric plexus and an increase in intestinal mucosal T lymphocytes have been observed, and it is generally accepted that the mucosal immune system seems to be activated, at least in a subset of patients with IBS.¹⁹ Irritable bowel syndrome associations with quiescent inflammatory bowel disease or postinfectious gastroenteritis provide 2 potential causes for the inflammation, but most IBS patients have had neither.²⁰ The mucosal lining of the intestine and immune system have vast exposure to intraluminal allergens in transit, and it is hypothesized that the same delayed-type hypersensitivity response elicited in the skin by patch testing is elicited in the intestine, resulting in the inflammation that triggers IBS symptoms.¹⁰ The results here add to the growing body of evidence that ingestion of type IV food allergens by previously sensitized individuals could, in fact, be the primary source of the inflammation observed in a large subpopulation of individuals who carry a diagnosis of IBS.

Food Allergens in Patch Testing
Many of the food allergens used in this study are commonly found in various nonfood products that may contact the skin. For example, many flavorings are used as fragrances, and many preservatives, binders, thickeners, emulsifiers, and stabilizers serve the same role in moisturizers, cosmetics, and topical medications. Likewise, nickel sulfate hexahydrate, ubiquitous in foods that arise from the earth, often is found in metal in jewelry, clothing components, and cell phones. All are potential sensitizers. Thus, the question may arise whether the causal relationship between the food allergens identified by patch testing and IBS symptoms might be more of a systemic effect akin to systemic contact dermatitis as sometimes follows ingestion of an allergen to which an individual has been topically sensitized, rather than the proposed localized immunologic response in the intestinal lining. We were unaware of patient history of allergic contact dermatitis to any of the patch test allergens in this study, but the dermatologist author here (M.S.) has unpublished experience with 2 other patients with IBS who have benefited from low-nickel diets after having had positive patch tests to nickel sulfate hexahydrate and who, in retrospect, did report a history of earring dermatitis. Future investigations using pre– and post–food challenge histologic assessments of the intestinal mucosa in patients who benefit from patch test–guided food avoidance diets should help to better define the mechanism.

Because IBS has not been traditionally associated with structural or biochemical abnormalities detectable with current routine diagnostic tools, it has long been viewed as a functional disorder. The findings published more recently,^9,10 in addition to this study’s results, would negate this functional classification in the subset of patients with IBS symptoms who experience sustained relief of their symptoms by patch test–directed food avoidance. The underlying delayed-type hypersensitivity pathogenesis of the IBS-like symptoms in these individuals would mandate an organic classification, aptly named allergic contact enteritis.¹⁰

Follow-up Data
The mean (SD) follow-up duration for this study and the 2018 report⁹ was 4.5 (3.0) months and 7.6 (3.9) months, respectively. The placebo effect is a concern for disorders such as IBS in which primarily subjective outcome measures are available,²¹ and in a retrospective analysis of 25 randomized, placebo-controlled IBS clinical trials, Spiller²² concluded the optimum length of such trials to be more than 3 months, which these studies exceed. Although not blinded or placebo controlled, the length of follow-up in the 2018 report⁹ and here enhances the validity of the results.

Limitation
The retrospective manner in which the self-assessments were reported in this study introduces the potential for recall bias, a variable that could affect results. The presence and direction of bias by any given individual cannot be known, making it difficult to determine any effect it may have had. Further investigation should include daily assessments and refine the primary study end points to include both abdominal pain and the defecation considerations that define IBS.

Conclusion

Food patch testing has the potential to offer a safe, cost-effective approach to the evaluation and management of IBS symptoms. Randomized clinical trials are needed to further investigate the validity of the proof-of-concept results to date. For patients who benefit from a patch test–guided avoidance diet, invasive and costly endoscopic, radiologic, and laboratory testing and pharmacologic management could be averted. Symptomatic relief could be attained simply by avoiding the implicated foods, essentially doing more by doing less. 


Irritable bowel syndrome (IBS) is one of the most common disorders managed by primary care physicians and gastroenterologists.¹ Characterized by abdominal pain coinciding with altered stool form and/or frequency as defined by the Rome IV diagnostic criteria,² symptoms range from mild to debilitating and may remarkably impair quality of life and work productivity.¹

The cause of IBS is poorly understood. Proposed pathophysiologic factors include impaired mucosal function, microbial imbalance, visceral hypersensitivity, psychologic dysfunction, genetic factors, neurotransmitter imbalance, postinfectious gastroenteritis, inflammation, and food intolerance, any or all of which may lead to the development and maintenance of IBS symptoms.³ More recent observations of inflammation in the intestinal lining^4,5 and proinflammatory peripherally circulating cytokines⁶ challenge its traditional classification as a functional disorder.

The cause of this inflammation is of intense interest, with speculation that the bacterial microbiota, bile acids, association with postinfectious gastroenteritis and inflammatory bowel disease cases, and/or foods may contribute. Although approximately 50% of individuals with IBS report that foods aggravate their symptoms,⁷ studies investigating type I antibody–mediated immediate hypersensitivity have largely failed to demonstrate a substantial link, prompting many authorities to regard these associations as food “intolerances” rather than true allergies. Based on this body of literature, a large 2010 consensus report on all aspects of food allergies advises against food allergy testing for IBS.⁸

In contrast, by utilizing type IV food allergen skin patch testing, 2 proof-of-concept studies^9,10 investigated a different allergic mechanism in IBS, namely cell-mediated delayed-type hypersensitivity. Because many foods and food additives are known to cause allergic contact dermatitis,¹¹ it was hypothesized that these foods may elicit a similar delayed-type hypersensitivity response in the intestinal lining in previously sensitized individuals. By following a patch test–guided food avoidance diet, a large subpopulation of patients with IBS experienced partial or complete IBS symptom relief.^9,10 Our study further investigates a role for food-related delayed-type hypersensitivities in the pathogenesis of IBS.

Methods

Patient Selection
This study was conducted in a secondary care community-based setting. All patients were self-referred over an 18-month period ending in October 2019, had physician-diagnosed IBS, and/or met the Rome IV criteria for IBS and presented expressly for the food patch testing on a fee-for-service basis. Subtype of IBS was determined on presentation by the self-reported historically predominant symptom. Duration of IBS symptoms was self-reported and was rounded to the nearest year for purposes of data collection.

Exclusion criteria included pregnancy, known allergy to adhesive tape or any of the food allergens used in the study, severe skin rash, symptoms that had a known cause other than IBS, or active treatment with systemic immunosuppressive medications.

Patch Testing
Skin patch testing was initiated using an extensive panel of 117 type IV food allergens (eTable)¹¹ identified in the literature,¹² most of which utilized standard compounded formulations¹³ or were available from reputable patch test manufacturers (Brial Allergen GmbH; Chemotechnique Diagnostics). This panel was not approved by the US Food and Drug Administration. The freeze-dried vegetable formulations were taken from the 2018 report.⁹ Standard skin patch test procedure protocols¹² were used, affixing the patches to the upper aspect of the back.

Following patch test application on day 1, two follow-up visits occurred on day 3 and either day 4 or day 5. On day 3, patches were removed, and the initial results were read by a board-certified dermatologist according to a standard grading system.¹⁴ Interpretation of patch tests included no reaction, questionable reaction consisting of macular erythema, weak reaction consisting of erythema and slight edema, or strong reaction consisting of erythema and marked edema. On day 4 or day 5, the final patch test reading was performed, and patients were informed of their results. Patients were advised to avoid ingestion of all foods that elicited a questionable or positive patch test response for at least 3 months, and information about the foods and their avoidance also was distributed and reviewed.

Food Avoidance Questionnaire
Patients with questionable or positive patch tests at 72 or 96 hours were advised of their eligibility to participate in an institutional review board–approved food avoidance questionnaire study investigating the utility of patch test–guided food avoidance on IBS symptoms. The questionnaire assessed the following: (1) baseline average abdominal pain prior to patch test–guided avoidance diet (0=no symptoms; 10=very severe); (2) average abdominal pain since initiation of patch test–guided avoidance diet (0=no symptoms; 10=very severe); (3) degree of improvement in overall IBS symptoms by the end of the food avoidance period (0=no improvement; 10=great improvement); (4) compliance with the avoidance diet for the duration of the avoidance period (completely, partially, not at all, or not sure).

Questionnaires and informed consent were mailed to patients via the US Postal Service 3 months after completing the patch testing. The questionnaire and consent were to be completed and returned after dietary avoidance of the identified allergens for at least 3 months. Patients were not compensated for participation in the study.

Statistical Analysis
Statistical analysis of data collected from study questionnaires was performed with Microsoft Excel. Mean abdominal pain and mean global improvement scores were reported along with 1 SD of the mean. For comparison of mean abdominal pain and improvement in global IBS symptoms from baseline to after 3 months of identified allergen avoidance, a Mann-Whitney U test was performed, with P<.05 being considered statistically significant.

Results

Thirty-seven consecutive patients underwent the testing and were eligible for the study. Nineteen patients were included in the study by virtue of completing and returning their posttest food avoidance questionnaire and informed consent. Eighteen patients were White and 1 was Asian. Subcategories of IBS were diarrhea predominant (9 [47.4%]), constipation predominant (3 [15.8%]), mixed type (5 [26.3%]), and undetermined type (2 [10.5%]). Questionnaire answers were reported after a mean (SD) duration of patch test–directed food avoidance of 4.5 (3.0) months (Table 1).

Overall Improvement
Fifteen (78.9%) patients reported at least slight to great improvement in their global IBS symptoms, and 4 (21.1%) reported no improvement (Table 2), with a mean (SD) improvement score of 5.1 (3.3)(P<.00001).

Comment

Despite intense research interest and a growing number of new medications for IBS approved by the US Food and Drug Administration, there remains a large void in the search for cost-effective and efficacious approaches for IBS evaluation and treatment. In addition to major disturbances in quality of life,^14,15 the cost to society in direct medical expenses and indirect costs associated with loss of productivity and work absenteeism is considerable; estimates range from $21 billion or more annually.¹⁶

Food Hypersensitivities Triggering IBS
This study further evaluated a role for skin patch testing to identify delayed-type (type IV) food hypersensitivities that trigger IBS symptoms and differed from the prior investigations^9,10 in that the symptoms used to define IBS were updated from the Rome III¹⁷ to the newer Rome IV² criteria. The data presented here show moderate to great improvement in global IBS symptoms in 58% (11/19) of patients, which is in line with a 2018 report of 40 study participants for whom follow-up at 3 or more months was available,⁹ providing additional support for a role for type IV food allergies in causing the same gastrointestinal tract symptoms that define IBS. The distinction between food-related studies, including this one, that implicate food allergies^9,10 and prior studies that did not support a role for food allergies in IBS pathogenesis⁸ can be accounted for by the type of allergy investigated. Conclusions that IBS flares after food ingestion were attributable to intolerance rather than true allergy were based on results investigating only the humoral arm and failed to consider the cell-mediated arm of the immune system. As such, foods that appear to trigger IBS symptoms on an allergic basis in our study are recognized in the literature¹² as type IV allergens that elicit cell-mediated immunologic responses rather than more widely recognized type I allergens, such as peanuts and shellfish, that elicit immediate-type hypersensitivity responses. Although any type IV food allergen(s) could be responsible, a pattern emerged in this study and the study published in 2018.⁹ Namely, some foods stood out as more frequently inducing patch test reactions, with the 3 most common being carmine, cinnamon bark oil, and sodium bisulfite (eTable). The sample size is relatively small, but the results raise the question of whether these foods are the most likely to trigger IBS symptoms in the general population. If so, is it the result of a higher innate sensitizing potential and/or a higher frequency of exposure in commonly eaten foods? Larger randomized clinical trials are needed.

Immune Response and IBS
There is mounting evidence that the immune system may play a role in the pathophysiology of IBS.¹⁸ Both lymphocyte infiltration of the myenteric plexus and an increase in intestinal mucosal T lymphocytes have been observed, and it is generally accepted that the mucosal immune system seems to be activated, at least in a subset of patients with IBS.¹⁹ Irritable bowel syndrome associations with quiescent inflammatory bowel disease or postinfectious gastroenteritis provide 2 potential causes for the inflammation, but most IBS patients have had neither.²⁰ The mucosal lining of the intestine and immune system have vast exposure to intraluminal allergens in transit, and it is hypothesized that the same delayed-type hypersensitivity response elicited in the skin by patch testing is elicited in the intestine, resulting in the inflammation that triggers IBS symptoms.¹⁰ The results here add to the growing body of evidence that ingestion of type IV food allergens by previously sensitized individuals could, in fact, be the primary source of the inflammation observed in a large subpopulation of individuals who carry a diagnosis of IBS.

Food Allergens in Patch Testing
Many of the food allergens used in this study are commonly found in various nonfood products that may contact the skin. For example, many flavorings are used as fragrances, and many preservatives, binders, thickeners, emulsifiers, and stabilizers serve the same role in moisturizers, cosmetics, and topical medications. Likewise, nickel sulfate hexahydrate, ubiquitous in foods that arise from the earth, often is found in metal in jewelry, clothing components, and cell phones. All are potential sensitizers. Thus, the question may arise whether the causal relationship between the food allergens identified by patch testing and IBS symptoms might be more of a systemic effect akin to systemic contact dermatitis as sometimes follows ingestion of an allergen to which an individual has been topically sensitized, rather than the proposed localized immunologic response in the intestinal lining. We were unaware of patient history of allergic contact dermatitis to any of the patch test allergens in this study, but the dermatologist author here (M.S.) has unpublished experience with 2 other patients with IBS who have benefited from low-nickel diets after having had positive patch tests to nickel sulfate hexahydrate and who, in retrospect, did report a history of earring dermatitis. Future investigations using pre– and post–food challenge histologic assessments of the intestinal mucosa in patients who benefit from patch test–guided food avoidance diets should help to better define the mechanism.

Because IBS has not been traditionally associated with structural or biochemical abnormalities detectable with current routine diagnostic tools, it has long been viewed as a functional disorder. The findings published more recently,^9,10 in addition to this study’s results, would negate this functional classification in the subset of patients with IBS symptoms who experience sustained relief of their symptoms by patch test–directed food avoidance. The underlying delayed-type hypersensitivity pathogenesis of the IBS-like symptoms in these individuals would mandate an organic classification, aptly named allergic contact enteritis.¹⁰

Follow-up Data
The mean (SD) follow-up duration for this study and the 2018 report⁹ was 4.5 (3.0) months and 7.6 (3.9) months, respectively. The placebo effect is a concern for disorders such as IBS in which primarily subjective outcome measures are available,²¹ and in a retrospective analysis of 25 randomized, placebo-controlled IBS clinical trials, Spiller²² concluded the optimum length of such trials to be more than 3 months, which these studies exceed. Although not blinded or placebo controlled, the length of follow-up in the 2018 report⁹ and here enhances the validity of the results.

Limitation
The retrospective manner in which the self-assessments were reported in this study introduces the potential for recall bias, a variable that could affect results. The presence and direction of bias by any given individual cannot be known, making it difficult to determine any effect it may have had. Further investigation should include daily assessments and refine the primary study end points to include both abdominal pain and the defecation considerations that define IBS.

Conclusion

Food patch testing has the potential to offer a safe, cost-effective approach to the evaluation and management of IBS symptoms. Randomized clinical trials are needed to further investigate the validity of the proof-of-concept results to date. For patients who benefit from a patch test–guided avoidance diet, invasive and costly endoscopic, radiologic, and laboratory testing and pharmacologic management could be averted. Symptomatic relief could be attained simply by avoiding the implicated foods, essentially doing more by doing less. 


Infection with hepatitis B virus (HBV) is associated with risk of potentially lethal, chronic infection and is a major public health problem. Infection from HBV has the potential to lead to liver failure, cirrhosis, and cancer.^1,2 Chronic HBV infection exists in as many as 2.2 million Americans, and in 2015 alone, HBV was estimated to be associated with 887,000 deaths worldwide.^1,3 Suppression of viral load is the basis of treatment, necessitating long-term use of medication for treatment.⁴ Nucleoside reverse transcriptase inhibitors (entecavir, lamivudine, telbivudine) and nucleotide reverse transcriptase inhibitors (adefovir, tenofovir), have improved the efficacy and tolerability of chronic HBV treatment compared with interferon-based agents.^4-7 However, concerns remain regarding long-term risk of nephrotoxicity, in particular with tenofovir disoproxil fumarate (TDF), which could lead to a limitation of safe and effective options for certain populations.^5,6,8 A newer formulation, tenofovir alafenamide fumarate (TAF), has improved the kidney risks, but expense remains a limiting factor for this agent.⁹

Nucleos(t)ide reverse transcriptase inhibitors (NRTIs) have demonstrated efficacy in reducing HBV viral load and other markers of improvement in chronic HBV, but entecavir and tenofovir have tended to demonstrate greater efficacy in clinical trials.^5-7 Several studies have suggested potential benefits of tenofovir-based treatment over other NRTIs, including greater viral load achievement compared with adefovir, efficacy in patients with previous failure of lamivudine or adefovir, and long-term efficacy in chronic HBV infection.^10-12 A 2019 systematic review suggests TDF and TAF are more effective than other NRTIs for achieving viral load suppression.¹³ Other NRTIs are not without their own risks, including mitochondrial dysfunction, mostly with lamivudine and telbivudine.⁴

Despite these data, guidelines have varied in their treatment recommendations in the context of chronic kidney disease partly due to variations in the evidence regarding nephrotoxicity.^7,14 Cohort studies and case reports have suggested association between TDF and acute kidney injury in patients with HIV infection as well as long-term reductions in kidney function.^15,16 In one study, 58% of patients treated with TDF did not return to baseline kidney function after an event of acute kidney injury.¹⁷ However, little data are available on whether this association exists for chronic HBV treatment in the absence of HIV infection. One retrospective analysis comparing TDF and entecavir in chronic HBV without HIV showed greater incidence of creatinine clearance < 60 mL/min with TDF but greater incidence of serum creatinine (SCr) ≥ 2.5 mg/dL in the entacavir group, making it difficult to reach a clear conclusion on risks.¹⁸ Other studies have either suffered from small cohorts with TDF or included patients with HIV coinfection.^19,20 Although a retrospective comparison of TDF and entecavir, randomly matched 1:2 to account for differences between groups, showed lower estimated glomerular filtration rate (eGFR) in the TDF group, more data are needed.²¹ Entecavir remains an option for many patient, but for those who have failed nucleosides, few options remain.

With the advantages available from TDF and the continued expense of TAF, more data regarding the risks of nephrotoxicity with TDF would be beneficial. The objective of this study was to compare treatment with TDF and other NRTIs in chronic HBV monoinfection to distinguish any differences in kidney function changes over time. With hopes of gathering enough data to distinguish between groups, information was gathered from across the Veterans Health Administration (VHA) system.

Methods

A nationwide, multicenter, retrospective, cohort study of veterans with HBV infection was conducted to compare the effects of various NRTIs on renal function. Patient were identified through the US Department of Veterans Affairs Corporate Data Warehouse (CDW), using data from July 1, 2005 to July 31, 2015. Patients were included who had positive HBV surface antigen (HBsAg) or newly prescribed NRTI. Multiple drug episodes could be included for each patient. That is, if a patient who had previously been included had another instance of a newly prescribed NRTI, this would be included in the analysis. Exclusion criteria were patients aged < 18 years, those with NRTI prescription for ≤ 1 month, and concurrent HIV infection. All patients with HBsAg were included for the study for increasing the sensitivity in gathering patients; however, those patients were included only if they received NRTI concurrent with the laboratory test results used for the primary endpoint (ie, SCr) to be included in the analysis.

How data are received from CDW bears some explanation. A basic way to understand the way data are received is that questions can be asked such as “for X population, at this point in time, was the patient on Y drug and what was the SCr value.” Therefore, inclusion and exclusion must first be specified to define the population, after which point certain data points can be received depending on the specifications made. For this reason, there is no way to determine, for example, whether a certain patient continued TDF use for the duration of the study, only at the defined points in time (described below) to receive the specific data.

For the patients included, information was retrieved from the first receipt of the NRTI prescription to 36 months after initiation. Baseline characteristics included age, sex, race, and ethnicity, and were defined at time of NRTI initiation. Values for SCr were compared at baseline, 3, 6, 12, 24, and 36 months after prescription of NRTI. The date of laboratory results was associated with the nearest date of comparison. Values for eGFR were determined by the modification of diet in renal disease equation. Values for eGFR are available in the CDW, whereas there is no direct means to calculate creatinine clearance with the available data, so eGFR was used for this study.

The primary endpoint was a change in eGFR in patients taking TDF after adjustment for time with the full cohort. Secondary analyses included the overall effect of time for the full cohort and change in renal function for each NRTI group. Mean and standard deviation for eGFR were determined for each NRTI group using the available data points. Analyses of the primary and secondary endpoints were completed using a linear mixed model with terms for time, to account for fixed effects, and specific NRTI used to account for random effects. A 2-sided α of .05 was used to determine statistical significance.

Results

A total of 413 drug episodes from 308 subjects met inclusion criteria for the study. Of these subjects, 229 were still living at the time of query. Most study participants were male (96%), the mean age was 62.1 years for males and 55.9 years for females; 49.5% were White and 39.7% were Black veterans (Table 1).

The NRTIs received by patients during the study period included TDF, TDF/emtricitabine, adefovir, entecavir, and lamivudine. No patients were on telbivudine. Formulations including TAF had not been approved by the US Food and Drug Administration (FDA) by the end of the study period, and as such were not found in the study.¹³ A plurality of participants received entecavir (94 of 223 at baseline), followed by TDF (n = 38) (Table 2). Of note, only 8 participants received TDF/emtricitabine at baseline. Differences were found between the groups in number of SCr data points available at 36 months vs baseline. The TDF group had the greatest reduction in data points available with 38 laboratory values at baseline vs 15 at 36 months (39.5% of baseline). From the available data, it is not possible to determine whether these represent medication discontinuations, missing values, lost to follow-up, or some other cause. Baseline eGFR was highest in the 2 TDF groups, with TDF alone at 77.7 mL/min (1.4-5.5 mL/min higher than the nontenofovir groups) and TDF/emtricitabine at 89.7 mL/min (13.4-17.5 mL/min higher than nontenofovir groups) (Table 3).

Discussion

This study demonstrated a decline in eGFR over time in a similar fashion for all NRTIs used in patients treated for HBV monoinfection, but no greater decline in renal function was found with use of TDF vs other NRTIs. A statistically significant decline in eGFR of −4.55 mL/min over the 36-month time frame of the study was demonstrated for the full cohort, but no statistically significant change in eGFR was found for any individual NRTI after accounting for the fixed effect of time. If TDF is not associated with additional risk of nephrotoxicity compared with other NRTIs, this could have important implications for treatment when considering the evidence that tenofovir-based treatment seems to be more effective than other medications for suppressing viral load.¹³

This result runs contrary to data in patients given NRTIs for HIV infection as well as a more recent cohort study in chronic HBV infectioin, which showed a statistically significant difference in kidney dysfunction between TDF and entecavir (-15.73 vs -5.96 mL/min/m², P < .001).^5-7,21 Possible mechanism for differences in response between HIV and HBV patients has not been elucidated, but the inherent risk of developing chronic kidney disease from HIV disease may play a role.²² The possibility remains that all NRTIs cause a degree of kidney impairment in patients treated for chronic HBV infection as evidenced by the statistically significant fixed effect for time in the present study. The cause of this effect is unknown but may be independently related to HBV infection or may be specific to NRTI therapy. No control group of patients not receiving NRTI therapy was included in this study, so conclusions cannot be drawn regarding whether all NRTIs are associated with decline in renal function in chronic HBV infection.

Limitations

Although this study did not detect a difference in change in eGFR between TDF and other NRTI treatments, it is possible that the length of data collection was not adequate to account for possible kidney injury from TDF. A study assessing renal tubular dysfunction in patients receiving adefovir or TDF showed a mean onset of dysfunction of 49 months.¹⁵ It is possible that participants in this study would go on to develop renal dysfunction in the future. This potential also was observed in a more recent retrospective cohort study in chronic HBV infection, which showed the greatest degree of decline in kidney function between 36 and 48 months (−11.87 to −15.73 mL/min/m² for the TDF group).²¹

The retrospective design created additional limitations. We attempted to account for some by using a matched cohort for the entecavir group, and there was no statistically significant difference between the groups in baseline characteristics. In HIV patients, a 10-year follow-up study continued to show decline in eGFR throughout the study, though the greatest degree of reduction occurred in the first year of the study.¹⁰ The higher baseline eGFR of the TDF recipients, 77.7 mL/min for the TDF alone group and 89.7 mL/min for the TDF/emtricitabine group vs 72.2 to 76.3 mL/min in the other NRTI groups, suggests high potential for selection bias. Some health care providers were likely to avoid TDF in patients with lower eGFR due to the data suggesting nephrotoxicity in other populations. Another limitation is that the reason for the missing laboratory values could not be determined. The TDF group had the greatest disparity in SCr data availability at baseline vs 36 months, with 39.5% concurrence with TDF alone compared with 50.0 to 63.6% in the other groups. Other treatment received outside the VHA system also could have influenced results.

Conclusions

This retrospective, multicenter, cohort study did not find a difference between TDF and other NRTIs for changes in renal function over time in patients with HBV infection without HIV. There was a fixed effect for time, ie, all NRTI groups showed some decline in renal function over time (−4.6 mL/min), but the effects were similar across groups. The results appear contrary to studies with comorbid HIV showing a decline in renal function with TDF, but present studies in HBV monotherapy have mixed results.

Further studies are needed to validate these results, as this and previous studies have several limitations. If these results are confirmed, a possible mechanism for these differences between patients with and without HIV should be examined. In addition, a study looking specifically at incidence of acute kidney injury rather than overall decline in renal function would add important data. If the results of this study are confirmed, there could be clinical implications in choice of agent with treatment of HBV monoinfection. This would add to the overall armament of medications available for chronic HBV infection and could create cost savings in certain situations if providers feel more comfortable continuing to use TDF instead of switching to the more expensive TAF.

Acknowledgments
Funding for this study was provided by the Veterans Health Administration.

Infection with hepatitis B virus (HBV) is associated with risk of potentially lethal, chronic infection and is a major public health problem. Infection from HBV has the potential to lead to liver failure, cirrhosis, and cancer.^1,2 Chronic HBV infection exists in as many as 2.2 million Americans, and in 2015 alone, HBV was estimated to be associated with 887,000 deaths worldwide.^1,3 Suppression of viral load is the basis of treatment, necessitating long-term use of medication for treatment.⁴ Nucleoside reverse transcriptase inhibitors (entecavir, lamivudine, telbivudine) and nucleotide reverse transcriptase inhibitors (adefovir, tenofovir), have improved the efficacy and tolerability of chronic HBV treatment compared with interferon-based agents.^4-7 However, concerns remain regarding long-term risk of nephrotoxicity, in particular with tenofovir disoproxil fumarate (TDF), which could lead to a limitation of safe and effective options for certain populations.^5,6,8 A newer formulation, tenofovir alafenamide fumarate (TAF), has improved the kidney risks, but expense remains a limiting factor for this agent.⁹

Nucleos(t)ide reverse transcriptase inhibitors (NRTIs) have demonstrated efficacy in reducing HBV viral load and other markers of improvement in chronic HBV, but entecavir and tenofovir have tended to demonstrate greater efficacy in clinical trials.^5-7 Several studies have suggested potential benefits of tenofovir-based treatment over other NRTIs, including greater viral load achievement compared with adefovir, efficacy in patients with previous failure of lamivudine or adefovir, and long-term efficacy in chronic HBV infection.^10-12 A 2019 systematic review suggests TDF and TAF are more effective than other NRTIs for achieving viral load suppression.¹³ Other NRTIs are not without their own risks, including mitochondrial dysfunction, mostly with lamivudine and telbivudine.⁴

Despite these data, guidelines have varied in their treatment recommendations in the context of chronic kidney disease partly due to variations in the evidence regarding nephrotoxicity.^7,14 Cohort studies and case reports have suggested association between TDF and acute kidney injury in patients with HIV infection as well as long-term reductions in kidney function.^15,16 In one study, 58% of patients treated with TDF did not return to baseline kidney function after an event of acute kidney injury.¹⁷ However, little data are available on whether this association exists for chronic HBV treatment in the absence of HIV infection. One retrospective analysis comparing TDF and entecavir in chronic HBV without HIV showed greater incidence of creatinine clearance < 60 mL/min with TDF but greater incidence of serum creatinine (SCr) ≥ 2.5 mg/dL in the entacavir group, making it difficult to reach a clear conclusion on risks.¹⁸ Other studies have either suffered from small cohorts with TDF or included patients with HIV coinfection.^19,20 Although a retrospective comparison of TDF and entecavir, randomly matched 1:2 to account for differences between groups, showed lower estimated glomerular filtration rate (eGFR) in the TDF group, more data are needed.²¹ Entecavir remains an option for many patient, but for those who have failed nucleosides, few options remain.

With the advantages available from TDF and the continued expense of TAF, more data regarding the risks of nephrotoxicity with TDF would be beneficial. The objective of this study was to compare treatment with TDF and other NRTIs in chronic HBV monoinfection to distinguish any differences in kidney function changes over time. With hopes of gathering enough data to distinguish between groups, information was gathered from across the Veterans Health Administration (VHA) system.

Methods

A nationwide, multicenter, retrospective, cohort study of veterans with HBV infection was conducted to compare the effects of various NRTIs on renal function. Patient were identified through the US Department of Veterans Affairs Corporate Data Warehouse (CDW), using data from July 1, 2005 to July 31, 2015. Patients were included who had positive HBV surface antigen (HBsAg) or newly prescribed NRTI. Multiple drug episodes could be included for each patient. That is, if a patient who had previously been included had another instance of a newly prescribed NRTI, this would be included in the analysis. Exclusion criteria were patients aged < 18 years, those with NRTI prescription for ≤ 1 month, and concurrent HIV infection. All patients with HBsAg were included for the study for increasing the sensitivity in gathering patients; however, those patients were included only if they received NRTI concurrent with the laboratory test results used for the primary endpoint (ie, SCr) to be included in the analysis.

How data are received from CDW bears some explanation. A basic way to understand the way data are received is that questions can be asked such as “for X population, at this point in time, was the patient on Y drug and what was the SCr value.” Therefore, inclusion and exclusion must first be specified to define the population, after which point certain data points can be received depending on the specifications made. For this reason, there is no way to determine, for example, whether a certain patient continued TDF use for the duration of the study, only at the defined points in time (described below) to receive the specific data.

For the patients included, information was retrieved from the first receipt of the NRTI prescription to 36 months after initiation. Baseline characteristics included age, sex, race, and ethnicity, and were defined at time of NRTI initiation. Values for SCr were compared at baseline, 3, 6, 12, 24, and 36 months after prescription of NRTI. The date of laboratory results was associated with the nearest date of comparison. Values for eGFR were determined by the modification of diet in renal disease equation. Values for eGFR are available in the CDW, whereas there is no direct means to calculate creatinine clearance with the available data, so eGFR was used for this study.

The primary endpoint was a change in eGFR in patients taking TDF after adjustment for time with the full cohort. Secondary analyses included the overall effect of time for the full cohort and change in renal function for each NRTI group. Mean and standard deviation for eGFR were determined for each NRTI group using the available data points. Analyses of the primary and secondary endpoints were completed using a linear mixed model with terms for time, to account for fixed effects, and specific NRTI used to account for random effects. A 2-sided α of .05 was used to determine statistical significance.

Results

A total of 413 drug episodes from 308 subjects met inclusion criteria for the study. Of these subjects, 229 were still living at the time of query. Most study participants were male (96%), the mean age was 62.1 years for males and 55.9 years for females; 49.5% were White and 39.7% were Black veterans (Table 1).

The NRTIs received by patients during the study period included TDF, TDF/emtricitabine, adefovir, entecavir, and lamivudine. No patients were on telbivudine. Formulations including TAF had not been approved by the US Food and Drug Administration (FDA) by the end of the study period, and as such were not found in the study.¹³ A plurality of participants received entecavir (94 of 223 at baseline), followed by TDF (n = 38) (Table 2). Of note, only 8 participants received TDF/emtricitabine at baseline. Differences were found between the groups in number of SCr data points available at 36 months vs baseline. The TDF group had the greatest reduction in data points available with 38 laboratory values at baseline vs 15 at 36 months (39.5% of baseline). From the available data, it is not possible to determine whether these represent medication discontinuations, missing values, lost to follow-up, or some other cause. Baseline eGFR was highest in the 2 TDF groups, with TDF alone at 77.7 mL/min (1.4-5.5 mL/min higher than the nontenofovir groups) and TDF/emtricitabine at 89.7 mL/min (13.4-17.5 mL/min higher than nontenofovir groups) (Table 3).

Discussion

This study demonstrated a decline in eGFR over time in a similar fashion for all NRTIs used in patients treated for HBV monoinfection, but no greater decline in renal function was found with use of TDF vs other NRTIs. A statistically significant decline in eGFR of −4.55 mL/min over the 36-month time frame of the study was demonstrated for the full cohort, but no statistically significant change in eGFR was found for any individual NRTI after accounting for the fixed effect of time. If TDF is not associated with additional risk of nephrotoxicity compared with other NRTIs, this could have important implications for treatment when considering the evidence that tenofovir-based treatment seems to be more effective than other medications for suppressing viral load.¹³

This result runs contrary to data in patients given NRTIs for HIV infection as well as a more recent cohort study in chronic HBV infectioin, which showed a statistically significant difference in kidney dysfunction between TDF and entecavir (-15.73 vs -5.96 mL/min/m², P < .001).^5-7,21 Possible mechanism for differences in response between HIV and HBV patients has not been elucidated, but the inherent risk of developing chronic kidney disease from HIV disease may play a role.²² The possibility remains that all NRTIs cause a degree of kidney impairment in patients treated for chronic HBV infection as evidenced by the statistically significant fixed effect for time in the present study. The cause of this effect is unknown but may be independently related to HBV infection or may be specific to NRTI therapy. No control group of patients not receiving NRTI therapy was included in this study, so conclusions cannot be drawn regarding whether all NRTIs are associated with decline in renal function in chronic HBV infection.

Limitations

Although this study did not detect a difference in change in eGFR between TDF and other NRTI treatments, it is possible that the length of data collection was not adequate to account for possible kidney injury from TDF. A study assessing renal tubular dysfunction in patients receiving adefovir or TDF showed a mean onset of dysfunction of 49 months.¹⁵ It is possible that participants in this study would go on to develop renal dysfunction in the future. This potential also was observed in a more recent retrospective cohort study in chronic HBV infection, which showed the greatest degree of decline in kidney function between 36 and 48 months (−11.87 to −15.73 mL/min/m² for the TDF group).²¹

The retrospective design created additional limitations. We attempted to account for some by using a matched cohort for the entecavir group, and there was no statistically significant difference between the groups in baseline characteristics. In HIV patients, a 10-year follow-up study continued to show decline in eGFR throughout the study, though the greatest degree of reduction occurred in the first year of the study.¹⁰ The higher baseline eGFR of the TDF recipients, 77.7 mL/min for the TDF alone group and 89.7 mL/min for the TDF/emtricitabine group vs 72.2 to 76.3 mL/min in the other NRTI groups, suggests high potential for selection bias. Some health care providers were likely to avoid TDF in patients with lower eGFR due to the data suggesting nephrotoxicity in other populations. Another limitation is that the reason for the missing laboratory values could not be determined. The TDF group had the greatest disparity in SCr data availability at baseline vs 36 months, with 39.5% concurrence with TDF alone compared with 50.0 to 63.6% in the other groups. Other treatment received outside the VHA system also could have influenced results.

Conclusions

This retrospective, multicenter, cohort study did not find a difference between TDF and other NRTIs for changes in renal function over time in patients with HBV infection without HIV. There was a fixed effect for time, ie, all NRTI groups showed some decline in renal function over time (−4.6 mL/min), but the effects were similar across groups. The results appear contrary to studies with comorbid HIV showing a decline in renal function with TDF, but present studies in HBV monotherapy have mixed results.

Further studies are needed to validate these results, as this and previous studies have several limitations. If these results are confirmed, a possible mechanism for these differences between patients with and without HIV should be examined. In addition, a study looking specifically at incidence of acute kidney injury rather than overall decline in renal function would add important data. If the results of this study are confirmed, there could be clinical implications in choice of agent with treatment of HBV monoinfection. This would add to the overall armament of medications available for chronic HBV infection and could create cost savings in certain situations if providers feel more comfortable continuing to use TDF instead of switching to the more expensive TAF.

Acknowledgments
Funding for this study was provided by the Veterans Health Administration.

Infection with hepatitis B virus (HBV) is associated with risk of potentially lethal, chronic infection and is a major public health problem. Infection from HBV has the potential to lead to liver failure, cirrhosis, and cancer.^1,2 Chronic HBV infection exists in as many as 2.2 million Americans, and in 2015 alone, HBV was estimated to be associated with 887,000 deaths worldwide.^1,3 Suppression of viral load is the basis of treatment, necessitating long-term use of medication for treatment.⁴ Nucleoside reverse transcriptase inhibitors (entecavir, lamivudine, telbivudine) and nucleotide reverse transcriptase inhibitors (adefovir, tenofovir), have improved the efficacy and tolerability of chronic HBV treatment compared with interferon-based agents.^4-7 However, concerns remain regarding long-term risk of nephrotoxicity, in particular with tenofovir disoproxil fumarate (TDF), which could lead to a limitation of safe and effective options for certain populations.^5,6,8 A newer formulation, tenofovir alafenamide fumarate (TAF), has improved the kidney risks, but expense remains a limiting factor for this agent.⁹

Nucleos(t)ide reverse transcriptase inhibitors (NRTIs) have demonstrated efficacy in reducing HBV viral load and other markers of improvement in chronic HBV, but entecavir and tenofovir have tended to demonstrate greater efficacy in clinical trials.^5-7 Several studies have suggested potential benefits of tenofovir-based treatment over other NRTIs, including greater viral load achievement compared with adefovir, efficacy in patients with previous failure of lamivudine or adefovir, and long-term efficacy in chronic HBV infection.^10-12 A 2019 systematic review suggests TDF and TAF are more effective than other NRTIs for achieving viral load suppression.¹³ Other NRTIs are not without their own risks, including mitochondrial dysfunction, mostly with lamivudine and telbivudine.⁴

Despite these data, guidelines have varied in their treatment recommendations in the context of chronic kidney disease partly due to variations in the evidence regarding nephrotoxicity.^7,14 Cohort studies and case reports have suggested association between TDF and acute kidney injury in patients with HIV infection as well as long-term reductions in kidney function.^15,16 In one study, 58% of patients treated with TDF did not return to baseline kidney function after an event of acute kidney injury.¹⁷ However, little data are available on whether this association exists for chronic HBV treatment in the absence of HIV infection. One retrospective analysis comparing TDF and entecavir in chronic HBV without HIV showed greater incidence of creatinine clearance < 60 mL/min with TDF but greater incidence of serum creatinine (SCr) ≥ 2.5 mg/dL in the entacavir group, making it difficult to reach a clear conclusion on risks.¹⁸ Other studies have either suffered from small cohorts with TDF or included patients with HIV coinfection.^19,20 Although a retrospective comparison of TDF and entecavir, randomly matched 1:2 to account for differences between groups, showed lower estimated glomerular filtration rate (eGFR) in the TDF group, more data are needed.²¹ Entecavir remains an option for many patient, but for those who have failed nucleosides, few options remain.

With the advantages available from TDF and the continued expense of TAF, more data regarding the risks of nephrotoxicity with TDF would be beneficial. The objective of this study was to compare treatment with TDF and other NRTIs in chronic HBV monoinfection to distinguish any differences in kidney function changes over time. With hopes of gathering enough data to distinguish between groups, information was gathered from across the Veterans Health Administration (VHA) system.

Methods

A nationwide, multicenter, retrospective, cohort study of veterans with HBV infection was conducted to compare the effects of various NRTIs on renal function. Patient were identified through the US Department of Veterans Affairs Corporate Data Warehouse (CDW), using data from July 1, 2005 to July 31, 2015. Patients were included who had positive HBV surface antigen (HBsAg) or newly prescribed NRTI. Multiple drug episodes could be included for each patient. That is, if a patient who had previously been included had another instance of a newly prescribed NRTI, this would be included in the analysis. Exclusion criteria were patients aged < 18 years, those with NRTI prescription for ≤ 1 month, and concurrent HIV infection. All patients with HBsAg were included for the study for increasing the sensitivity in gathering patients; however, those patients were included only if they received NRTI concurrent with the laboratory test results used for the primary endpoint (ie, SCr) to be included in the analysis.

How data are received from CDW bears some explanation. A basic way to understand the way data are received is that questions can be asked such as “for X population, at this point in time, was the patient on Y drug and what was the SCr value.” Therefore, inclusion and exclusion must first be specified to define the population, after which point certain data points can be received depending on the specifications made. For this reason, there is no way to determine, for example, whether a certain patient continued TDF use for the duration of the study, only at the defined points in time (described below) to receive the specific data.

For the patients included, information was retrieved from the first receipt of the NRTI prescription to 36 months after initiation. Baseline characteristics included age, sex, race, and ethnicity, and were defined at time of NRTI initiation. Values for SCr were compared at baseline, 3, 6, 12, 24, and 36 months after prescription of NRTI. The date of laboratory results was associated with the nearest date of comparison. Values for eGFR were determined by the modification of diet in renal disease equation. Values for eGFR are available in the CDW, whereas there is no direct means to calculate creatinine clearance with the available data, so eGFR was used for this study.

The primary endpoint was a change in eGFR in patients taking TDF after adjustment for time with the full cohort. Secondary analyses included the overall effect of time for the full cohort and change in renal function for each NRTI group. Mean and standard deviation for eGFR were determined for each NRTI group using the available data points. Analyses of the primary and secondary endpoints were completed using a linear mixed model with terms for time, to account for fixed effects, and specific NRTI used to account for random effects. A 2-sided α of .05 was used to determine statistical significance.

Results

A total of 413 drug episodes from 308 subjects met inclusion criteria for the study. Of these subjects, 229 were still living at the time of query. Most study participants were male (96%), the mean age was 62.1 years for males and 55.9 years for females; 49.5% were White and 39.7% were Black veterans (Table 1).

The NRTIs received by patients during the study period included TDF, TDF/emtricitabine, adefovir, entecavir, and lamivudine. No patients were on telbivudine. Formulations including TAF had not been approved by the US Food and Drug Administration (FDA) by the end of the study period, and as such were not found in the study.¹³ A plurality of participants received entecavir (94 of 223 at baseline), followed by TDF (n = 38) (Table 2). Of note, only 8 participants received TDF/emtricitabine at baseline. Differences were found between the groups in number of SCr data points available at 36 months vs baseline. The TDF group had the greatest reduction in data points available with 38 laboratory values at baseline vs 15 at 36 months (39.5% of baseline). From the available data, it is not possible to determine whether these represent medication discontinuations, missing values, lost to follow-up, or some other cause. Baseline eGFR was highest in the 2 TDF groups, with TDF alone at 77.7 mL/min (1.4-5.5 mL/min higher than the nontenofovir groups) and TDF/emtricitabine at 89.7 mL/min (13.4-17.5 mL/min higher than nontenofovir groups) (Table 3).

Discussion

This study demonstrated a decline in eGFR over time in a similar fashion for all NRTIs used in patients treated for HBV monoinfection, but no greater decline in renal function was found with use of TDF vs other NRTIs. A statistically significant decline in eGFR of −4.55 mL/min over the 36-month time frame of the study was demonstrated for the full cohort, but no statistically significant change in eGFR was found for any individual NRTI after accounting for the fixed effect of time. If TDF is not associated with additional risk of nephrotoxicity compared with other NRTIs, this could have important implications for treatment when considering the evidence that tenofovir-based treatment seems to be more effective than other medications for suppressing viral load.¹³

This result runs contrary to data in patients given NRTIs for HIV infection as well as a more recent cohort study in chronic HBV infectioin, which showed a statistically significant difference in kidney dysfunction between TDF and entecavir (-15.73 vs -5.96 mL/min/m², P < .001).^5-7,21 Possible mechanism for differences in response between HIV and HBV patients has not been elucidated, but the inherent risk of developing chronic kidney disease from HIV disease may play a role.²² The possibility remains that all NRTIs cause a degree of kidney impairment in patients treated for chronic HBV infection as evidenced by the statistically significant fixed effect for time in the present study. The cause of this effect is unknown but may be independently related to HBV infection or may be specific to NRTI therapy. No control group of patients not receiving NRTI therapy was included in this study, so conclusions cannot be drawn regarding whether all NRTIs are associated with decline in renal function in chronic HBV infection.

Limitations

Although this study did not detect a difference in change in eGFR between TDF and other NRTI treatments, it is possible that the length of data collection was not adequate to account for possible kidney injury from TDF. A study assessing renal tubular dysfunction in patients receiving adefovir or TDF showed a mean onset of dysfunction of 49 months.¹⁵ It is possible that participants in this study would go on to develop renal dysfunction in the future. This potential also was observed in a more recent retrospective cohort study in chronic HBV infection, which showed the greatest degree of decline in kidney function between 36 and 48 months (−11.87 to −15.73 mL/min/m² for the TDF group).²¹

The retrospective design created additional limitations. We attempted to account for some by using a matched cohort for the entecavir group, and there was no statistically significant difference between the groups in baseline characteristics. In HIV patients, a 10-year follow-up study continued to show decline in eGFR throughout the study, though the greatest degree of reduction occurred in the first year of the study.¹⁰ The higher baseline eGFR of the TDF recipients, 77.7 mL/min for the TDF alone group and 89.7 mL/min for the TDF/emtricitabine group vs 72.2 to 76.3 mL/min in the other NRTI groups, suggests high potential for selection bias. Some health care providers were likely to avoid TDF in patients with lower eGFR due to the data suggesting nephrotoxicity in other populations. Another limitation is that the reason for the missing laboratory values could not be determined. The TDF group had the greatest disparity in SCr data availability at baseline vs 36 months, with 39.5% concurrence with TDF alone compared with 50.0 to 63.6% in the other groups. Other treatment received outside the VHA system also could have influenced results.

Conclusions

This retrospective, multicenter, cohort study did not find a difference between TDF and other NRTIs for changes in renal function over time in patients with HBV infection without HIV. There was a fixed effect for time, ie, all NRTI groups showed some decline in renal function over time (−4.6 mL/min), but the effects were similar across groups. The results appear contrary to studies with comorbid HIV showing a decline in renal function with TDF, but present studies in HBV monotherapy have mixed results.

Further studies are needed to validate these results, as this and previous studies have several limitations. If these results are confirmed, a possible mechanism for these differences between patients with and without HIV should be examined. In addition, a study looking specifically at incidence of acute kidney injury rather than overall decline in renal function would add important data. If the results of this study are confirmed, there could be clinical implications in choice of agent with treatment of HBV monoinfection. This would add to the overall armament of medications available for chronic HBV infection and could create cost savings in certain situations if providers feel more comfortable continuing to use TDF instead of switching to the more expensive TAF.

Acknowledgments
Funding for this study was provided by the Veterans Health Administration.