Approximately 15% to 20% of inpatients develop significant adverse events, which are often preceded by a change in the patient's condition.[1, 2] Many hospitals utilize medical emergency teams (METs) to deliver prompt care for deteriorating patients. METs have an emerging role in end‐of‐life care; up to 35% of patients who die in the hospital with a do not resuscitate (DNR) order have a MET activation during the admission.[3] Thus, METs are well positioned to discuss preferences for cardiopulmonary resuscitation in a population at high risk for cardiac arrest.[4]

MET activations involve a change in code status in approximately 3% to 10% of cases.[5, 6, 7, 8] However, previous studies primarily involved hospitals in Australia, making the results difficult to generalize to other countries.[3, 6, 7] There may be important cultural differences in patient and clinician attitudes toward end‐of‐life care among different countries.[9] In addition, little is known about METs and hospital resources utilized in patients with MET‐implemented changes in code status. Anecdotal evidence suggests that MET activations addressing code status are more time consuming.[3] Conversely, by identifying patients unlikely to benefit from restorative care, METs may facilitate end‐of‐life comfort care, thereby reducing unplanned intensive care unit (ICU) admissions and hospital length of stay. Finally, preliminary evidence suggests that METs may improve the quality of end‐of‐life care.[10] However, the use of end‐of‐life resources, including inpatient palliative care consultation and hospice care following MET activation has not been well studied.

The purpose of our study was to examine the role of a US MET in end‐of‐life care. First, we assessed the proportion of MET calls that resulted in a new DNR order in a US hospital and compared this to previous reports in other countries. We also examined MET and hospital resource utilization in MET activations involving changes in code status by evaluating the duration of MET activations, the need for telemetry or ICU transfer, and hospital length of stay following MET activation. Finally, we explored the quality of end‐of‐life care in patients with MET‐implemented DNR orders by assessing the utilization of inpatient palliative care consultation and hospice care compared to patients with a preexisting DNR order.

MATERIALS AND METHODS

We conducted this study at Lahey Clinic, a 350‐bed academic, tertiary care center. The institutional review board approved this study and waived the need for informed consent. We performed a retrospective review of a prospectively collected MET registry. We included consecutive, adult (>18 years old) inpatient MET activations and excluded nonhospitalized patients. Data were recorded in an intranet registry at the time of the event by the MET nurse, including preexisting code status (full code or DNR), any change to the patient's code status (full code to DNR or DNR to full code) during the MET event, date of the event, disposition after the event (no transfer, transferred to ICU, or transfer to telemetry), and a description of MET interventions. The primary reason for the event was also recorded, including cardiovascular (systolic blood pressure <90 mm Hg, pulse <40 or >130 bpm), respiratory (respiratory rate <10 or >24 breaths per minute, need for noninvasive positive pressure ventilation, oxygen saturation <90%), neurologic (loss of consciousness, change in mental status, seizure, or suspected stroke), or clinical deterioration causing staff to become worried. MET activation occurs at our institution by utilizing a text paging system that records the date and time of the call, along with the patient's location. The event start time was recorded from the paging system. We considered the event stop time to be the time at which the patient is transferred to a different level of care (ie, ICU) or the MET members leave the bedside and transfer care back to the primary service. MET nurses recorded the duration of MET activation in the intranet database immediately following the activation. Data were collected from the medical record, including age, gender, race, medical insurance, religion, admission source (home, assisted living, rehabilitation, or other hospital), admission team (internal medicine or surgery), admission date and time, discharge or death date and time, disposition at discharge (home, rehabilitation, death with or without a DNR order, or hospice care), and admission diagnosis. We categorized patients discharged to a hospice bridge program as being discharged with hospice care. We also recorded whether or not code status was discussed at the time of admission. Our hospital policy requires that an advanced directives order (either full code or DNR) must be placed on all patients at the time of admission. However, we only considered a code status discussion to have occurred if there was explicit documentation of a code status discussion in the medical record at the time of hospital admission. Data from the time of hospital admission were collected to calculate a Charlson Comorbidity Index, a well‐validated predictor of mortality that uses a weighted sum of 17 medical conditions, with scores ranging from 0 to 37.[11] Higher scores indicate a greater burden of illness. We also recorded whether or not the inpatient palliative care service was consulted following MET activation by examining the medical record for a consult note. The inpatient palliative care service at our institution was implemented in 2005 and receives over 700 new inpatient consults per year. The most common reason for consultation is to assist the primary service with discussing goals of care and code status.

Our MET was established in 2005 and responds to approximately 30 to 40 events per 1000 admissions. There are 4 members on our MET, including a critical care nurse, a nursing supervisor, a respiratory therapist, and a team leader who, depending on a predetermined schedule, is either an attending hospital medicine physician, an attending critical care physician, a critical care training physician, or a critical care physician assistant.

The data were transferred from the intranet registry to an Excel (Microsoft Corp., Redmond, WA) spreadsheet for statistical analysis. The primary outcome was the proportion of activations resulting in a MET‐implemented DNR order. Secondary outcomes included the duration of MET activation, need for transfer to telemetry or the ICU, hospital length of stay following MET activation (time from the end of the MET activation to hospital discharge or death), and the frequency with which inpatient palliative care consultation and outpatient hospice care were utilized. For repeat MET activations in a single patient, we considered each MET activation as a separate event as the code status could potentially change more than once. We used SAS version 9.2 for Windows (SAS Institute Inc., Cary, NC) statistical analysis software for data analysis. The 2 method was used for categorical variables, and either a 2‐sample t test or Wilcoxon rank sum test was utilized for continuous variables. A P value 0.05 was considered significant.

RESULTS

We observed 1156 MET activations in 998 patients. The mean age was 67 years and 57% (565/998) were male (Table 1). The mean Charlson Comorbidity Index was 5.4. Most patients were admitted from home (76%, 760/998), to a medical service (72%, 720/998), and to a teaching service (73%, 732/998). Sepsis (11%, 109/998) and trauma (11%, 105/998) were the most common admission diagnoses. A cardiovascular abnormality was the most common (35%, 399/1156) reason for activation. A code status discussion was documented on admission in 44% (440/998) of all patients.

MET activation resulted in a DNR order in 5% (58/1156) of cases (Figure 1). In activations involving a change in code status, 21% (15/73) were changed from DNR to full code. When compared to patients with a preexisting DNR order, patients with a MET‐implemented DNR order where younger (70 vs 81 years, P<0.0001), more commonly admitted from home (60% vs 44%, P=0.04), less frequently from a nursing home (1% vs 9%, P<0.01), and had a lower Charlson index (5.7 vs 7.7, P<0.001) (Table 1). Moderate to severe liver disease was more common in patients with a MET‐implemented DNR order (10% vs 0%, P=0.001). Admission diagnoses were similar between patients with a preexisting DNR and a MET‐implemented DNR order (Table 1).

Figure 1

The median time spent on activations with a change in code status was significantly longer than activations without a change (66 vs 60 minutes, P=0.05). The rates of telemetry (6% vs 3%, P=0.24) and ICU transfer (40% vs 41%, P=0.8) were similar between patients with a change in code status and patients without a change (Table 2). Patients with a MET‐implemented DNR order were more frequently transferred to the ICU than patients with a preexisting DNR order (36% vs 17%, P<0.01). The median hospital length of stay following MET activation was shorter in patients with a change in code status compared to patients with no change (3 vs 5 days, P<0.0001).

The inpatient mortality was 17% (165/998). Most patients who died had the focus of care changed to comfort measures only (88%, 146/165). When examining the group of patients who died in the hospital with comfort care, we found that 58% (92/159) were transferred to the ICU following the MET call, 5% (8/159) were changed to comfort care during the MET call, and 18% (29/159) had a palliative care consult. We also observed that 16% (25/159) patients who died with comfort care were made DNR during MET activation. The inpatient mortality was significantly higher in patients with a change in code status compared to patients with no change in code status (44% vs 14%, 133/926, P<0.0001). Patients with a MET‐implemented DNR order had a higher inpatient mortality than patients with a preexisting DNR (43% vs 27%, P=0.04). Twenty‐five patients with a MET‐implemented DNR order died in the hospital. When examining a subgroup of patients who required end‐of‐life care (died or discharged from the hospital with hospice), we found patients with a MET‐implemented DNR order were less likely to be discharged with hospice care than patients with a preexisting DNR (4% vs 29%, P=0.01). There was no difference in the use of inpatient palliative care consultation at the end of life in patients with a preexisting DNR versus MET‐implemented DNR order (20% vs 12%, P=0.39). Patients with a MET‐implemented DNR order also had a significantly shorter median time from implementation of comfort care orders to death or discharge with hospice compared to patients with a preexisting DNR order (7 hours, interquartile range [IQR], 416 hours vs 22 hours, IQR 939 hours).

DISCUSSION

We observed a MET‐implemented DNR order in 5% of activations. Little is known about the role of METs in end‐of‐life discussions in the United States, and past experience has primarily come from Australian hospitals. Important differences in end‐of‐life care exist among different countries, particularly with regard to placing limitations on treatment.[9] Our observed rate is similar to the 3% to 10% rate of MET‐implemented DNR orders in previous reports worldwide.[3, 5, 8, 12, 13] Recent data from the United States suggest that METs initiate a DNR order in 28% of cases.[14] However, most DNR orders in that study were placed in the ICU days to weeks after MET activation, likely accounting for the high DNR rate. Our data add to the growing body of evidence that METs play an important role in end‐of‐life discussions among different countries throughout the world, including the United States.

To our knowledge, no prior study has evaluated the impact of code status discussions on MET resource utilization. Our MET spent 6 minutes longer on activations involving a change in code status when compared to activations with no changes made to code status. Presumably, some of this time was spent discussing goals of care. In our opinion, the additional time spent on these activations was invaluable, particularly when considering MET‐initiated end‐of‐life discussions may have prevented several unwanted resuscitations (25 patients died with a MET‐implemented DNR order). Interestingly, less than half of patients had code status discussions at the time of hospital admission. This finding suggests that clinicians could be more vigilant about discussing preferences for resuscitation at the time of admission in patients at risk for clinical deterioration. We suspect that in some cases code status discussions may have occurred between the patient and the primary service later in the patient's hospitalization, which were not captured in our study.

Surprisingly, when examining the use of hospital resources, we found no difference in the rate of unplanned ICU transfer in patients with a change in code status. In fact, we observed a higher rate of ICU transfer in patients with a MET‐implemented DNR order compared to those with a preexisting DNR order (36% vs 17%). These results were at odds with our hypothesis of a lower rate of ICU transfer in patients with MET‐implemented limitations in care. When compared to patients with a preexisting DNR order, patients with a MET‐implemented DNR order were younger, more commonly admitted from home, and had a lower Charlson index. Despite evidence of a lower burden of chronic illness and younger age, patients with a MET‐implemented DNR order had higher inpatient mortality than patients with a preexisting DNR order (43% vs 27%), suggesting an acute and rapidly progressive disease process. These observations may have compelled the MET to advocate for aggressive ICU‐level care in patients with a MET‐implemented DNR order. Another possible explanation for the relatively high rate of ICU transfer is that the MET is, in part, led by ICU staff. Thus, our MET may have made the decision to transfer the patient to the ICU and then subsequently initiated end‐of‐life discussions only after taking ownership of the patient. Furthermore, almost 20% of MET‐implemented changes to code status involved reversing status from DNR to full code. These data suggest that METs are not merely serving as a resource to review code status, but rather providing intensive treatment for acutely ill patients and simultaneously initiating end‐of‐life discussions in a population with a high inpatient mortality rate. The practice pattern observed in our study of transferring patients to the ICU for a trial of intensive therapy at the end of life is consistent with the overall trend in the United States for increased inpatient treatment intensity at the end of life.[14, 15, 16]

Our data suggest that the increased use of ICU resources in patients with a MET‐implemented DNR may be balanced by a shorter hospital length of stay following MET activation. In a multicenter observational study, Jones et al. found hospital length of stay to be similar in patients with and without a MET‐implemented limitation of medical therapy.[3] The authors did not examine length of stay specifically in patients with a DNR order, but rather examined patients with any limitation in medical therapy, including not for ICU admission. We suspect that the shorter length of stay following MET activation in our study was related to the fact that patients with a change in code status had a significantly higher inpatient mortality.

We observed several interesting findings with regard to end‐of‐life care following MET‐implemented DNR orders. First, the inpatient mortality in this population was remarkably high at 43%, compared to 27% in patients with a preexisting DNR order. Interestingly, there was no difference in the rate of palliative care consultation between the 2 groups despite the fact that all 25 patients who died following a MET‐implemented DNR order did so with a comfort measures only order. We also found that patients with a preexisting DNR also had a higher rate of discharge with hospice compared to patients with a MET‐implemented DNR order (29% vs 4%). Thus, our data suggest that inpatient palliative care consultation and hospice services are not resources that are routinely utilized in patients with MET‐initiated DNR orders. It may be the case that the acuity and severity of illness or patient preferences may have precluded the possibility of discharging some patients in our study home with hospice care or implementing comfort care earlier in the hospital course. Patients with MET‐implemented DNR orders were younger, had fewer comorbidities, and died sooner after comfort care orders were written. The overall rate of comfort care provided to patients who died was high at 88%. We have an inpatient comfort measures only order set at our hospital, which may account for the large proportion of patients receiving comfort care at the end of life. In addition, this order set may also help to improve the quality of end‐of‐life care and thus limit the need for palliative care consultation to some extent. However, we found that patients with a MET‐implemented DNR order had a shorter time from comfort care orders to death than patients with a preexisting DNR order. This finding suggests that patients with MET‐implemented DNR orders may have had comfort care implemented relatively late in the course of illness and had less‐than‐optimal end‐of‐life care. Vazquez et al. reported improved quality of end‐of‐life care after implementation of a MET.[10] However, an inpatient palliative care service was not available in that study, and it is not clear whether or not a comfort care order set was available. Evidence suggests that utilization of palliative care resources improves end‐of‐life care in the ICU.[17, 18, 19] We found that more than half of patients who died with comfort care in the hospital did so after being transferred to the ICU for a trial of aggressive care, suggesting that this population may have benefited from more involvement of our palliative care service. In summary, our data on end‐of‐life care following MET activation suggest that the METs are able to take advantage of an opportunity to identify patients who would not want resuscitation efforts because of personal preferences or futility of treatments. However, our surrogate measures of the quality of end‐of‐life care suggest that patients with MET‐implemented DNR orders may benefit from coordinated care with inpatient palliative care services, timelier implementation of comfort care orders, and possibly increased referrals for hospice care to help improve the quality of end‐of‐life care in this population.

Our study is subject to a number of limitations. This was a single‐center study, making the results difficult to generalize. The retrospective nature of our study makes it subject to the limitations inherent in this study design, including bias and confounding. The duration of MET activation was difficult to accurately and objectively measure and is subject to reporting bias. The event stop time, in particular, was subjectively measured by the MET nurse and is difficult to accurately assess, because MET members occasionally leave the bedside and return to reevaluate the response to therapy. We tried to account for this by clearly defining the MET stop time as the point at which MET members leave the bedside and transfer care back to the primary service or physically transfer the patient to a higher level of care. It also bears mentioning that the nurses performing data entry were not aware of the study hypothesis at the time of data entry. Despite including over 1100 MET calls in our analysis, the number of patients with a MET‐implemented DNR order was relatively small, which may have limited our ability to detect differences among subgroups during our analysis. We also did not document the clinical circumstances surrounding the MET‐implemented DNR order. Although we hypothesized that these patients had a higher mortality due to a higher acuity of illness, we were unable to support this hypothesis with the data available in our retrospective study. We did not record which providers were involved in code status discussions and the exact amount time spent on these discussions, making it difficult to accurately quantify the MET resources utilized on the calls. Our MET works closely with the patient's primary service, and it is possible that some of the changes to code status were implemented by the primary service and not MET providers. The patient's primary service may have a preexisting relationship with the patient and would be in a better position to discuss goals of care than MET providers who have had no prior relationship with the patient. However, even in this case, the clinical deterioration prompting MET activation was likely the event that triggered end‐of‐life discussions. Prospective studies would be helpful not only to identify the individuals involved in code status discussions during MET activations, but also to objectively measure the time spent on such discussions. Finally, our study population consisted primarily of Caucasian patients. Preferences for end‐of‐life care may differ among socioeconomic and ethnic groups, thus limiting the generalizability of our study findings.[20, 21]

In conclusion, we found the rate of MET‐implemented DNR orders in the United States to be similar to that of previous reports in other countries. MET events involving a change in code status are associated with increased utilization of MET and ICU resources, but a shorter hospital length of stay. Despite a high inpatient mortality rate, patients with a MET‐implemented DNR had a relatively low utilization of end‐of‐life resources, including palliative care and home hospice services. Coordinated care between METs and palliative care may help to improve of end‐of‐life care in patients with a change in code status following MET activation.

Approximately 15% to 20% of inpatients develop significant adverse events, which are often preceded by a change in the patient's condition.[1, 2] Many hospitals utilize medical emergency teams (METs) to deliver prompt care for deteriorating patients. METs have an emerging role in end‐of‐life care; up to 35% of patients who die in the hospital with a do not resuscitate (DNR) order have a MET activation during the admission.[3] Thus, METs are well positioned to discuss preferences for cardiopulmonary resuscitation in a population at high risk for cardiac arrest.[4]

MET activations involve a change in code status in approximately 3% to 10% of cases.[5, 6, 7, 8] However, previous studies primarily involved hospitals in Australia, making the results difficult to generalize to other countries.[3, 6, 7] There may be important cultural differences in patient and clinician attitudes toward end‐of‐life care among different countries.[9] In addition, little is known about METs and hospital resources utilized in patients with MET‐implemented changes in code status. Anecdotal evidence suggests that MET activations addressing code status are more time consuming.[3] Conversely, by identifying patients unlikely to benefit from restorative care, METs may facilitate end‐of‐life comfort care, thereby reducing unplanned intensive care unit (ICU) admissions and hospital length of stay. Finally, preliminary evidence suggests that METs may improve the quality of end‐of‐life care.[10] However, the use of end‐of‐life resources, including inpatient palliative care consultation and hospice care following MET activation has not been well studied.

The purpose of our study was to examine the role of a US MET in end‐of‐life care. First, we assessed the proportion of MET calls that resulted in a new DNR order in a US hospital and compared this to previous reports in other countries. We also examined MET and hospital resource utilization in MET activations involving changes in code status by evaluating the duration of MET activations, the need for telemetry or ICU transfer, and hospital length of stay following MET activation. Finally, we explored the quality of end‐of‐life care in patients with MET‐implemented DNR orders by assessing the utilization of inpatient palliative care consultation and hospice care compared to patients with a preexisting DNR order.

MATERIALS AND METHODS

We conducted this study at Lahey Clinic, a 350‐bed academic, tertiary care center. The institutional review board approved this study and waived the need for informed consent. We performed a retrospective review of a prospectively collected MET registry. We included consecutive, adult (>18 years old) inpatient MET activations and excluded nonhospitalized patients. Data were recorded in an intranet registry at the time of the event by the MET nurse, including preexisting code status (full code or DNR), any change to the patient's code status (full code to DNR or DNR to full code) during the MET event, date of the event, disposition after the event (no transfer, transferred to ICU, or transfer to telemetry), and a description of MET interventions. The primary reason for the event was also recorded, including cardiovascular (systolic blood pressure <90 mm Hg, pulse <40 or >130 bpm), respiratory (respiratory rate <10 or >24 breaths per minute, need for noninvasive positive pressure ventilation, oxygen saturation <90%), neurologic (loss of consciousness, change in mental status, seizure, or suspected stroke), or clinical deterioration causing staff to become worried. MET activation occurs at our institution by utilizing a text paging system that records the date and time of the call, along with the patient's location. The event start time was recorded from the paging system. We considered the event stop time to be the time at which the patient is transferred to a different level of care (ie, ICU) or the MET members leave the bedside and transfer care back to the primary service. MET nurses recorded the duration of MET activation in the intranet database immediately following the activation. Data were collected from the medical record, including age, gender, race, medical insurance, religion, admission source (home, assisted living, rehabilitation, or other hospital), admission team (internal medicine or surgery), admission date and time, discharge or death date and time, disposition at discharge (home, rehabilitation, death with or without a DNR order, or hospice care), and admission diagnosis. We categorized patients discharged to a hospice bridge program as being discharged with hospice care. We also recorded whether or not code status was discussed at the time of admission. Our hospital policy requires that an advanced directives order (either full code or DNR) must be placed on all patients at the time of admission. However, we only considered a code status discussion to have occurred if there was explicit documentation of a code status discussion in the medical record at the time of hospital admission. Data from the time of hospital admission were collected to calculate a Charlson Comorbidity Index, a well‐validated predictor of mortality that uses a weighted sum of 17 medical conditions, with scores ranging from 0 to 37.[11] Higher scores indicate a greater burden of illness. We also recorded whether or not the inpatient palliative care service was consulted following MET activation by examining the medical record for a consult note. The inpatient palliative care service at our institution was implemented in 2005 and receives over 700 new inpatient consults per year. The most common reason for consultation is to assist the primary service with discussing goals of care and code status.

Our MET was established in 2005 and responds to approximately 30 to 40 events per 1000 admissions. There are 4 members on our MET, including a critical care nurse, a nursing supervisor, a respiratory therapist, and a team leader who, depending on a predetermined schedule, is either an attending hospital medicine physician, an attending critical care physician, a critical care training physician, or a critical care physician assistant.

The data were transferred from the intranet registry to an Excel (Microsoft Corp., Redmond, WA) spreadsheet for statistical analysis. The primary outcome was the proportion of activations resulting in a MET‐implemented DNR order. Secondary outcomes included the duration of MET activation, need for transfer to telemetry or the ICU, hospital length of stay following MET activation (time from the end of the MET activation to hospital discharge or death), and the frequency with which inpatient palliative care consultation and outpatient hospice care were utilized. For repeat MET activations in a single patient, we considered each MET activation as a separate event as the code status could potentially change more than once. We used SAS version 9.2 for Windows (SAS Institute Inc., Cary, NC) statistical analysis software for data analysis. The 2 method was used for categorical variables, and either a 2‐sample t test or Wilcoxon rank sum test was utilized for continuous variables. A P value 0.05 was considered significant.

RESULTS

We observed 1156 MET activations in 998 patients. The mean age was 67 years and 57% (565/998) were male (Table 1). The mean Charlson Comorbidity Index was 5.4. Most patients were admitted from home (76%, 760/998), to a medical service (72%, 720/998), and to a teaching service (73%, 732/998). Sepsis (11%, 109/998) and trauma (11%, 105/998) were the most common admission diagnoses. A cardiovascular abnormality was the most common (35%, 399/1156) reason for activation. A code status discussion was documented on admission in 44% (440/998) of all patients.

MET activation resulted in a DNR order in 5% (58/1156) of cases (Figure 1). In activations involving a change in code status, 21% (15/73) were changed from DNR to full code. When compared to patients with a preexisting DNR order, patients with a MET‐implemented DNR order where younger (70 vs 81 years, P<0.0001), more commonly admitted from home (60% vs 44%, P=0.04), less frequently from a nursing home (1% vs 9%, P<0.01), and had a lower Charlson index (5.7 vs 7.7, P<0.001) (Table 1). Moderate to severe liver disease was more common in patients with a MET‐implemented DNR order (10% vs 0%, P=0.001). Admission diagnoses were similar between patients with a preexisting DNR and a MET‐implemented DNR order (Table 1).

Figure 1

The median time spent on activations with a change in code status was significantly longer than activations without a change (66 vs 60 minutes, P=0.05). The rates of telemetry (6% vs 3%, P=0.24) and ICU transfer (40% vs 41%, P=0.8) were similar between patients with a change in code status and patients without a change (Table 2). Patients with a MET‐implemented DNR order were more frequently transferred to the ICU than patients with a preexisting DNR order (36% vs 17%, P<0.01). The median hospital length of stay following MET activation was shorter in patients with a change in code status compared to patients with no change (3 vs 5 days, P<0.0001).

The inpatient mortality was 17% (165/998). Most patients who died had the focus of care changed to comfort measures only (88%, 146/165). When examining the group of patients who died in the hospital with comfort care, we found that 58% (92/159) were transferred to the ICU following the MET call, 5% (8/159) were changed to comfort care during the MET call, and 18% (29/159) had a palliative care consult. We also observed that 16% (25/159) patients who died with comfort care were made DNR during MET activation. The inpatient mortality was significantly higher in patients with a change in code status compared to patients with no change in code status (44% vs 14%, 133/926, P<0.0001). Patients with a MET‐implemented DNR order had a higher inpatient mortality than patients with a preexisting DNR (43% vs 27%, P=0.04). Twenty‐five patients with a MET‐implemented DNR order died in the hospital. When examining a subgroup of patients who required end‐of‐life care (died or discharged from the hospital with hospice), we found patients with a MET‐implemented DNR order were less likely to be discharged with hospice care than patients with a preexisting DNR (4% vs 29%, P=0.01). There was no difference in the use of inpatient palliative care consultation at the end of life in patients with a preexisting DNR versus MET‐implemented DNR order (20% vs 12%, P=0.39). Patients with a MET‐implemented DNR order also had a significantly shorter median time from implementation of comfort care orders to death or discharge with hospice compared to patients with a preexisting DNR order (7 hours, interquartile range [IQR], 416 hours vs 22 hours, IQR 939 hours).

DISCUSSION

We observed a MET‐implemented DNR order in 5% of activations. Little is known about the role of METs in end‐of‐life discussions in the United States, and past experience has primarily come from Australian hospitals. Important differences in end‐of‐life care exist among different countries, particularly with regard to placing limitations on treatment.[9] Our observed rate is similar to the 3% to 10% rate of MET‐implemented DNR orders in previous reports worldwide.[3, 5, 8, 12, 13] Recent data from the United States suggest that METs initiate a DNR order in 28% of cases.[14] However, most DNR orders in that study were placed in the ICU days to weeks after MET activation, likely accounting for the high DNR rate. Our data add to the growing body of evidence that METs play an important role in end‐of‐life discussions among different countries throughout the world, including the United States.

To our knowledge, no prior study has evaluated the impact of code status discussions on MET resource utilization. Our MET spent 6 minutes longer on activations involving a change in code status when compared to activations with no changes made to code status. Presumably, some of this time was spent discussing goals of care. In our opinion, the additional time spent on these activations was invaluable, particularly when considering MET‐initiated end‐of‐life discussions may have prevented several unwanted resuscitations (25 patients died with a MET‐implemented DNR order). Interestingly, less than half of patients had code status discussions at the time of hospital admission. This finding suggests that clinicians could be more vigilant about discussing preferences for resuscitation at the time of admission in patients at risk for clinical deterioration. We suspect that in some cases code status discussions may have occurred between the patient and the primary service later in the patient's hospitalization, which were not captured in our study.

Surprisingly, when examining the use of hospital resources, we found no difference in the rate of unplanned ICU transfer in patients with a change in code status. In fact, we observed a higher rate of ICU transfer in patients with a MET‐implemented DNR order compared to those with a preexisting DNR order (36% vs 17%). These results were at odds with our hypothesis of a lower rate of ICU transfer in patients with MET‐implemented limitations in care. When compared to patients with a preexisting DNR order, patients with a MET‐implemented DNR order were younger, more commonly admitted from home, and had a lower Charlson index. Despite evidence of a lower burden of chronic illness and younger age, patients with a MET‐implemented DNR order had higher inpatient mortality than patients with a preexisting DNR order (43% vs 27%), suggesting an acute and rapidly progressive disease process. These observations may have compelled the MET to advocate for aggressive ICU‐level care in patients with a MET‐implemented DNR order. Another possible explanation for the relatively high rate of ICU transfer is that the MET is, in part, led by ICU staff. Thus, our MET may have made the decision to transfer the patient to the ICU and then subsequently initiated end‐of‐life discussions only after taking ownership of the patient. Furthermore, almost 20% of MET‐implemented changes to code status involved reversing status from DNR to full code. These data suggest that METs are not merely serving as a resource to review code status, but rather providing intensive treatment for acutely ill patients and simultaneously initiating end‐of‐life discussions in a population with a high inpatient mortality rate. The practice pattern observed in our study of transferring patients to the ICU for a trial of intensive therapy at the end of life is consistent with the overall trend in the United States for increased inpatient treatment intensity at the end of life.[14, 15, 16]

Our data suggest that the increased use of ICU resources in patients with a MET‐implemented DNR may be balanced by a shorter hospital length of stay following MET activation. In a multicenter observational study, Jones et al. found hospital length of stay to be similar in patients with and without a MET‐implemented limitation of medical therapy.[3] The authors did not examine length of stay specifically in patients with a DNR order, but rather examined patients with any limitation in medical therapy, including not for ICU admission. We suspect that the shorter length of stay following MET activation in our study was related to the fact that patients with a change in code status had a significantly higher inpatient mortality.

We observed several interesting findings with regard to end‐of‐life care following MET‐implemented DNR orders. First, the inpatient mortality in this population was remarkably high at 43%, compared to 27% in patients with a preexisting DNR order. Interestingly, there was no difference in the rate of palliative care consultation between the 2 groups despite the fact that all 25 patients who died following a MET‐implemented DNR order did so with a comfort measures only order. We also found that patients with a preexisting DNR also had a higher rate of discharge with hospice compared to patients with a MET‐implemented DNR order (29% vs 4%). Thus, our data suggest that inpatient palliative care consultation and hospice services are not resources that are routinely utilized in patients with MET‐initiated DNR orders. It may be the case that the acuity and severity of illness or patient preferences may have precluded the possibility of discharging some patients in our study home with hospice care or implementing comfort care earlier in the hospital course. Patients with MET‐implemented DNR orders were younger, had fewer comorbidities, and died sooner after comfort care orders were written. The overall rate of comfort care provided to patients who died was high at 88%. We have an inpatient comfort measures only order set at our hospital, which may account for the large proportion of patients receiving comfort care at the end of life. In addition, this order set may also help to improve the quality of end‐of‐life care and thus limit the need for palliative care consultation to some extent. However, we found that patients with a MET‐implemented DNR order had a shorter time from comfort care orders to death than patients with a preexisting DNR order. This finding suggests that patients with MET‐implemented DNR orders may have had comfort care implemented relatively late in the course of illness and had less‐than‐optimal end‐of‐life care. Vazquez et al. reported improved quality of end‐of‐life care after implementation of a MET.[10] However, an inpatient palliative care service was not available in that study, and it is not clear whether or not a comfort care order set was available. Evidence suggests that utilization of palliative care resources improves end‐of‐life care in the ICU.[17, 18, 19] We found that more than half of patients who died with comfort care in the hospital did so after being transferred to the ICU for a trial of aggressive care, suggesting that this population may have benefited from more involvement of our palliative care service. In summary, our data on end‐of‐life care following MET activation suggest that the METs are able to take advantage of an opportunity to identify patients who would not want resuscitation efforts because of personal preferences or futility of treatments. However, our surrogate measures of the quality of end‐of‐life care suggest that patients with MET‐implemented DNR orders may benefit from coordinated care with inpatient palliative care services, timelier implementation of comfort care orders, and possibly increased referrals for hospice care to help improve the quality of end‐of‐life care in this population.

Our study is subject to a number of limitations. This was a single‐center study, making the results difficult to generalize. The retrospective nature of our study makes it subject to the limitations inherent in this study design, including bias and confounding. The duration of MET activation was difficult to accurately and objectively measure and is subject to reporting bias. The event stop time, in particular, was subjectively measured by the MET nurse and is difficult to accurately assess, because MET members occasionally leave the bedside and return to reevaluate the response to therapy. We tried to account for this by clearly defining the MET stop time as the point at which MET members leave the bedside and transfer care back to the primary service or physically transfer the patient to a higher level of care. It also bears mentioning that the nurses performing data entry were not aware of the study hypothesis at the time of data entry. Despite including over 1100 MET calls in our analysis, the number of patients with a MET‐implemented DNR order was relatively small, which may have limited our ability to detect differences among subgroups during our analysis. We also did not document the clinical circumstances surrounding the MET‐implemented DNR order. Although we hypothesized that these patients had a higher mortality due to a higher acuity of illness, we were unable to support this hypothesis with the data available in our retrospective study. We did not record which providers were involved in code status discussions and the exact amount time spent on these discussions, making it difficult to accurately quantify the MET resources utilized on the calls. Our MET works closely with the patient's primary service, and it is possible that some of the changes to code status were implemented by the primary service and not MET providers. The patient's primary service may have a preexisting relationship with the patient and would be in a better position to discuss goals of care than MET providers who have had no prior relationship with the patient. However, even in this case, the clinical deterioration prompting MET activation was likely the event that triggered end‐of‐life discussions. Prospective studies would be helpful not only to identify the individuals involved in code status discussions during MET activations, but also to objectively measure the time spent on such discussions. Finally, our study population consisted primarily of Caucasian patients. Preferences for end‐of‐life care may differ among socioeconomic and ethnic groups, thus limiting the generalizability of our study findings.[20, 21]

In conclusion, we found the rate of MET‐implemented DNR orders in the United States to be similar to that of previous reports in other countries. MET events involving a change in code status are associated with increased utilization of MET and ICU resources, but a shorter hospital length of stay. Despite a high inpatient mortality rate, patients with a MET‐implemented DNR had a relatively low utilization of end‐of‐life resources, including palliative care and home hospice services. Coordinated care between METs and palliative care may help to improve of end‐of‐life care in patients with a change in code status following MET activation.

Approximately 15% to 20% of inpatients develop significant adverse events, which are often preceded by a change in the patient's condition.[1, 2] Many hospitals utilize medical emergency teams (METs) to deliver prompt care for deteriorating patients. METs have an emerging role in end‐of‐life care; up to 35% of patients who die in the hospital with a do not resuscitate (DNR) order have a MET activation during the admission.[3] Thus, METs are well positioned to discuss preferences for cardiopulmonary resuscitation in a population at high risk for cardiac arrest.[4]

MET activations involve a change in code status in approximately 3% to 10% of cases.[5, 6, 7, 8] However, previous studies primarily involved hospitals in Australia, making the results difficult to generalize to other countries.[3, 6, 7] There may be important cultural differences in patient and clinician attitudes toward end‐of‐life care among different countries.[9] In addition, little is known about METs and hospital resources utilized in patients with MET‐implemented changes in code status. Anecdotal evidence suggests that MET activations addressing code status are more time consuming.[3] Conversely, by identifying patients unlikely to benefit from restorative care, METs may facilitate end‐of‐life comfort care, thereby reducing unplanned intensive care unit (ICU) admissions and hospital length of stay. Finally, preliminary evidence suggests that METs may improve the quality of end‐of‐life care.[10] However, the use of end‐of‐life resources, including inpatient palliative care consultation and hospice care following MET activation has not been well studied.

The purpose of our study was to examine the role of a US MET in end‐of‐life care. First, we assessed the proportion of MET calls that resulted in a new DNR order in a US hospital and compared this to previous reports in other countries. We also examined MET and hospital resource utilization in MET activations involving changes in code status by evaluating the duration of MET activations, the need for telemetry or ICU transfer, and hospital length of stay following MET activation. Finally, we explored the quality of end‐of‐life care in patients with MET‐implemented DNR orders by assessing the utilization of inpatient palliative care consultation and hospice care compared to patients with a preexisting DNR order.

MATERIALS AND METHODS

We conducted this study at Lahey Clinic, a 350‐bed academic, tertiary care center. The institutional review board approved this study and waived the need for informed consent. We performed a retrospective review of a prospectively collected MET registry. We included consecutive, adult (>18 years old) inpatient MET activations and excluded nonhospitalized patients. Data were recorded in an intranet registry at the time of the event by the MET nurse, including preexisting code status (full code or DNR), any change to the patient's code status (full code to DNR or DNR to full code) during the MET event, date of the event, disposition after the event (no transfer, transferred to ICU, or transfer to telemetry), and a description of MET interventions. The primary reason for the event was also recorded, including cardiovascular (systolic blood pressure <90 mm Hg, pulse <40 or >130 bpm), respiratory (respiratory rate <10 or >24 breaths per minute, need for noninvasive positive pressure ventilation, oxygen saturation <90%), neurologic (loss of consciousness, change in mental status, seizure, or suspected stroke), or clinical deterioration causing staff to become worried. MET activation occurs at our institution by utilizing a text paging system that records the date and time of the call, along with the patient's location. The event start time was recorded from the paging system. We considered the event stop time to be the time at which the patient is transferred to a different level of care (ie, ICU) or the MET members leave the bedside and transfer care back to the primary service. MET nurses recorded the duration of MET activation in the intranet database immediately following the activation. Data were collected from the medical record, including age, gender, race, medical insurance, religion, admission source (home, assisted living, rehabilitation, or other hospital), admission team (internal medicine or surgery), admission date and time, discharge or death date and time, disposition at discharge (home, rehabilitation, death with or without a DNR order, or hospice care), and admission diagnosis. We categorized patients discharged to a hospice bridge program as being discharged with hospice care. We also recorded whether or not code status was discussed at the time of admission. Our hospital policy requires that an advanced directives order (either full code or DNR) must be placed on all patients at the time of admission. However, we only considered a code status discussion to have occurred if there was explicit documentation of a code status discussion in the medical record at the time of hospital admission. Data from the time of hospital admission were collected to calculate a Charlson Comorbidity Index, a well‐validated predictor of mortality that uses a weighted sum of 17 medical conditions, with scores ranging from 0 to 37.[11] Higher scores indicate a greater burden of illness. We also recorded whether or not the inpatient palliative care service was consulted following MET activation by examining the medical record for a consult note. The inpatient palliative care service at our institution was implemented in 2005 and receives over 700 new inpatient consults per year. The most common reason for consultation is to assist the primary service with discussing goals of care and code status.

Our MET was established in 2005 and responds to approximately 30 to 40 events per 1000 admissions. There are 4 members on our MET, including a critical care nurse, a nursing supervisor, a respiratory therapist, and a team leader who, depending on a predetermined schedule, is either an attending hospital medicine physician, an attending critical care physician, a critical care training physician, or a critical care physician assistant.

The data were transferred from the intranet registry to an Excel (Microsoft Corp., Redmond, WA) spreadsheet for statistical analysis. The primary outcome was the proportion of activations resulting in a MET‐implemented DNR order. Secondary outcomes included the duration of MET activation, need for transfer to telemetry or the ICU, hospital length of stay following MET activation (time from the end of the MET activation to hospital discharge or death), and the frequency with which inpatient palliative care consultation and outpatient hospice care were utilized. For repeat MET activations in a single patient, we considered each MET activation as a separate event as the code status could potentially change more than once. We used SAS version 9.2 for Windows (SAS Institute Inc., Cary, NC) statistical analysis software for data analysis. The 2 method was used for categorical variables, and either a 2‐sample t test or Wilcoxon rank sum test was utilized for continuous variables. A P value 0.05 was considered significant.

RESULTS

We observed 1156 MET activations in 998 patients. The mean age was 67 years and 57% (565/998) were male (Table 1). The mean Charlson Comorbidity Index was 5.4. Most patients were admitted from home (76%, 760/998), to a medical service (72%, 720/998), and to a teaching service (73%, 732/998). Sepsis (11%, 109/998) and trauma (11%, 105/998) were the most common admission diagnoses. A cardiovascular abnormality was the most common (35%, 399/1156) reason for activation. A code status discussion was documented on admission in 44% (440/998) of all patients.

MET activation resulted in a DNR order in 5% (58/1156) of cases (Figure 1). In activations involving a change in code status, 21% (15/73) were changed from DNR to full code. When compared to patients with a preexisting DNR order, patients with a MET‐implemented DNR order where younger (70 vs 81 years, P<0.0001), more commonly admitted from home (60% vs 44%, P=0.04), less frequently from a nursing home (1% vs 9%, P<0.01), and had a lower Charlson index (5.7 vs 7.7, P<0.001) (Table 1). Moderate to severe liver disease was more common in patients with a MET‐implemented DNR order (10% vs 0%, P=0.001). Admission diagnoses were similar between patients with a preexisting DNR and a MET‐implemented DNR order (Table 1).

Figure 1

The median time spent on activations with a change in code status was significantly longer than activations without a change (66 vs 60 minutes, P=0.05). The rates of telemetry (6% vs 3%, P=0.24) and ICU transfer (40% vs 41%, P=0.8) were similar between patients with a change in code status and patients without a change (Table 2). Patients with a MET‐implemented DNR order were more frequently transferred to the ICU than patients with a preexisting DNR order (36% vs 17%, P<0.01). The median hospital length of stay following MET activation was shorter in patients with a change in code status compared to patients with no change (3 vs 5 days, P<0.0001).

The inpatient mortality was 17% (165/998). Most patients who died had the focus of care changed to comfort measures only (88%, 146/165). When examining the group of patients who died in the hospital with comfort care, we found that 58% (92/159) were transferred to the ICU following the MET call, 5% (8/159) were changed to comfort care during the MET call, and 18% (29/159) had a palliative care consult. We also observed that 16% (25/159) patients who died with comfort care were made DNR during MET activation. The inpatient mortality was significantly higher in patients with a change in code status compared to patients with no change in code status (44% vs 14%, 133/926, P<0.0001). Patients with a MET‐implemented DNR order had a higher inpatient mortality than patients with a preexisting DNR (43% vs 27%, P=0.04). Twenty‐five patients with a MET‐implemented DNR order died in the hospital. When examining a subgroup of patients who required end‐of‐life care (died or discharged from the hospital with hospice), we found patients with a MET‐implemented DNR order were less likely to be discharged with hospice care than patients with a preexisting DNR (4% vs 29%, P=0.01). There was no difference in the use of inpatient palliative care consultation at the end of life in patients with a preexisting DNR versus MET‐implemented DNR order (20% vs 12%, P=0.39). Patients with a MET‐implemented DNR order also had a significantly shorter median time from implementation of comfort care orders to death or discharge with hospice compared to patients with a preexisting DNR order (7 hours, interquartile range [IQR], 416 hours vs 22 hours, IQR 939 hours).

DISCUSSION

We observed a MET‐implemented DNR order in 5% of activations. Little is known about the role of METs in end‐of‐life discussions in the United States, and past experience has primarily come from Australian hospitals. Important differences in end‐of‐life care exist among different countries, particularly with regard to placing limitations on treatment.[9] Our observed rate is similar to the 3% to 10% rate of MET‐implemented DNR orders in previous reports worldwide.[3, 5, 8, 12, 13] Recent data from the United States suggest that METs initiate a DNR order in 28% of cases.[14] However, most DNR orders in that study were placed in the ICU days to weeks after MET activation, likely accounting for the high DNR rate. Our data add to the growing body of evidence that METs play an important role in end‐of‐life discussions among different countries throughout the world, including the United States.

To our knowledge, no prior study has evaluated the impact of code status discussions on MET resource utilization. Our MET spent 6 minutes longer on activations involving a change in code status when compared to activations with no changes made to code status. Presumably, some of this time was spent discussing goals of care. In our opinion, the additional time spent on these activations was invaluable, particularly when considering MET‐initiated end‐of‐life discussions may have prevented several unwanted resuscitations (25 patients died with a MET‐implemented DNR order). Interestingly, less than half of patients had code status discussions at the time of hospital admission. This finding suggests that clinicians could be more vigilant about discussing preferences for resuscitation at the time of admission in patients at risk for clinical deterioration. We suspect that in some cases code status discussions may have occurred between the patient and the primary service later in the patient's hospitalization, which were not captured in our study.

Surprisingly, when examining the use of hospital resources, we found no difference in the rate of unplanned ICU transfer in patients with a change in code status. In fact, we observed a higher rate of ICU transfer in patients with a MET‐implemented DNR order compared to those with a preexisting DNR order (36% vs 17%). These results were at odds with our hypothesis of a lower rate of ICU transfer in patients with MET‐implemented limitations in care. When compared to patients with a preexisting DNR order, patients with a MET‐implemented DNR order were younger, more commonly admitted from home, and had a lower Charlson index. Despite evidence of a lower burden of chronic illness and younger age, patients with a MET‐implemented DNR order had higher inpatient mortality than patients with a preexisting DNR order (43% vs 27%), suggesting an acute and rapidly progressive disease process. These observations may have compelled the MET to advocate for aggressive ICU‐level care in patients with a MET‐implemented DNR order. Another possible explanation for the relatively high rate of ICU transfer is that the MET is, in part, led by ICU staff. Thus, our MET may have made the decision to transfer the patient to the ICU and then subsequently initiated end‐of‐life discussions only after taking ownership of the patient. Furthermore, almost 20% of MET‐implemented changes to code status involved reversing status from DNR to full code. These data suggest that METs are not merely serving as a resource to review code status, but rather providing intensive treatment for acutely ill patients and simultaneously initiating end‐of‐life discussions in a population with a high inpatient mortality rate. The practice pattern observed in our study of transferring patients to the ICU for a trial of intensive therapy at the end of life is consistent with the overall trend in the United States for increased inpatient treatment intensity at the end of life.[14, 15, 16]

Our data suggest that the increased use of ICU resources in patients with a MET‐implemented DNR may be balanced by a shorter hospital length of stay following MET activation. In a multicenter observational study, Jones et al. found hospital length of stay to be similar in patients with and without a MET‐implemented limitation of medical therapy.[3] The authors did not examine length of stay specifically in patients with a DNR order, but rather examined patients with any limitation in medical therapy, including not for ICU admission. We suspect that the shorter length of stay following MET activation in our study was related to the fact that patients with a change in code status had a significantly higher inpatient mortality.

We observed several interesting findings with regard to end‐of‐life care following MET‐implemented DNR orders. First, the inpatient mortality in this population was remarkably high at 43%, compared to 27% in patients with a preexisting DNR order. Interestingly, there was no difference in the rate of palliative care consultation between the 2 groups despite the fact that all 25 patients who died following a MET‐implemented DNR order did so with a comfort measures only order. We also found that patients with a preexisting DNR also had a higher rate of discharge with hospice compared to patients with a MET‐implemented DNR order (29% vs 4%). Thus, our data suggest that inpatient palliative care consultation and hospice services are not resources that are routinely utilized in patients with MET‐initiated DNR orders. It may be the case that the acuity and severity of illness or patient preferences may have precluded the possibility of discharging some patients in our study home with hospice care or implementing comfort care earlier in the hospital course. Patients with MET‐implemented DNR orders were younger, had fewer comorbidities, and died sooner after comfort care orders were written. The overall rate of comfort care provided to patients who died was high at 88%. We have an inpatient comfort measures only order set at our hospital, which may account for the large proportion of patients receiving comfort care at the end of life. In addition, this order set may also help to improve the quality of end‐of‐life care and thus limit the need for palliative care consultation to some extent. However, we found that patients with a MET‐implemented DNR order had a shorter time from comfort care orders to death than patients with a preexisting DNR order. This finding suggests that patients with MET‐implemented DNR orders may have had comfort care implemented relatively late in the course of illness and had less‐than‐optimal end‐of‐life care. Vazquez et al. reported improved quality of end‐of‐life care after implementation of a MET.[10] However, an inpatient palliative care service was not available in that study, and it is not clear whether or not a comfort care order set was available. Evidence suggests that utilization of palliative care resources improves end‐of‐life care in the ICU.[17, 18, 19] We found that more than half of patients who died with comfort care in the hospital did so after being transferred to the ICU for a trial of aggressive care, suggesting that this population may have benefited from more involvement of our palliative care service. In summary, our data on end‐of‐life care following MET activation suggest that the METs are able to take advantage of an opportunity to identify patients who would not want resuscitation efforts because of personal preferences or futility of treatments. However, our surrogate measures of the quality of end‐of‐life care suggest that patients with MET‐implemented DNR orders may benefit from coordinated care with inpatient palliative care services, timelier implementation of comfort care orders, and possibly increased referrals for hospice care to help improve the quality of end‐of‐life care in this population.

Our study is subject to a number of limitations. This was a single‐center study, making the results difficult to generalize. The retrospective nature of our study makes it subject to the limitations inherent in this study design, including bias and confounding. The duration of MET activation was difficult to accurately and objectively measure and is subject to reporting bias. The event stop time, in particular, was subjectively measured by the MET nurse and is difficult to accurately assess, because MET members occasionally leave the bedside and return to reevaluate the response to therapy. We tried to account for this by clearly defining the MET stop time as the point at which MET members leave the bedside and transfer care back to the primary service or physically transfer the patient to a higher level of care. It also bears mentioning that the nurses performing data entry were not aware of the study hypothesis at the time of data entry. Despite including over 1100 MET calls in our analysis, the number of patients with a MET‐implemented DNR order was relatively small, which may have limited our ability to detect differences among subgroups during our analysis. We also did not document the clinical circumstances surrounding the MET‐implemented DNR order. Although we hypothesized that these patients had a higher mortality due to a higher acuity of illness, we were unable to support this hypothesis with the data available in our retrospective study. We did not record which providers were involved in code status discussions and the exact amount time spent on these discussions, making it difficult to accurately quantify the MET resources utilized on the calls. Our MET works closely with the patient's primary service, and it is possible that some of the changes to code status were implemented by the primary service and not MET providers. The patient's primary service may have a preexisting relationship with the patient and would be in a better position to discuss goals of care than MET providers who have had no prior relationship with the patient. However, even in this case, the clinical deterioration prompting MET activation was likely the event that triggered end‐of‐life discussions. Prospective studies would be helpful not only to identify the individuals involved in code status discussions during MET activations, but also to objectively measure the time spent on such discussions. Finally, our study population consisted primarily of Caucasian patients. Preferences for end‐of‐life care may differ among socioeconomic and ethnic groups, thus limiting the generalizability of our study findings.[20, 21]

In conclusion, we found the rate of MET‐implemented DNR orders in the United States to be similar to that of previous reports in other countries. MET events involving a change in code status are associated with increased utilization of MET and ICU resources, but a shorter hospital length of stay. Despite a high inpatient mortality rate, patients with a MET‐implemented DNR had a relatively low utilization of end‐of‐life resources, including palliative care and home hospice services. Coordinated care between METs and palliative care may help to improve of end‐of‐life care in patients with a change in code status following MET activation.

Unplanned hospital readmission within 30 days is an important marker of the quality of care provided in the hospital and in the immediate posthospital setting.[1] In the United States, 1 in 5 Medicare patients is readmitted within 30 days, and related Medicare costs are estimated to be $17 billion annually.[2] Public policy is attempting to drive down healthcare costs in the United States via the Affordable Care Act by reducing payments to hospitals that have high 30‐day readmission rates.[3]

The prevalence of depression is 6.7% of adults,[4] and depressive symptomatology has been linked to hospital readmission.[5, 6] Depressive symptoms are associated with poor health outcomes and increased utilization. Patients with cardiac disease and depressive symptoms have worse outcomes.[7] Mild symptoms are associated with increased primary care and mental health care visits.[8] Both mild and moderate‐to‐severe depressive symptomatology are associated with symptom burden, physical limitation, quality of life, and overall health.[9] Importantly, many hospitalized patients, who do not have a diagnosis of depression, display depressive symptomatology.[10] A gap in knowledge exists as to the utility of stratifying depressive symptomatology between mild and moderate to severe in determining risk for hospital readmission.

Although the causal pathway to worse outcomes in patients with a diagnosis of depression has been studied,[11] the reasons for poor outcomes of patients with depressive symptomatology have not been well described, particularly in hospitalized general medical patients. Nonadherence to physician recommendations, decreased cognitive function, and poor adherence to self‐care recommendations[12, 13, 14] may explain increased healthcare utilization. Physiological models hypothesize heightened levels of proinflammatory markers[15, 16] and vascular depression[17] may play a role. There remains a gap in knowledge as to the postdischarge utilization of hospitalized patients with depressive symptomatology.

We studied the rate of hospital readmission among hospitalized adult patients with mild and moderate‐to‐severe depressive symptoms as defined by the 9‐Item Patient Health Questionnaire (PHQ‐9) depression screening tool.[18]

METHODS

RESULTS

Of the total study population, 15.9% (225/1418) demonstrated mild depressive symptoms, and 23.7% (336/1418) demonstrated moderate‐to‐severe symptoms. Of those in our study, 38.9% (551/1418) self‐reported being told by a healthcare professional that they had depression. Of those self‐reporting depression, 27.9% (273/540) were currently taking antidepressant medication at the time of the index hospitalization. Table 1 shows baseline characteristics by dichotomized utilization outcome. Mean age, marital status, health insurance, employment status, mean length of stay, admissions in the previous 6 months, mean Charlson score, and substance abuse were significantly associated with hospital readmission. Of these characteristics, all but mean length of stay and previous history of substance abuse were significantly associated with ED utilization.

Table 2 shows the unadjusted 30 day hospital readmission, ED utilization, and PCP follow‐up rates. Participants with mild symptoms had higher readmission rates than those without symptoms (0.20 vs 0.13). In other words, 20 readmissions occurred per 100 subjects with mild symptoms, compared with 13 readmissions per 100 subjects without symptoms (P<0.001). The readmission rate was 0.21 for subjects with moderate‐to‐severe depression. The rate of ED utilization for subjects with mild symptoms was 0.18. This was significantly different from ED utilization rates of those with no depression and those with moderate‐to‐severe symptoms, which were 0.16 and 0.28, respectively (P<0.001). The postdischarge follow‐up rates were different for those without depression compared to those with mild and moderate‐to‐severe symptoms (58.7, 49.5, and 51.1, respectively), but this did not reach statistical significance (P=0.06).

Poisson analyses were conducted to control for potential confounding in the relationship between symptom severity and readmission or ED utilization (Table 3). Compared to subjects with no depression, the association between mild symptoms and readmission remained significant (adjusted IRR: 1.49; 95% confidence interval [CI]: 1.11‐2.00) after controlling for relevant confounders. For those with moderate‐to‐severe symptoms, the adjusted IRR was 1.96 (95% CI: 1.51‐2.49). When compared to those without depression, the adjusted IRR for ED reutilization was not found to be significant for those with mild symptoms (1.30; 95% CI: 0.96‐1.76) and significant for those participants with moderate‐to‐severe symptoms (1.48; 95% CI: 1.16‐1.89).

Figure 1 depicts the hazard curve generated for the time to first hospital readmission, stratified by depressive symptom severity. A readmission within 30 days following an index discharge date occurred in 10% of participants without depression, 14% of those with mild symptoms, and 19% of those with moderate‐to‐severe symptoms (P=0.03).

Figure 1

DISCUSSION

Our study shows hospitalized medical patients at an urban academic hospital with a positive screen for depressive symptoms are significantly more likely to be readmitted within 30 days of discharge as compared to those who do not screen positive. The significant association of depressive symptoms and readmission remains even after stratifying by severity and controlling for relevant confounders. Further, there appears to be a dose‐response relationship between depressive symptom severity and readmission. This graded effect makes the distinction between mild and moderate‐to‐severe depressive symptoms a better instrument at predicting rehospitalization than a diagnostic code for depression. Few studies have analyzed the readmission of general medical patients stratified by depressive symptomatology, and even fewer have addressed the presence of mild depressive symptomatology as it relates to readmission. A diagnosis of mild depression is associated with similar though less severe outcomes as compared to major depression, including negative effects on quality of life, functional disability, health status, and mortality.[29] Patients with heart failure and mild depressive symptoms have higher rates of readmission at 3 months and 1 year as compared with those without depressive symptoms, but these findings were not found to be significant.[30] Mild depressive symptoms may contribute to readmission, accrued medical cost, and burden of disease.

We extend previous research[5, 31, 32] by showing that, compared to those without and those with mild symptoms, the readmission risk is even greater for those who screen positive for moderate‐to‐severe symptoms. The mechanism linking depressive symptoms and readmission is not well understood. Behavioral mechanisms such as physical symptom amplification or anxiety about symptoms link depressive symptoms to healthcare utilization after discharge.[33] Depressive symptoms among patients with diabetes, asthma, hypertension, or human immunodeficiency virus (HIV) impairs medication adherence and self‐care behavior.[14, 34, 35, 36] Depressed patients might have reduced social support leading to increased stress, worsened symptoms, and prolonged recovery.[37] These mechanisms may prompt patients to present to hospitals for reevaluation. The direct physiologic consequences of depressive symptoms may be similar to that of the diagnosis of depression. Patients with cardiovascular disease and depression have poor outcomes, which may be related to decreased heart rate variability, hypercoagulability, high burden of inflammatory markers, and severity of left ventricular dysfunction.[38, 39] Among patients with HIV/acquired immunodeficiency syndrome and coronary artery disease, depression is linked to increased proinflammatory marker levels and less favorable outcomes, which may signal a more severe form of the disease or an impaired response to treatment.[15, 16]

Our data have several implications. Though disease burden may play a role in the presence of depressive symptomatology in hospitalized patients, screen‐positive patients still experienced more readmission events as compared to those without depressive symptoms after controlling for relevant confounders. Further, there appears to be a dose‐dependent relationship between depressive symptom severity and rate of readmission. Use of a categorical ICD‐9 code often implies that the diagnosis of depression has been confirmed. Rather than using administrative ICD‐9 codes to account for readmission risk, hospitals may consider screening patients for depressive symptoms during hospitalizations to both identify and risk‐stratify patients at high risk for readmission. Procedures should be implemented to address barriers to safe transitions in care in the screen‐positive population. The relationship between symptom severity and readmission rate may aid in the decision to devote resources to those at highest risk of readmission. Lastly, though research on the treatment of depressive symptoms in medical inpatients has been inconclusive in determining whether this approach is better than usual care or structured pharmacotherapies,[40] further study is needed to determine whether treatment of mild and moderate‐to‐severe depressive symptoms during an acute medical hospitalization will decrease readmission.

Strengths of the current study include the large dataset, the broad range of covariates available for analyses, and the inclusive nature of the sample, which was not restricted to factors such as age or medical condition.

Several limitations should be noted. We did not conduct a psychiatric evaluation to evaluate screen‐positive patients who met diagnostic criteria for minor or major depressive disorder, nor did we reconfirm the presence of symptoms at the time of or following hospital discharge. Our data, then, may not reflect patients' depressive symptomatology prior to the index hospitalization or at the time of discharge. Although such data might further refine the use of depressive symptomatology in identifying patients at high risk for readmission, our findings demonstrate that simply screening positive for depressive symptomatology at time of admission is associated with increased risk of readmission. We do not know the direction of the reported associations. If depressive symptoms are the consequence of higher disease burden, treatment of the underlying disease may be the most important intervention. Although this is possible, our model does include variables (eg, length of stay, Charlson Comorbidity Index), which are likely to adjust for disease severity, pointing to the likelihood that depressive symptoms truly predict hospital readmission independent of disease severity. Data on utilization outside Boston Medical Center (about 9% of outcomes) were determined by patient self‐report and not confirmed by document review. Our results may not be generalizable to populations other than those served by an urban safety‐net hospital or other populations excluded from analysis (eg, nonEnglish‐speaking patients, patients from nursing homes). Finally, social factors such as social support may residually confound the relationship between depressive symptom severity and readmission.

Our finding linking both mild and moderate‐to‐severe depressive symptoms to increased readmission when compared to those without depressive symptoms is significant for future policy. If future studies demonstrate that the initiation of treatment of patients who screen positive for depressive symptoms during an acute hospitalization leads to reduced readmission, policymakers should increase support for mental health screening and programming as an integral portion of general medical patient management.

In conclusion, screening positive for mild or moderate‐to‐severe depressive symptoms is associated with an increased rate of early hospital readmission as compared to those without depressive symptoms, even after controlling for relevant confounders. The rate and hazard of hospital readmission increase with symptom severity. This finding has important implications for future research for hospital screening programming and interventions for patients who screen positive for depressive symptoms.

Disclosures: This research was funded in part by grants from the Agency for Healthcare Research and Quality (NCT00252057) and the National Heart, Lung, and Blood Institute (NCT00217867). Dr. Cancino has been paid for consulting work for PracticeUpdate, a subsidiary of Elsevier. Dr. Culpepper has been paid for participation in advisory boards by AstraZeneca, Eli Lilly and Co., Boehringer Ingelheim Pharmaceuticals Inc., Forest Labs, Janssen Pharmaceuticals, Inc., Jazz Pharmaceuticals plc, H. Lundbeck A/S, Merck & Co., Pfizer Inc., Reckitt Benckiser Pharmaceuticals Inc., Sunovion Pharmaceuticals Inc., and Takeda Pharmaceuticals Inc. He has received payment for educational presentations regarding hospital readmission without mention of any pharmaceutical or other products from Merck. Dr. Mitchell has received honoraria from Merck for lectures on health behavior counseling. Trial registration: NCT00252057, NCT00217867.

Unplanned hospital readmission within 30 days is an important marker of the quality of care provided in the hospital and in the immediate posthospital setting.[1] In the United States, 1 in 5 Medicare patients is readmitted within 30 days, and related Medicare costs are estimated to be $17 billion annually.[2] Public policy is attempting to drive down healthcare costs in the United States via the Affordable Care Act by reducing payments to hospitals that have high 30‐day readmission rates.[3]

The prevalence of depression is 6.7% of adults,[4] and depressive symptomatology has been linked to hospital readmission.[5, 6] Depressive symptoms are associated with poor health outcomes and increased utilization. Patients with cardiac disease and depressive symptoms have worse outcomes.[7] Mild symptoms are associated with increased primary care and mental health care visits.[8] Both mild and moderate‐to‐severe depressive symptomatology are associated with symptom burden, physical limitation, quality of life, and overall health.[9] Importantly, many hospitalized patients, who do not have a diagnosis of depression, display depressive symptomatology.[10] A gap in knowledge exists as to the utility of stratifying depressive symptomatology between mild and moderate to severe in determining risk for hospital readmission.

Although the causal pathway to worse outcomes in patients with a diagnosis of depression has been studied,[11] the reasons for poor outcomes of patients with depressive symptomatology have not been well described, particularly in hospitalized general medical patients. Nonadherence to physician recommendations, decreased cognitive function, and poor adherence to self‐care recommendations[12, 13, 14] may explain increased healthcare utilization. Physiological models hypothesize heightened levels of proinflammatory markers[15, 16] and vascular depression[17] may play a role. There remains a gap in knowledge as to the postdischarge utilization of hospitalized patients with depressive symptomatology.

We studied the rate of hospital readmission among hospitalized adult patients with mild and moderate‐to‐severe depressive symptoms as defined by the 9‐Item Patient Health Questionnaire (PHQ‐9) depression screening tool.[18]

METHODS

RESULTS

Of the total study population, 15.9% (225/1418) demonstrated mild depressive symptoms, and 23.7% (336/1418) demonstrated moderate‐to‐severe symptoms. Of those in our study, 38.9% (551/1418) self‐reported being told by a healthcare professional that they had depression. Of those self‐reporting depression, 27.9% (273/540) were currently taking antidepressant medication at the time of the index hospitalization. Table 1 shows baseline characteristics by dichotomized utilization outcome. Mean age, marital status, health insurance, employment status, mean length of stay, admissions in the previous 6 months, mean Charlson score, and substance abuse were significantly associated with hospital readmission. Of these characteristics, all but mean length of stay and previous history of substance abuse were significantly associated with ED utilization.

Table 2 shows the unadjusted 30 day hospital readmission, ED utilization, and PCP follow‐up rates. Participants with mild symptoms had higher readmission rates than those without symptoms (0.20 vs 0.13). In other words, 20 readmissions occurred per 100 subjects with mild symptoms, compared with 13 readmissions per 100 subjects without symptoms (P<0.001). The readmission rate was 0.21 for subjects with moderate‐to‐severe depression. The rate of ED utilization for subjects with mild symptoms was 0.18. This was significantly different from ED utilization rates of those with no depression and those with moderate‐to‐severe symptoms, which were 0.16 and 0.28, respectively (P<0.001). The postdischarge follow‐up rates were different for those without depression compared to those with mild and moderate‐to‐severe symptoms (58.7, 49.5, and 51.1, respectively), but this did not reach statistical significance (P=0.06).

Poisson analyses were conducted to control for potential confounding in the relationship between symptom severity and readmission or ED utilization (Table 3). Compared to subjects with no depression, the association between mild symptoms and readmission remained significant (adjusted IRR: 1.49; 95% confidence interval [CI]: 1.11‐2.00) after controlling for relevant confounders. For those with moderate‐to‐severe symptoms, the adjusted IRR was 1.96 (95% CI: 1.51‐2.49). When compared to those without depression, the adjusted IRR for ED reutilization was not found to be significant for those with mild symptoms (1.30; 95% CI: 0.96‐1.76) and significant for those participants with moderate‐to‐severe symptoms (1.48; 95% CI: 1.16‐1.89).

Figure 1 depicts the hazard curve generated for the time to first hospital readmission, stratified by depressive symptom severity. A readmission within 30 days following an index discharge date occurred in 10% of participants without depression, 14% of those with mild symptoms, and 19% of those with moderate‐to‐severe symptoms (P=0.03).

Figure 1

DISCUSSION

Our study shows hospitalized medical patients at an urban academic hospital with a positive screen for depressive symptoms are significantly more likely to be readmitted within 30 days of discharge as compared to those who do not screen positive. The significant association of depressive symptoms and readmission remains even after stratifying by severity and controlling for relevant confounders. Further, there appears to be a dose‐response relationship between depressive symptom severity and readmission. This graded effect makes the distinction between mild and moderate‐to‐severe depressive symptoms a better instrument at predicting rehospitalization than a diagnostic code for depression. Few studies have analyzed the readmission of general medical patients stratified by depressive symptomatology, and even fewer have addressed the presence of mild depressive symptomatology as it relates to readmission. A diagnosis of mild depression is associated with similar though less severe outcomes as compared to major depression, including negative effects on quality of life, functional disability, health status, and mortality.[29] Patients with heart failure and mild depressive symptoms have higher rates of readmission at 3 months and 1 year as compared with those without depressive symptoms, but these findings were not found to be significant.[30] Mild depressive symptoms may contribute to readmission, accrued medical cost, and burden of disease.

We extend previous research[5, 31, 32] by showing that, compared to those without and those with mild symptoms, the readmission risk is even greater for those who screen positive for moderate‐to‐severe symptoms. The mechanism linking depressive symptoms and readmission is not well understood. Behavioral mechanisms such as physical symptom amplification or anxiety about symptoms link depressive symptoms to healthcare utilization after discharge.[33] Depressive symptoms among patients with diabetes, asthma, hypertension, or human immunodeficiency virus (HIV) impairs medication adherence and self‐care behavior.[14, 34, 35, 36] Depressed patients might have reduced social support leading to increased stress, worsened symptoms, and prolonged recovery.[37] These mechanisms may prompt patients to present to hospitals for reevaluation. The direct physiologic consequences of depressive symptoms may be similar to that of the diagnosis of depression. Patients with cardiovascular disease and depression have poor outcomes, which may be related to decreased heart rate variability, hypercoagulability, high burden of inflammatory markers, and severity of left ventricular dysfunction.[38, 39] Among patients with HIV/acquired immunodeficiency syndrome and coronary artery disease, depression is linked to increased proinflammatory marker levels and less favorable outcomes, which may signal a more severe form of the disease or an impaired response to treatment.[15, 16]

Our data have several implications. Though disease burden may play a role in the presence of depressive symptomatology in hospitalized patients, screen‐positive patients still experienced more readmission events as compared to those without depressive symptoms after controlling for relevant confounders. Further, there appears to be a dose‐dependent relationship between depressive symptom severity and rate of readmission. Use of a categorical ICD‐9 code often implies that the diagnosis of depression has been confirmed. Rather than using administrative ICD‐9 codes to account for readmission risk, hospitals may consider screening patients for depressive symptoms during hospitalizations to both identify and risk‐stratify patients at high risk for readmission. Procedures should be implemented to address barriers to safe transitions in care in the screen‐positive population. The relationship between symptom severity and readmission rate may aid in the decision to devote resources to those at highest risk of readmission. Lastly, though research on the treatment of depressive symptoms in medical inpatients has been inconclusive in determining whether this approach is better than usual care or structured pharmacotherapies,[40] further study is needed to determine whether treatment of mild and moderate‐to‐severe depressive symptoms during an acute medical hospitalization will decrease readmission.

Strengths of the current study include the large dataset, the broad range of covariates available for analyses, and the inclusive nature of the sample, which was not restricted to factors such as age or medical condition.

Several limitations should be noted. We did not conduct a psychiatric evaluation to evaluate screen‐positive patients who met diagnostic criteria for minor or major depressive disorder, nor did we reconfirm the presence of symptoms at the time of or following hospital discharge. Our data, then, may not reflect patients' depressive symptomatology prior to the index hospitalization or at the time of discharge. Although such data might further refine the use of depressive symptomatology in identifying patients at high risk for readmission, our findings demonstrate that simply screening positive for depressive symptomatology at time of admission is associated with increased risk of readmission. We do not know the direction of the reported associations. If depressive symptoms are the consequence of higher disease burden, treatment of the underlying disease may be the most important intervention. Although this is possible, our model does include variables (eg, length of stay, Charlson Comorbidity Index), which are likely to adjust for disease severity, pointing to the likelihood that depressive symptoms truly predict hospital readmission independent of disease severity. Data on utilization outside Boston Medical Center (about 9% of outcomes) were determined by patient self‐report and not confirmed by document review. Our results may not be generalizable to populations other than those served by an urban safety‐net hospital or other populations excluded from analysis (eg, nonEnglish‐speaking patients, patients from nursing homes). Finally, social factors such as social support may residually confound the relationship between depressive symptom severity and readmission.

Our finding linking both mild and moderate‐to‐severe depressive symptoms to increased readmission when compared to those without depressive symptoms is significant for future policy. If future studies demonstrate that the initiation of treatment of patients who screen positive for depressive symptoms during an acute hospitalization leads to reduced readmission, policymakers should increase support for mental health screening and programming as an integral portion of general medical patient management.

In conclusion, screening positive for mild or moderate‐to‐severe depressive symptoms is associated with an increased rate of early hospital readmission as compared to those without depressive symptoms, even after controlling for relevant confounders. The rate and hazard of hospital readmission increase with symptom severity. This finding has important implications for future research for hospital screening programming and interventions for patients who screen positive for depressive symptoms.

Disclosures: This research was funded in part by grants from the Agency for Healthcare Research and Quality (NCT00252057) and the National Heart, Lung, and Blood Institute (NCT00217867). Dr. Cancino has been paid for consulting work for PracticeUpdate, a subsidiary of Elsevier. Dr. Culpepper has been paid for participation in advisory boards by AstraZeneca, Eli Lilly and Co., Boehringer Ingelheim Pharmaceuticals Inc., Forest Labs, Janssen Pharmaceuticals, Inc., Jazz Pharmaceuticals plc, H. Lundbeck A/S, Merck & Co., Pfizer Inc., Reckitt Benckiser Pharmaceuticals Inc., Sunovion Pharmaceuticals Inc., and Takeda Pharmaceuticals Inc. He has received payment for educational presentations regarding hospital readmission without mention of any pharmaceutical or other products from Merck. Dr. Mitchell has received honoraria from Merck for lectures on health behavior counseling. Trial registration: NCT00252057, NCT00217867.

Unplanned hospital readmission within 30 days is an important marker of the quality of care provided in the hospital and in the immediate posthospital setting.[1] In the United States, 1 in 5 Medicare patients is readmitted within 30 days, and related Medicare costs are estimated to be $17 billion annually.[2] Public policy is attempting to drive down healthcare costs in the United States via the Affordable Care Act by reducing payments to hospitals that have high 30‐day readmission rates.[3]

The prevalence of depression is 6.7% of adults,[4] and depressive symptomatology has been linked to hospital readmission.[5, 6] Depressive symptoms are associated with poor health outcomes and increased utilization. Patients with cardiac disease and depressive symptoms have worse outcomes.[7] Mild symptoms are associated with increased primary care and mental health care visits.[8] Both mild and moderate‐to‐severe depressive symptomatology are associated with symptom burden, physical limitation, quality of life, and overall health.[9] Importantly, many hospitalized patients, who do not have a diagnosis of depression, display depressive symptomatology.[10] A gap in knowledge exists as to the utility of stratifying depressive symptomatology between mild and moderate to severe in determining risk for hospital readmission.

Although the causal pathway to worse outcomes in patients with a diagnosis of depression has been studied,[11] the reasons for poor outcomes of patients with depressive symptomatology have not been well described, particularly in hospitalized general medical patients. Nonadherence to physician recommendations, decreased cognitive function, and poor adherence to self‐care recommendations[12, 13, 14] may explain increased healthcare utilization. Physiological models hypothesize heightened levels of proinflammatory markers[15, 16] and vascular depression[17] may play a role. There remains a gap in knowledge as to the postdischarge utilization of hospitalized patients with depressive symptomatology.

We studied the rate of hospital readmission among hospitalized adult patients with mild and moderate‐to‐severe depressive symptoms as defined by the 9‐Item Patient Health Questionnaire (PHQ‐9) depression screening tool.[18]

METHODS

RESULTS

Of the total study population, 15.9% (225/1418) demonstrated mild depressive symptoms, and 23.7% (336/1418) demonstrated moderate‐to‐severe symptoms. Of those in our study, 38.9% (551/1418) self‐reported being told by a healthcare professional that they had depression. Of those self‐reporting depression, 27.9% (273/540) were currently taking antidepressant medication at the time of the index hospitalization. Table 1 shows baseline characteristics by dichotomized utilization outcome. Mean age, marital status, health insurance, employment status, mean length of stay, admissions in the previous 6 months, mean Charlson score, and substance abuse were significantly associated with hospital readmission. Of these characteristics, all but mean length of stay and previous history of substance abuse were significantly associated with ED utilization.

Table 2 shows the unadjusted 30 day hospital readmission, ED utilization, and PCP follow‐up rates. Participants with mild symptoms had higher readmission rates than those without symptoms (0.20 vs 0.13). In other words, 20 readmissions occurred per 100 subjects with mild symptoms, compared with 13 readmissions per 100 subjects without symptoms (P<0.001). The readmission rate was 0.21 for subjects with moderate‐to‐severe depression. The rate of ED utilization for subjects with mild symptoms was 0.18. This was significantly different from ED utilization rates of those with no depression and those with moderate‐to‐severe symptoms, which were 0.16 and 0.28, respectively (P<0.001). The postdischarge follow‐up rates were different for those without depression compared to those with mild and moderate‐to‐severe symptoms (58.7, 49.5, and 51.1, respectively), but this did not reach statistical significance (P=0.06).

Poisson analyses were conducted to control for potential confounding in the relationship between symptom severity and readmission or ED utilization (Table 3). Compared to subjects with no depression, the association between mild symptoms and readmission remained significant (adjusted IRR: 1.49; 95% confidence interval [CI]: 1.11‐2.00) after controlling for relevant confounders. For those with moderate‐to‐severe symptoms, the adjusted IRR was 1.96 (95% CI: 1.51‐2.49). When compared to those without depression, the adjusted IRR for ED reutilization was not found to be significant for those with mild symptoms (1.30; 95% CI: 0.96‐1.76) and significant for those participants with moderate‐to‐severe symptoms (1.48; 95% CI: 1.16‐1.89).

Figure 1 depicts the hazard curve generated for the time to first hospital readmission, stratified by depressive symptom severity. A readmission within 30 days following an index discharge date occurred in 10% of participants without depression, 14% of those with mild symptoms, and 19% of those with moderate‐to‐severe symptoms (P=0.03).

Figure 1

DISCUSSION

Our study shows hospitalized medical patients at an urban academic hospital with a positive screen for depressive symptoms are significantly more likely to be readmitted within 30 days of discharge as compared to those who do not screen positive. The significant association of depressive symptoms and readmission remains even after stratifying by severity and controlling for relevant confounders. Further, there appears to be a dose‐response relationship between depressive symptom severity and readmission. This graded effect makes the distinction between mild and moderate‐to‐severe depressive symptoms a better instrument at predicting rehospitalization than a diagnostic code for depression. Few studies have analyzed the readmission of general medical patients stratified by depressive symptomatology, and even fewer have addressed the presence of mild depressive symptomatology as it relates to readmission. A diagnosis of mild depression is associated with similar though less severe outcomes as compared to major depression, including negative effects on quality of life, functional disability, health status, and mortality.[29] Patients with heart failure and mild depressive symptoms have higher rates of readmission at 3 months and 1 year as compared with those without depressive symptoms, but these findings were not found to be significant.[30] Mild depressive symptoms may contribute to readmission, accrued medical cost, and burden of disease.

We extend previous research[5, 31, 32] by showing that, compared to those without and those with mild symptoms, the readmission risk is even greater for those who screen positive for moderate‐to‐severe symptoms. The mechanism linking depressive symptoms and readmission is not well understood. Behavioral mechanisms such as physical symptom amplification or anxiety about symptoms link depressive symptoms to healthcare utilization after discharge.[33] Depressive symptoms among patients with diabetes, asthma, hypertension, or human immunodeficiency virus (HIV) impairs medication adherence and self‐care behavior.[14, 34, 35, 36] Depressed patients might have reduced social support leading to increased stress, worsened symptoms, and prolonged recovery.[37] These mechanisms may prompt patients to present to hospitals for reevaluation. The direct physiologic consequences of depressive symptoms may be similar to that of the diagnosis of depression. Patients with cardiovascular disease and depression have poor outcomes, which may be related to decreased heart rate variability, hypercoagulability, high burden of inflammatory markers, and severity of left ventricular dysfunction.[38, 39] Among patients with HIV/acquired immunodeficiency syndrome and coronary artery disease, depression is linked to increased proinflammatory marker levels and less favorable outcomes, which may signal a more severe form of the disease or an impaired response to treatment.[15, 16]

Our data have several implications. Though disease burden may play a role in the presence of depressive symptomatology in hospitalized patients, screen‐positive patients still experienced more readmission events as compared to those without depressive symptoms after controlling for relevant confounders. Further, there appears to be a dose‐dependent relationship between depressive symptom severity and rate of readmission. Use of a categorical ICD‐9 code often implies that the diagnosis of depression has been confirmed. Rather than using administrative ICD‐9 codes to account for readmission risk, hospitals may consider screening patients for depressive symptoms during hospitalizations to both identify and risk‐stratify patients at high risk for readmission. Procedures should be implemented to address barriers to safe transitions in care in the screen‐positive population. The relationship between symptom severity and readmission rate may aid in the decision to devote resources to those at highest risk of readmission. Lastly, though research on the treatment of depressive symptoms in medical inpatients has been inconclusive in determining whether this approach is better than usual care or structured pharmacotherapies,[40] further study is needed to determine whether treatment of mild and moderate‐to‐severe depressive symptoms during an acute medical hospitalization will decrease readmission.

Strengths of the current study include the large dataset, the broad range of covariates available for analyses, and the inclusive nature of the sample, which was not restricted to factors such as age or medical condition.

Several limitations should be noted. We did not conduct a psychiatric evaluation to evaluate screen‐positive patients who met diagnostic criteria for minor or major depressive disorder, nor did we reconfirm the presence of symptoms at the time of or following hospital discharge. Our data, then, may not reflect patients' depressive symptomatology prior to the index hospitalization or at the time of discharge. Although such data might further refine the use of depressive symptomatology in identifying patients at high risk for readmission, our findings demonstrate that simply screening positive for depressive symptomatology at time of admission is associated with increased risk of readmission. We do not know the direction of the reported associations. If depressive symptoms are the consequence of higher disease burden, treatment of the underlying disease may be the most important intervention. Although this is possible, our model does include variables (eg, length of stay, Charlson Comorbidity Index), which are likely to adjust for disease severity, pointing to the likelihood that depressive symptoms truly predict hospital readmission independent of disease severity. Data on utilization outside Boston Medical Center (about 9% of outcomes) were determined by patient self‐report and not confirmed by document review. Our results may not be generalizable to populations other than those served by an urban safety‐net hospital or other populations excluded from analysis (eg, nonEnglish‐speaking patients, patients from nursing homes). Finally, social factors such as social support may residually confound the relationship between depressive symptom severity and readmission.

Our finding linking both mild and moderate‐to‐severe depressive symptoms to increased readmission when compared to those without depressive symptoms is significant for future policy. If future studies demonstrate that the initiation of treatment of patients who screen positive for depressive symptoms during an acute hospitalization leads to reduced readmission, policymakers should increase support for mental health screening and programming as an integral portion of general medical patient management.

In conclusion, screening positive for mild or moderate‐to‐severe depressive symptoms is associated with an increased rate of early hospital readmission as compared to those without depressive symptoms, even after controlling for relevant confounders. The rate and hazard of hospital readmission increase with symptom severity. This finding has important implications for future research for hospital screening programming and interventions for patients who screen positive for depressive symptoms.

Disclosures: This research was funded in part by grants from the Agency for Healthcare Research and Quality (NCT00252057) and the National Heart, Lung, and Blood Institute (NCT00217867). Dr. Cancino has been paid for consulting work for PracticeUpdate, a subsidiary of Elsevier. Dr. Culpepper has been paid for participation in advisory boards by AstraZeneca, Eli Lilly and Co., Boehringer Ingelheim Pharmaceuticals Inc., Forest Labs, Janssen Pharmaceuticals, Inc., Jazz Pharmaceuticals plc, H. Lundbeck A/S, Merck & Co., Pfizer Inc., Reckitt Benckiser Pharmaceuticals Inc., Sunovion Pharmaceuticals Inc., and Takeda Pharmaceuticals Inc. He has received payment for educational presentations regarding hospital readmission without mention of any pharmaceutical or other products from Merck. Dr. Mitchell has received honoraria from Merck for lectures on health behavior counseling. Trial registration: NCT00252057, NCT00217867.