Original Research

Given its safety profile and bactericidal activity against the predominant organisms causing surgical site infections (SSIs), cefazolin remains the most popular choice for surgical prophylaxis.¹ Cefazolin offers protection against the pathogens most likely to contaminate the surgical site while minimizing inappropriate methicillin- resistant Staphylococcus aureus coverage that occurs with alternatives such as vancomycin and clindamycin. Documented allergies to Β-lactam antibiotics have historically forced clinicians to avoid the use of cephalosporins due to the potential risk of cross-reactivity. True type 1 (immunoglobin E [IgE]-mediated) cross-allergic reactions between penicillin and cephalosporins are rare, and previously reported data indicate cross-reactivity as a result of antibody recognition is more closely related to the side-chain identity rather than the Β-lactam ring.^2,3

About 10% of US patients report having a penicillin allergy; however, < 1% of the population has a true IgE-mediated allergic reaction.⁴ Previous research that has challenged penicillin allergies with cefazolin for surgical prophylaxis has reported minimal rates of allergic reactions.^2-5

In previous trials, patients with a history of delayed skin reactions, such as Stevens-Johnson syndrome (SJS), toxic epidermal necrolysis (TEN), and drug reaction with eosinophilia and systemic symptoms (DRESS), were excluded. Additionally, patients with an allergy to cefazolin including those with urticaria, angioedema, bronchospasm, or anaphylaxis, were excluded from perioperative retrial of cefazolin. Grant et al found that cefazolin can be safely given to patients with IgE-mediated reactions to penicillin and other cephalosporins due to a structurally different side chain.³

In January 2023, the Veteran Health Indiana (VHI) pharmacy team in conjunction with surgery, infectious disease, and anesthesiology, implemented a screening tool as an amendment to perioperative antibiotic guidance to help determine which patients with a documented penicillin allergy could be candidates for perioperative cefazolin. The implemented screening tool (Allergy Clarification for Cefazolin Evidence-Based Prescribing Tool) has been described by Lam et al, who reported that an increased proportion of patients with documented penicillin allergy received cefazolin without more adverse drug reactions (ADRs).⁵ Patients with a Β-lactam allergy were eligible to receive cefazolin unless the ADR was SJS, TEN, or DRESS, or the offending agent was cefazolin and the patient experienced urticaria, angioedema, bronchospasm, or anaphylaxis. If the reaction was not from cefazolin or was unknown, patients were eligible to receive cefazolin (Figure).

To date, minimal data exist to evaluate the incidence of ADRs when cefazolin is given perioperatively to patients with a previously documented penicillin allergy. The purpose of this study was to evaluate the incidence of allergic ADRs in patients who had a documented penicillin allergy and received periprocedural antibiotics.

Methods

This single-center, retrospective chart review used the US Department of Veterans Affairs (VA) Computerized Patient Record System (CPRS) to identify patients with a documented penicillin allergy who underwent an operation and received periprocedural antibiotics between February 1, 2023, and January 31, 2024. This study was reviewed and approved by the Indiana University Health Institutional Review Board and the VHI Research and Development Committee.

Patients were enrolled if they were aged ≥ 18 years, had a documented penicillin allergy, underwent a surgical intervention, and received perioperative antibiotics during the study period. Patients were excluded if they had a documented penicillin allergy resulting in severe delayed skin reactions (ie, SJS, TEN, or DRESS). These criteria produced 197 surgical procedures. Data were collected for each surgical procedure, so patients could be included more than once. Patient history of allergic reaction to penicillin was obtained through CPRS.

The primary endpoint was the percentage of allergic ADRs in patients with penicillin allergies receiving cefazolin perioperatively. Secondary outcomes included the appropriateness of the antibiotic regimen in congruence with American System of Health Pharmacists (ASHP) recommendations, incidence of SSIs within 30 days of the procedure, incidence of ADRs in those with a history of anaphylaxis vs nonanaphylaxis allergy, incidence of allergic reaction requiring pharmacologic and nonpharmacologic interventions, and incidence of acute kidney injury (AKI). AKI was defined as an increase in serum creatinine by ≥ 0.3 mg/dL within 48 hours or an increase in serum creatinine to ≥ 1.5 times baseline.

Demographic data included sex, age, race, preoperative serum creatinine, and postoperative serum creatinine. Anaphylaxis was defined as an acute onset of illness (within minutes to several hours) with involvement of skin, mucosal tissue, or both involving either respiratory compromise or reduced blood pressures. Allergic reactions were defined as facial, tongue, throat, airway, lip, mouth, periorbital, or eye swelling, urticaria, angioedema, dyspnea, anaphylaxis, or a positive penicillin skin test. Additionally, data collected included the description and severity of postprophylactic antibiotic reaction, antibiotic choice, interventions required for the allergic reaction, SSI occurrence, date of SSI, operating specialty, and postoperative change in renal function.

Descriptive statistics, including mean, SD, and percentages were reported for baseline characteristics of the study population. Percentages were used to demonstrate the differences in primary and secondary outcomes for each study group. Fisher exact tests were used for incidence of ADRs in patients with penicillin allergy who received cefazolin and reported incidence of SSIs.

Results

A total of 197 surgical procedures in patients with a documented penicillin allergy were included; 127 procedures used cefazolin perioperatively, 3 procedures used cefazolin plus gentamicin, and 67 procedures used other antibiotics. Most patients were White (n = 160; 81.2%), male (n = 158; 80.2%), and had a mean age of 64.9 years. Urology was the most common surgical specialty (n = 59; 29.9%) (Table 1). Of the 16 patients with documented penicillin anaphylaxis reaction, 8 received cefazolin and 8 received a different antibiotic. A total of 181 patients reported a nonanaphylaxis allergy. One hundred fifty-one patients (68.6%) reported a reaction history of hives, rash, or swelling (Table 2). Patients could report ≥ 1 reaction. The most prevalent antibiotics used were cefazolin, which was used by 130 patients (61.3%), and clindamycin which was used by 33 patients (15.6%) (Table 3). Patients could receive ≥ 1 antibiotic.

For the primary outcome, the incidence of allergic reactions in patients allergic to penicillin, there was no incidence of allergic reactions in either the cefazolin or other group. Given the absence of reactions, no interventions were required.

There were no ADRs in those with history of anaphylaxis or nonanaphylaxis allergy. In the cefazolin group, 126 of 127 surgical procedure regimens (99.2%) were congruent with ASHP recommendations, all 3 surgical procedures regimens in the cefazolin plus gentamicin group were congruent with ASHP recommendations, and 58 of 67 surgical procedure regimens (86.6%) in the other antibiotic group were congruent with ASHP recommendations. None of the 127 patients in the cefazolin group or of the 3 patients in the cefazolin plus gentamicin group reported an SSI, and 3 of 67 patients (4.5%) had an SSI in the other antibiotic group. One procedure that resulted in SSI was not congruent with ASHP recommendations. Twenty-four patients had 2 serum creatinine levels drawn within 48 hours of surgery. One of 12 patients (8.3%) and 0 of 12 patients had an AKI in the cefazolin and other antibiotic group, respectively (Table 4).

Discussion

Implementation of a screening tool at VHI allowed patients with documented penicillin allergy, including anaphylaxis, to receive cefazolin perioperatively. Broad spectrum antibiotics such as vancomycin, clindamycin, and fluoroquinolones are frequently used in patients allergic to penicillin, which can increase health care costs, risk of toxicity, and antimicrobial resistance.⁴ There was no incidence of allergic reactions noted in patients allergic to penicillin who received cefazolin. When comparing the incidence of observed allergic reactions to received perioperative antibiotics in the cefazolin group to previously published literature, no difference in allergy rates (P = .09) was found.³ Most antibiotics administered were congruent with ASHP guideline recommendations, and most patients eligible for cefazolin received it perioperatively.

Similar to this study, Goodman et al concluded that cefazolin appears to be a safe regimen in patients with documented penicillin anaphylactic reaction for surgical prophylaxis with only 1 (0.2%) potential allergic reaction.⁶ Patients who received cefazolin perioperatively had a statistically significant decrease in SSI rates. There were no clinically or statistically significant differences found between the proportion of allergic reactions or ADRs when compared to alternative antibiotics. Lessard et al concluded that a pharmacist-led interdisciplinary collaborative practice agreement increased cefazolin use in patients allergic to penicillin, including those with urticaria and anaphylaxis, with no reported ADRs.⁷ This study further demonstrated the safety of cefazolin use in patients with anaphylaxis to penicillin.

Limitations

This study’s single-center, retrospective design, patient population, and small sample size limit the generalizability of its results. The data collected are dependent on documentation in the chart. No ADRs were reported from the antibiotics patients received perioperatively. When considering safety data, information such as serum creatinine were available only in CPRS and some patients did not receive a postprocedure serum creatinine level. Additionally, this study did not investigate whether there was an increase in preferred preoperative antimicrobial prophylaxis after implementation of this protocol.

Conclusions

The results of this study support the use of cefazolin perioperatively in patients allergic to penicillin, including those with a history of anaphylaxis. Additional research should be conducted to validate data given the low incidence of ADRs. The primary outcome did not reach statistical significance, but the results may be clinically significant from a stewardship and safety perspective. VHI continues to use the screening tool described in this article.

Given its safety profile and bactericidal activity against the predominant organisms causing surgical site infections (SSIs), cefazolin remains the most popular choice for surgical prophylaxis.¹ Cefazolin offers protection against the pathogens most likely to contaminate the surgical site while minimizing inappropriate methicillin- resistant Staphylococcus aureus coverage that occurs with alternatives such as vancomycin and clindamycin. Documented allergies to Β-lactam antibiotics have historically forced clinicians to avoid the use of cephalosporins due to the potential risk of cross-reactivity. True type 1 (immunoglobin E [IgE]-mediated) cross-allergic reactions between penicillin and cephalosporins are rare, and previously reported data indicate cross-reactivity as a result of antibody recognition is more closely related to the side-chain identity rather than the Β-lactam ring.^2,3

About 10% of US patients report having a penicillin allergy; however, < 1% of the population has a true IgE-mediated allergic reaction.⁴ Previous research that has challenged penicillin allergies with cefazolin for surgical prophylaxis has reported minimal rates of allergic reactions.^2-5

In previous trials, patients with a history of delayed skin reactions, such as Stevens-Johnson syndrome (SJS), toxic epidermal necrolysis (TEN), and drug reaction with eosinophilia and systemic symptoms (DRESS), were excluded. Additionally, patients with an allergy to cefazolin including those with urticaria, angioedema, bronchospasm, or anaphylaxis, were excluded from perioperative retrial of cefazolin. Grant et al found that cefazolin can be safely given to patients with IgE-mediated reactions to penicillin and other cephalosporins due to a structurally different side chain.³

In January 2023, the Veteran Health Indiana (VHI) pharmacy team in conjunction with surgery, infectious disease, and anesthesiology, implemented a screening tool as an amendment to perioperative antibiotic guidance to help determine which patients with a documented penicillin allergy could be candidates for perioperative cefazolin. The implemented screening tool (Allergy Clarification for Cefazolin Evidence-Based Prescribing Tool) has been described by Lam et al, who reported that an increased proportion of patients with documented penicillin allergy received cefazolin without more adverse drug reactions (ADRs).⁵ Patients with a Β-lactam allergy were eligible to receive cefazolin unless the ADR was SJS, TEN, or DRESS, or the offending agent was cefazolin and the patient experienced urticaria, angioedema, bronchospasm, or anaphylaxis. If the reaction was not from cefazolin or was unknown, patients were eligible to receive cefazolin (Figure).

To date, minimal data exist to evaluate the incidence of ADRs when cefazolin is given perioperatively to patients with a previously documented penicillin allergy. The purpose of this study was to evaluate the incidence of allergic ADRs in patients who had a documented penicillin allergy and received periprocedural antibiotics.

Methods

This single-center, retrospective chart review used the US Department of Veterans Affairs (VA) Computerized Patient Record System (CPRS) to identify patients with a documented penicillin allergy who underwent an operation and received periprocedural antibiotics between February 1, 2023, and January 31, 2024. This study was reviewed and approved by the Indiana University Health Institutional Review Board and the VHI Research and Development Committee.

Patients were enrolled if they were aged ≥ 18 years, had a documented penicillin allergy, underwent a surgical intervention, and received perioperative antibiotics during the study period. Patients were excluded if they had a documented penicillin allergy resulting in severe delayed skin reactions (ie, SJS, TEN, or DRESS). These criteria produced 197 surgical procedures. Data were collected for each surgical procedure, so patients could be included more than once. Patient history of allergic reaction to penicillin was obtained through CPRS.

The primary endpoint was the percentage of allergic ADRs in patients with penicillin allergies receiving cefazolin perioperatively. Secondary outcomes included the appropriateness of the antibiotic regimen in congruence with American System of Health Pharmacists (ASHP) recommendations, incidence of SSIs within 30 days of the procedure, incidence of ADRs in those with a history of anaphylaxis vs nonanaphylaxis allergy, incidence of allergic reaction requiring pharmacologic and nonpharmacologic interventions, and incidence of acute kidney injury (AKI). AKI was defined as an increase in serum creatinine by ≥ 0.3 mg/dL within 48 hours or an increase in serum creatinine to ≥ 1.5 times baseline.

Demographic data included sex, age, race, preoperative serum creatinine, and postoperative serum creatinine. Anaphylaxis was defined as an acute onset of illness (within minutes to several hours) with involvement of skin, mucosal tissue, or both involving either respiratory compromise or reduced blood pressures. Allergic reactions were defined as facial, tongue, throat, airway, lip, mouth, periorbital, or eye swelling, urticaria, angioedema, dyspnea, anaphylaxis, or a positive penicillin skin test. Additionally, data collected included the description and severity of postprophylactic antibiotic reaction, antibiotic choice, interventions required for the allergic reaction, SSI occurrence, date of SSI, operating specialty, and postoperative change in renal function.

Descriptive statistics, including mean, SD, and percentages were reported for baseline characteristics of the study population. Percentages were used to demonstrate the differences in primary and secondary outcomes for each study group. Fisher exact tests were used for incidence of ADRs in patients with penicillin allergy who received cefazolin and reported incidence of SSIs.

Results

A total of 197 surgical procedures in patients with a documented penicillin allergy were included; 127 procedures used cefazolin perioperatively, 3 procedures used cefazolin plus gentamicin, and 67 procedures used other antibiotics. Most patients were White (n = 160; 81.2%), male (n = 158; 80.2%), and had a mean age of 64.9 years. Urology was the most common surgical specialty (n = 59; 29.9%) (Table 1). Of the 16 patients with documented penicillin anaphylaxis reaction, 8 received cefazolin and 8 received a different antibiotic. A total of 181 patients reported a nonanaphylaxis allergy. One hundred fifty-one patients (68.6%) reported a reaction history of hives, rash, or swelling (Table 2). Patients could report ≥ 1 reaction. The most prevalent antibiotics used were cefazolin, which was used by 130 patients (61.3%), and clindamycin which was used by 33 patients (15.6%) (Table 3). Patients could receive ≥ 1 antibiotic.

For the primary outcome, the incidence of allergic reactions in patients allergic to penicillin, there was no incidence of allergic reactions in either the cefazolin or other group. Given the absence of reactions, no interventions were required.

There were no ADRs in those with history of anaphylaxis or nonanaphylaxis allergy. In the cefazolin group, 126 of 127 surgical procedure regimens (99.2%) were congruent with ASHP recommendations, all 3 surgical procedures regimens in the cefazolin plus gentamicin group were congruent with ASHP recommendations, and 58 of 67 surgical procedure regimens (86.6%) in the other antibiotic group were congruent with ASHP recommendations. None of the 127 patients in the cefazolin group or of the 3 patients in the cefazolin plus gentamicin group reported an SSI, and 3 of 67 patients (4.5%) had an SSI in the other antibiotic group. One procedure that resulted in SSI was not congruent with ASHP recommendations. Twenty-four patients had 2 serum creatinine levels drawn within 48 hours of surgery. One of 12 patients (8.3%) and 0 of 12 patients had an AKI in the cefazolin and other antibiotic group, respectively (Table 4).

Discussion

Implementation of a screening tool at VHI allowed patients with documented penicillin allergy, including anaphylaxis, to receive cefazolin perioperatively. Broad spectrum antibiotics such as vancomycin, clindamycin, and fluoroquinolones are frequently used in patients allergic to penicillin, which can increase health care costs, risk of toxicity, and antimicrobial resistance.⁴ There was no incidence of allergic reactions noted in patients allergic to penicillin who received cefazolin. When comparing the incidence of observed allergic reactions to received perioperative antibiotics in the cefazolin group to previously published literature, no difference in allergy rates (P = .09) was found.³ Most antibiotics administered were congruent with ASHP guideline recommendations, and most patients eligible for cefazolin received it perioperatively.

Similar to this study, Goodman et al concluded that cefazolin appears to be a safe regimen in patients with documented penicillin anaphylactic reaction for surgical prophylaxis with only 1 (0.2%) potential allergic reaction.⁶ Patients who received cefazolin perioperatively had a statistically significant decrease in SSI rates. There were no clinically or statistically significant differences found between the proportion of allergic reactions or ADRs when compared to alternative antibiotics. Lessard et al concluded that a pharmacist-led interdisciplinary collaborative practice agreement increased cefazolin use in patients allergic to penicillin, including those with urticaria and anaphylaxis, with no reported ADRs.⁷ This study further demonstrated the safety of cefazolin use in patients with anaphylaxis to penicillin.

Limitations

This study’s single-center, retrospective design, patient population, and small sample size limit the generalizability of its results. The data collected are dependent on documentation in the chart. No ADRs were reported from the antibiotics patients received perioperatively. When considering safety data, information such as serum creatinine were available only in CPRS and some patients did not receive a postprocedure serum creatinine level. Additionally, this study did not investigate whether there was an increase in preferred preoperative antimicrobial prophylaxis after implementation of this protocol.

Conclusions

The results of this study support the use of cefazolin perioperatively in patients allergic to penicillin, including those with a history of anaphylaxis. Additional research should be conducted to validate data given the low incidence of ADRs. The primary outcome did not reach statistical significance, but the results may be clinically significant from a stewardship and safety perspective. VHI continues to use the screening tool described in this article.

Given its safety profile and bactericidal activity against the predominant organisms causing surgical site infections (SSIs), cefazolin remains the most popular choice for surgical prophylaxis.¹ Cefazolin offers protection against the pathogens most likely to contaminate the surgical site while minimizing inappropriate methicillin- resistant Staphylococcus aureus coverage that occurs with alternatives such as vancomycin and clindamycin. Documented allergies to Β-lactam antibiotics have historically forced clinicians to avoid the use of cephalosporins due to the potential risk of cross-reactivity. True type 1 (immunoglobin E [IgE]-mediated) cross-allergic reactions between penicillin and cephalosporins are rare, and previously reported data indicate cross-reactivity as a result of antibody recognition is more closely related to the side-chain identity rather than the Β-lactam ring.^2,3

About 10% of US patients report having a penicillin allergy; however, < 1% of the population has a true IgE-mediated allergic reaction.⁴ Previous research that has challenged penicillin allergies with cefazolin for surgical prophylaxis has reported minimal rates of allergic reactions.^2-5

In previous trials, patients with a history of delayed skin reactions, such as Stevens-Johnson syndrome (SJS), toxic epidermal necrolysis (TEN), and drug reaction with eosinophilia and systemic symptoms (DRESS), were excluded. Additionally, patients with an allergy to cefazolin including those with urticaria, angioedema, bronchospasm, or anaphylaxis, were excluded from perioperative retrial of cefazolin. Grant et al found that cefazolin can be safely given to patients with IgE-mediated reactions to penicillin and other cephalosporins due to a structurally different side chain.³

In January 2023, the Veteran Health Indiana (VHI) pharmacy team in conjunction with surgery, infectious disease, and anesthesiology, implemented a screening tool as an amendment to perioperative antibiotic guidance to help determine which patients with a documented penicillin allergy could be candidates for perioperative cefazolin. The implemented screening tool (Allergy Clarification for Cefazolin Evidence-Based Prescribing Tool) has been described by Lam et al, who reported that an increased proportion of patients with documented penicillin allergy received cefazolin without more adverse drug reactions (ADRs).⁵ Patients with a Β-lactam allergy were eligible to receive cefazolin unless the ADR was SJS, TEN, or DRESS, or the offending agent was cefazolin and the patient experienced urticaria, angioedema, bronchospasm, or anaphylaxis. If the reaction was not from cefazolin or was unknown, patients were eligible to receive cefazolin (Figure).

To date, minimal data exist to evaluate the incidence of ADRs when cefazolin is given perioperatively to patients with a previously documented penicillin allergy. The purpose of this study was to evaluate the incidence of allergic ADRs in patients who had a documented penicillin allergy and received periprocedural antibiotics.

Methods

This single-center, retrospective chart review used the US Department of Veterans Affairs (VA) Computerized Patient Record System (CPRS) to identify patients with a documented penicillin allergy who underwent an operation and received periprocedural antibiotics between February 1, 2023, and January 31, 2024. This study was reviewed and approved by the Indiana University Health Institutional Review Board and the VHI Research and Development Committee.

Patients were enrolled if they were aged ≥ 18 years, had a documented penicillin allergy, underwent a surgical intervention, and received perioperative antibiotics during the study period. Patients were excluded if they had a documented penicillin allergy resulting in severe delayed skin reactions (ie, SJS, TEN, or DRESS). These criteria produced 197 surgical procedures. Data were collected for each surgical procedure, so patients could be included more than once. Patient history of allergic reaction to penicillin was obtained through CPRS.

The primary endpoint was the percentage of allergic ADRs in patients with penicillin allergies receiving cefazolin perioperatively. Secondary outcomes included the appropriateness of the antibiotic regimen in congruence with American System of Health Pharmacists (ASHP) recommendations, incidence of SSIs within 30 days of the procedure, incidence of ADRs in those with a history of anaphylaxis vs nonanaphylaxis allergy, incidence of allergic reaction requiring pharmacologic and nonpharmacologic interventions, and incidence of acute kidney injury (AKI). AKI was defined as an increase in serum creatinine by ≥ 0.3 mg/dL within 48 hours or an increase in serum creatinine to ≥ 1.5 times baseline.

Demographic data included sex, age, race, preoperative serum creatinine, and postoperative serum creatinine. Anaphylaxis was defined as an acute onset of illness (within minutes to several hours) with involvement of skin, mucosal tissue, or both involving either respiratory compromise or reduced blood pressures. Allergic reactions were defined as facial, tongue, throat, airway, lip, mouth, periorbital, or eye swelling, urticaria, angioedema, dyspnea, anaphylaxis, or a positive penicillin skin test. Additionally, data collected included the description and severity of postprophylactic antibiotic reaction, antibiotic choice, interventions required for the allergic reaction, SSI occurrence, date of SSI, operating specialty, and postoperative change in renal function.

Descriptive statistics, including mean, SD, and percentages were reported for baseline characteristics of the study population. Percentages were used to demonstrate the differences in primary and secondary outcomes for each study group. Fisher exact tests were used for incidence of ADRs in patients with penicillin allergy who received cefazolin and reported incidence of SSIs.

Results

A total of 197 surgical procedures in patients with a documented penicillin allergy were included; 127 procedures used cefazolin perioperatively, 3 procedures used cefazolin plus gentamicin, and 67 procedures used other antibiotics. Most patients were White (n = 160; 81.2%), male (n = 158; 80.2%), and had a mean age of 64.9 years. Urology was the most common surgical specialty (n = 59; 29.9%) (Table 1). Of the 16 patients with documented penicillin anaphylaxis reaction, 8 received cefazolin and 8 received a different antibiotic. A total of 181 patients reported a nonanaphylaxis allergy. One hundred fifty-one patients (68.6%) reported a reaction history of hives, rash, or swelling (Table 2). Patients could report ≥ 1 reaction. The most prevalent antibiotics used were cefazolin, which was used by 130 patients (61.3%), and clindamycin which was used by 33 patients (15.6%) (Table 3). Patients could receive ≥ 1 antibiotic.

For the primary outcome, the incidence of allergic reactions in patients allergic to penicillin, there was no incidence of allergic reactions in either the cefazolin or other group. Given the absence of reactions, no interventions were required.

There were no ADRs in those with history of anaphylaxis or nonanaphylaxis allergy. In the cefazolin group, 126 of 127 surgical procedure regimens (99.2%) were congruent with ASHP recommendations, all 3 surgical procedures regimens in the cefazolin plus gentamicin group were congruent with ASHP recommendations, and 58 of 67 surgical procedure regimens (86.6%) in the other antibiotic group were congruent with ASHP recommendations. None of the 127 patients in the cefazolin group or of the 3 patients in the cefazolin plus gentamicin group reported an SSI, and 3 of 67 patients (4.5%) had an SSI in the other antibiotic group. One procedure that resulted in SSI was not congruent with ASHP recommendations. Twenty-four patients had 2 serum creatinine levels drawn within 48 hours of surgery. One of 12 patients (8.3%) and 0 of 12 patients had an AKI in the cefazolin and other antibiotic group, respectively (Table 4).

Discussion

Implementation of a screening tool at VHI allowed patients with documented penicillin allergy, including anaphylaxis, to receive cefazolin perioperatively. Broad spectrum antibiotics such as vancomycin, clindamycin, and fluoroquinolones are frequently used in patients allergic to penicillin, which can increase health care costs, risk of toxicity, and antimicrobial resistance.⁴ There was no incidence of allergic reactions noted in patients allergic to penicillin who received cefazolin. When comparing the incidence of observed allergic reactions to received perioperative antibiotics in the cefazolin group to previously published literature, no difference in allergy rates (P = .09) was found.³ Most antibiotics administered were congruent with ASHP guideline recommendations, and most patients eligible for cefazolin received it perioperatively.

Similar to this study, Goodman et al concluded that cefazolin appears to be a safe regimen in patients with documented penicillin anaphylactic reaction for surgical prophylaxis with only 1 (0.2%) potential allergic reaction.⁶ Patients who received cefazolin perioperatively had a statistically significant decrease in SSI rates. There were no clinically or statistically significant differences found between the proportion of allergic reactions or ADRs when compared to alternative antibiotics. Lessard et al concluded that a pharmacist-led interdisciplinary collaborative practice agreement increased cefazolin use in patients allergic to penicillin, including those with urticaria and anaphylaxis, with no reported ADRs.⁷ This study further demonstrated the safety of cefazolin use in patients with anaphylaxis to penicillin.

Limitations

This study’s single-center, retrospective design, patient population, and small sample size limit the generalizability of its results. The data collected are dependent on documentation in the chart. No ADRs were reported from the antibiotics patients received perioperatively. When considering safety data, information such as serum creatinine were available only in CPRS and some patients did not receive a postprocedure serum creatinine level. Additionally, this study did not investigate whether there was an increase in preferred preoperative antimicrobial prophylaxis after implementation of this protocol.

Conclusions

The results of this study support the use of cefazolin perioperatively in patients allergic to penicillin, including those with a history of anaphylaxis. Additional research should be conducted to validate data given the low incidence of ADRs. The primary outcome did not reach statistical significance, but the results may be clinically significant from a stewardship and safety perspective. VHI continues to use the screening tool described in this article.

Roughly 8.6% of the 17.4 million US veterans live in poverty. About 11.1% are considered food insecure (ie, unable to acquire adequate food for ≥1 household members), with another 5.3% considered very food insecure (ie, eating patterns of ≥1 household members were disrupted and their food intake was reduced at least some time during the year). Compared with nonveterans, veterans are 7.4% more likely to be food insecure.¹ This high prevalence of food insecurity and poverty has a negative impact on veteran diets.

Veterans’ diets contained more added sugars and solid fats and scored lower compared with nonveterans when assessed for diet quality with the Healthy Eating Index.² Veterans have a higher prevalence of diet-related chronic disease, including diabetes, hypertension, and obesity compared with the nonveterans.^3-5 Given the critical role of diet in health and disease risk, enhancing diet quality among veterans has garnered significant attention and calls to action.^2,6,7 While there are many factors that contribute to diet, any veteran can receive a consultation or self-refer to receive nutrition counseling effective for improving diet quality, within the US Department of Veterans Affairs (VA).

The NOVA food classification system describes diet quality by categorizing food items by processing methods and ingredients into 4 food groups.⁸ The first is unprocessed and minimally processed items (MPFs) such as fresh fruits, vegetables, and meats. MPFs consist of whole foods which can also be minimally processed (eg, chopping, drying, grinding, heating, chilling). Culinary processed foods (CPFs) are processed foods for cooking (eg, salt, butter, and vinegar) and are typically eaten in small quantities along with MPFs. Processed foods (PRFs) include canned and smoked foods, while ultra-processed foods (UPFs) are distinguished by industrial ingredients, requiring specialized tools and processing techniques, and hyper-palatability related to color, flavor, and packaging.⁸ Examples of UPFs include mass-produced breads found at grocery stores, prepackaged snacks and meals, and hydrogenated oils. UPF consumption is associated with higher risk for negative cardiometabolic outcomes, common mental disorders, and all-cause mortality.⁹ To date, only a study by Powell et al has used the NOVA classification system in a veteran population, and it was limited to a comparison of the price of UPFs and veteran body mass index (BMI).¹⁰ Therefore, it remains unknown what percentage of total energy intake (TEI) comes from UPFs in the diets of veterans.

This study sought to quantify the proportion of TEI from UPFs among a sample of patients from the VA Phoenix Health Care System (VAPHCS). Results from a 2021 global meta-analysis reveal that the US and United Kingdom have the highest intakes of UPFs in the world.¹¹ Specifically, within the US, 15 studies with 234,890 participants reveal that the majority of TEI (about 55%) comes from UPFs.¹¹ We hypothesized that this veteran sample would have a higher proportion of TEI from UPFs, possibly due to a higher prevalence of poverty and food insecurity among veterans compared with nonveterans.¹ If the percentage of TEI coming from UPF is higher or even similar to nonveterans, further efforts to increase veterans’ use of the available nutritional services would be warranted to minimize nutrition-related disease among veterans.

Methods

This is a cross-sectional, secondary data analysis of baseline 3-day food records collected from 2017 to 2020 from 92 patients recruited at VAPHCS to participate in a whole-food plant-based diet study.¹² The original study was reviewed and approved by the VAPHCS Institutional Review Board (1593830). Recruitment methods included clinician recommendation, a recorded advertisement played while phone calls were on hold, and flyers distributed throughout VAPHCS. Patients were included if they were aged 18 to 90 years, had a BMI 25.1 to 39.9, had a diagnosis of nutrition-related chronic disease (hypertension, diabetes, or hyperlipidemia), an interest and desire to make a lifestyle change, active telephone contact information (either landline or cell phone), no contraindication to be on a whole-food plant-based diet, access to transportation and a functioning kitchen, ability to prepare meals independently, access to a computer or tablet with internet access, and a digital camera or smartphone. Exclusion criteria included significant unplanned weight loss within 6 months, uncontrolled insulin-dependent diabetes with a current hemoglobin A_1c > 9%, pregnancy/lactation, taking prescribed weight loss medication, currently following a diet (eg, plant-based diet, vegan, or medical weight loss program diet), celiac disease diagnosed within 6 months, end-stage hepatic disease or renal disease requiring dialysis, active cancer or receiving chemotherapy or radiation therapy, active alcohol or substance use disorder, history of eating disorders, fasting triglyceride level > 350 mg/dL, any psychological issues that prevent adherence, inability to speak English, limited mobility, and homeless or in housing with limited kitchen access. A baseline 3-day food record was collected from the participants and used in this secondary analysis.

Diet Analysis

Food records were analyzed using Esha Research Food Processor 4.0 to identify calorie and macronutrient information. To limit bias, food items were coded independently by 2 researchers into 4 food processing groups determined by the NOVA classification: MPF, CPF, PRF, and UPF.⁸ When possible, specific ingredient information was collected using internet searches for brand product websites. Initial coding had an 89% agreement rate for food item coding between the 2 researchers. As coding was done in duplicate, a third researcher resolved disagreements. The number of food items for each processing group was determined and the mean (SD) percentage of TEI for each NOVA group was provided across participants. A 1-way analysis of variance and Tukey Multiple Comparisons Test were used to determine significance between groups with an α = .05 using Prism V9.

Results

Of the 92 participants in the original study, only 79 met inclusion criteria and had baseline diet data. The 79 veterans had a mean (SD) age of 61 (13) years and 59 (75%) were male (Table 1). Mean (SD) TEI was 1921 (815) kcal. The mean (SD) percentage of calories from carbohydrate, fat, and protein were 46% (21%), 39% (20%), and 16% (6%), respectively (Table 2).

A mean (SD) of 36 (12) food items were analyzed from the 3-day food records. The majority of food items were UPFs (56%), 33% were MPFs, 8% were PRFs, and 3% were CPFs. In total, 75% of TEI came from UPFs (P < .001); only 14% of TEI came from minimally processed foods (Figure).

Discussion

To our knowledge, this is the first analysis of UPF consumption among US veterans. TEIs coming from UPFs appear to be about 20% higher among veterans compared to nonveterans: 75% vs 55%.¹¹ Coupled with high UPF consumption, MPFs (14%) and PRFs (9%) represent smaller sources of TEI among surveyed veterans. Top caloric sources of UPFs in the US include sandwiches (including burgers), sweet bakery products, savory snacks, pizza, sweetened beverages, and breads, rolls, and tortillas, and likely reflect the major sources of UPFs in the veteran diet.¹³ As the statistical comparison between the veteran data and nonveteran data is not feasible in the present study, a future study with a much larger sample size would be needed for a direct comparison.

While the exact cause of higher UPF consumption among sampled veterans remains unknown and likely multifactorial (eg, cost, food insecurity, access, cooking skills, nutrition knowledge), veterans can receive a consult or self-refer to a registered dietitian nutritionist (RDN) for nutrition education. Counseling has been shown to be an effective way to improve diet quality and increase daily fruit and vegetable intake.¹⁴ High consumption of UPFs, which are generally energy-dense and nutrient-poor, contributes to the low diet quality observed in veterans, and future research examining the relationship between UPF intake and overall diet quality among veterans is warranted.^2,15 As nutrition knowledge is associated with higher diet quality among veterans, increased use of nutrition services (ie, nutrition education or food supplement programs) has the potential to influence consumption of MPFs and decrease consumption of UPFs.¹⁶ Subsequently, UPF-targeted interventions developed by VA RDNs hold the promise to decrease consumption of UPFs and increase intake of MPFs and PRFs.

Veterans have a high prevalence of diabetes, hypertension, and obesity.⁹ The high UPF intake observed in this sample of veterans may increase the risk for these chronic diseases and overall mortality. The high percentage of TEI from UPFs among veterans is also of concern not only due to potential negative health outcomes, but also associated costs of treating veterans with multimorbidities.¹⁷ Targeting UPF intake via nutritional education may promote health and decrease the financial burden needed to support the health of veterans.

Improving veteran health and well-being, including enhancing health care accessibility in underserved areas, are pivotal objectives of the VA strategic plan for 2026 to 2030. Public policy aims to tackle food insecurity within the veteran population during the first 5 years of civilian life.¹⁸ In alignment with the White House Strategy on Hunger, Nutrition, and Health, VA established a Food Security Office (FSO) in 2023. The FSO mission is to use an interdisciplinary approach to provide resources to ensure veteran food security and create an environment where all veterans are food and nutrition secure.

Limitations

This study has several limitations. As the Food Processor software database does not include all brand items, similar brands were used to mirror the nutrient profile. While food records are common among veteran diet studies, accuracy may be reduced due to self-reporting bias.¹⁹ Different interpretation of the NOVA classification designation for various food items is possible, however, 89% of foods were coded the same by the research team which suggests high accuracy in food coding. Specific ingredient information was not collected from the 3-day food records; thus, these records were not produced in such a way to improve the accuracy of the NOVA classification designation. This study was limited by its small sample size (N = 79); although, this analysis is larger than other studies of UPF consumption in the US.^20,21 In addition, the generalizability of this study is limited as this population sample was from a single VA hospital and may not reflect the overall veteran population. Participants in this study were recruited only from those receiving VA care, thus their diet quality may not represent the quality consumed by veterans not participating in VA services. Further research on UPF consumption among veterans is warranted with a larger, more representative study sample size.

Conclusions

As this is the highest observed UPF intake documented in the US, these results should be of concern for the VA and its RDNs. More research is needed to better understand why UPF consumption is so high among veterans, what barriers veterans face to decreasing UPF consumption, and what intervention(s) veterans would welcome to improve their diet quality. Presently, veterans are provided with access to a variety of effective nutrition education and counseling options and should be encouraged to use these services. VA RDNs should be aware of the high intake of UPFs in the veteran population and familiarize themselves with education and counseling strategies that promote behavior change to replace UPFs with more nutrient-dense foods choices.

Roughly 8.6% of the 17.4 million US veterans live in poverty. About 11.1% are considered food insecure (ie, unable to acquire adequate food for ≥1 household members), with another 5.3% considered very food insecure (ie, eating patterns of ≥1 household members were disrupted and their food intake was reduced at least some time during the year). Compared with nonveterans, veterans are 7.4% more likely to be food insecure.¹ This high prevalence of food insecurity and poverty has a negative impact on veteran diets.

Veterans’ diets contained more added sugars and solid fats and scored lower compared with nonveterans when assessed for diet quality with the Healthy Eating Index.² Veterans have a higher prevalence of diet-related chronic disease, including diabetes, hypertension, and obesity compared with the nonveterans.^3-5 Given the critical role of diet in health and disease risk, enhancing diet quality among veterans has garnered significant attention and calls to action.^2,6,7 While there are many factors that contribute to diet, any veteran can receive a consultation or self-refer to receive nutrition counseling effective for improving diet quality, within the US Department of Veterans Affairs (VA).

The NOVA food classification system describes diet quality by categorizing food items by processing methods and ingredients into 4 food groups.⁸ The first is unprocessed and minimally processed items (MPFs) such as fresh fruits, vegetables, and meats. MPFs consist of whole foods which can also be minimally processed (eg, chopping, drying, grinding, heating, chilling). Culinary processed foods (CPFs) are processed foods for cooking (eg, salt, butter, and vinegar) and are typically eaten in small quantities along with MPFs. Processed foods (PRFs) include canned and smoked foods, while ultra-processed foods (UPFs) are distinguished by industrial ingredients, requiring specialized tools and processing techniques, and hyper-palatability related to color, flavor, and packaging.⁸ Examples of UPFs include mass-produced breads found at grocery stores, prepackaged snacks and meals, and hydrogenated oils. UPF consumption is associated with higher risk for negative cardiometabolic outcomes, common mental disorders, and all-cause mortality.⁹ To date, only a study by Powell et al has used the NOVA classification system in a veteran population, and it was limited to a comparison of the price of UPFs and veteran body mass index (BMI).¹⁰ Therefore, it remains unknown what percentage of total energy intake (TEI) comes from UPFs in the diets of veterans.

This study sought to quantify the proportion of TEI from UPFs among a sample of patients from the VA Phoenix Health Care System (VAPHCS). Results from a 2021 global meta-analysis reveal that the US and United Kingdom have the highest intakes of UPFs in the world.¹¹ Specifically, within the US, 15 studies with 234,890 participants reveal that the majority of TEI (about 55%) comes from UPFs.¹¹ We hypothesized that this veteran sample would have a higher proportion of TEI from UPFs, possibly due to a higher prevalence of poverty and food insecurity among veterans compared with nonveterans.¹ If the percentage of TEI coming from UPF is higher or even similar to nonveterans, further efforts to increase veterans’ use of the available nutritional services would be warranted to minimize nutrition-related disease among veterans.

Methods

This is a cross-sectional, secondary data analysis of baseline 3-day food records collected from 2017 to 2020 from 92 patients recruited at VAPHCS to participate in a whole-food plant-based diet study.¹² The original study was reviewed and approved by the VAPHCS Institutional Review Board (1593830). Recruitment methods included clinician recommendation, a recorded advertisement played while phone calls were on hold, and flyers distributed throughout VAPHCS. Patients were included if they were aged 18 to 90 years, had a BMI 25.1 to 39.9, had a diagnosis of nutrition-related chronic disease (hypertension, diabetes, or hyperlipidemia), an interest and desire to make a lifestyle change, active telephone contact information (either landline or cell phone), no contraindication to be on a whole-food plant-based diet, access to transportation and a functioning kitchen, ability to prepare meals independently, access to a computer or tablet with internet access, and a digital camera or smartphone. Exclusion criteria included significant unplanned weight loss within 6 months, uncontrolled insulin-dependent diabetes with a current hemoglobin A_1c > 9%, pregnancy/lactation, taking prescribed weight loss medication, currently following a diet (eg, plant-based diet, vegan, or medical weight loss program diet), celiac disease diagnosed within 6 months, end-stage hepatic disease or renal disease requiring dialysis, active cancer or receiving chemotherapy or radiation therapy, active alcohol or substance use disorder, history of eating disorders, fasting triglyceride level > 350 mg/dL, any psychological issues that prevent adherence, inability to speak English, limited mobility, and homeless or in housing with limited kitchen access. A baseline 3-day food record was collected from the participants and used in this secondary analysis.

Diet Analysis

Food records were analyzed using Esha Research Food Processor 4.0 to identify calorie and macronutrient information. To limit bias, food items were coded independently by 2 researchers into 4 food processing groups determined by the NOVA classification: MPF, CPF, PRF, and UPF.⁸ When possible, specific ingredient information was collected using internet searches for brand product websites. Initial coding had an 89% agreement rate for food item coding between the 2 researchers. As coding was done in duplicate, a third researcher resolved disagreements. The number of food items for each processing group was determined and the mean (SD) percentage of TEI for each NOVA group was provided across participants. A 1-way analysis of variance and Tukey Multiple Comparisons Test were used to determine significance between groups with an α = .05 using Prism V9.

Results

Of the 92 participants in the original study, only 79 met inclusion criteria and had baseline diet data. The 79 veterans had a mean (SD) age of 61 (13) years and 59 (75%) were male (Table 1). Mean (SD) TEI was 1921 (815) kcal. The mean (SD) percentage of calories from carbohydrate, fat, and protein were 46% (21%), 39% (20%), and 16% (6%), respectively (Table 2).

A mean (SD) of 36 (12) food items were analyzed from the 3-day food records. The majority of food items were UPFs (56%), 33% were MPFs, 8% were PRFs, and 3% were CPFs. In total, 75% of TEI came from UPFs (P < .001); only 14% of TEI came from minimally processed foods (Figure).

Discussion

To our knowledge, this is the first analysis of UPF consumption among US veterans. TEIs coming from UPFs appear to be about 20% higher among veterans compared to nonveterans: 75% vs 55%.¹¹ Coupled with high UPF consumption, MPFs (14%) and PRFs (9%) represent smaller sources of TEI among surveyed veterans. Top caloric sources of UPFs in the US include sandwiches (including burgers), sweet bakery products, savory snacks, pizza, sweetened beverages, and breads, rolls, and tortillas, and likely reflect the major sources of UPFs in the veteran diet.¹³ As the statistical comparison between the veteran data and nonveteran data is not feasible in the present study, a future study with a much larger sample size would be needed for a direct comparison.

While the exact cause of higher UPF consumption among sampled veterans remains unknown and likely multifactorial (eg, cost, food insecurity, access, cooking skills, nutrition knowledge), veterans can receive a consult or self-refer to a registered dietitian nutritionist (RDN) for nutrition education. Counseling has been shown to be an effective way to improve diet quality and increase daily fruit and vegetable intake.¹⁴ High consumption of UPFs, which are generally energy-dense and nutrient-poor, contributes to the low diet quality observed in veterans, and future research examining the relationship between UPF intake and overall diet quality among veterans is warranted.^2,15 As nutrition knowledge is associated with higher diet quality among veterans, increased use of nutrition services (ie, nutrition education or food supplement programs) has the potential to influence consumption of MPFs and decrease consumption of UPFs.¹⁶ Subsequently, UPF-targeted interventions developed by VA RDNs hold the promise to decrease consumption of UPFs and increase intake of MPFs and PRFs.

Veterans have a high prevalence of diabetes, hypertension, and obesity.⁹ The high UPF intake observed in this sample of veterans may increase the risk for these chronic diseases and overall mortality. The high percentage of TEI from UPFs among veterans is also of concern not only due to potential negative health outcomes, but also associated costs of treating veterans with multimorbidities.¹⁷ Targeting UPF intake via nutritional education may promote health and decrease the financial burden needed to support the health of veterans.

Improving veteran health and well-being, including enhancing health care accessibility in underserved areas, are pivotal objectives of the VA strategic plan for 2026 to 2030. Public policy aims to tackle food insecurity within the veteran population during the first 5 years of civilian life.¹⁸ In alignment with the White House Strategy on Hunger, Nutrition, and Health, VA established a Food Security Office (FSO) in 2023. The FSO mission is to use an interdisciplinary approach to provide resources to ensure veteran food security and create an environment where all veterans are food and nutrition secure.

Limitations

This study has several limitations. As the Food Processor software database does not include all brand items, similar brands were used to mirror the nutrient profile. While food records are common among veteran diet studies, accuracy may be reduced due to self-reporting bias.¹⁹ Different interpretation of the NOVA classification designation for various food items is possible, however, 89% of foods were coded the same by the research team which suggests high accuracy in food coding. Specific ingredient information was not collected from the 3-day food records; thus, these records were not produced in such a way to improve the accuracy of the NOVA classification designation. This study was limited by its small sample size (N = 79); although, this analysis is larger than other studies of UPF consumption in the US.^20,21 In addition, the generalizability of this study is limited as this population sample was from a single VA hospital and may not reflect the overall veteran population. Participants in this study were recruited only from those receiving VA care, thus their diet quality may not represent the quality consumed by veterans not participating in VA services. Further research on UPF consumption among veterans is warranted with a larger, more representative study sample size.

Conclusions

As this is the highest observed UPF intake documented in the US, these results should be of concern for the VA and its RDNs. More research is needed to better understand why UPF consumption is so high among veterans, what barriers veterans face to decreasing UPF consumption, and what intervention(s) veterans would welcome to improve their diet quality. Presently, veterans are provided with access to a variety of effective nutrition education and counseling options and should be encouraged to use these services. VA RDNs should be aware of the high intake of UPFs in the veteran population and familiarize themselves with education and counseling strategies that promote behavior change to replace UPFs with more nutrient-dense foods choices.

Roughly 8.6% of the 17.4 million US veterans live in poverty. About 11.1% are considered food insecure (ie, unable to acquire adequate food for ≥1 household members), with another 5.3% considered very food insecure (ie, eating patterns of ≥1 household members were disrupted and their food intake was reduced at least some time during the year). Compared with nonveterans, veterans are 7.4% more likely to be food insecure.¹ This high prevalence of food insecurity and poverty has a negative impact on veteran diets.

Veterans’ diets contained more added sugars and solid fats and scored lower compared with nonveterans when assessed for diet quality with the Healthy Eating Index.² Veterans have a higher prevalence of diet-related chronic disease, including diabetes, hypertension, and obesity compared with the nonveterans.^3-5 Given the critical role of diet in health and disease risk, enhancing diet quality among veterans has garnered significant attention and calls to action.^2,6,7 While there are many factors that contribute to diet, any veteran can receive a consultation or self-refer to receive nutrition counseling effective for improving diet quality, within the US Department of Veterans Affairs (VA).

The NOVA food classification system describes diet quality by categorizing food items by processing methods and ingredients into 4 food groups.⁸ The first is unprocessed and minimally processed items (MPFs) such as fresh fruits, vegetables, and meats. MPFs consist of whole foods which can also be minimally processed (eg, chopping, drying, grinding, heating, chilling). Culinary processed foods (CPFs) are processed foods for cooking (eg, salt, butter, and vinegar) and are typically eaten in small quantities along with MPFs. Processed foods (PRFs) include canned and smoked foods, while ultra-processed foods (UPFs) are distinguished by industrial ingredients, requiring specialized tools and processing techniques, and hyper-palatability related to color, flavor, and packaging.⁸ Examples of UPFs include mass-produced breads found at grocery stores, prepackaged snacks and meals, and hydrogenated oils. UPF consumption is associated with higher risk for negative cardiometabolic outcomes, common mental disorders, and all-cause mortality.⁹ To date, only a study by Powell et al has used the NOVA classification system in a veteran population, and it was limited to a comparison of the price of UPFs and veteran body mass index (BMI).¹⁰ Therefore, it remains unknown what percentage of total energy intake (TEI) comes from UPFs in the diets of veterans.

This study sought to quantify the proportion of TEI from UPFs among a sample of patients from the VA Phoenix Health Care System (VAPHCS). Results from a 2021 global meta-analysis reveal that the US and United Kingdom have the highest intakes of UPFs in the world.¹¹ Specifically, within the US, 15 studies with 234,890 participants reveal that the majority of TEI (about 55%) comes from UPFs.¹¹ We hypothesized that this veteran sample would have a higher proportion of TEI from UPFs, possibly due to a higher prevalence of poverty and food insecurity among veterans compared with nonveterans.¹ If the percentage of TEI coming from UPF is higher or even similar to nonveterans, further efforts to increase veterans’ use of the available nutritional services would be warranted to minimize nutrition-related disease among veterans.

Methods

This is a cross-sectional, secondary data analysis of baseline 3-day food records collected from 2017 to 2020 from 92 patients recruited at VAPHCS to participate in a whole-food plant-based diet study.¹² The original study was reviewed and approved by the VAPHCS Institutional Review Board (1593830). Recruitment methods included clinician recommendation, a recorded advertisement played while phone calls were on hold, and flyers distributed throughout VAPHCS. Patients were included if they were aged 18 to 90 years, had a BMI 25.1 to 39.9, had a diagnosis of nutrition-related chronic disease (hypertension, diabetes, or hyperlipidemia), an interest and desire to make a lifestyle change, active telephone contact information (either landline or cell phone), no contraindication to be on a whole-food plant-based diet, access to transportation and a functioning kitchen, ability to prepare meals independently, access to a computer or tablet with internet access, and a digital camera or smartphone. Exclusion criteria included significant unplanned weight loss within 6 months, uncontrolled insulin-dependent diabetes with a current hemoglobin A_1c > 9%, pregnancy/lactation, taking prescribed weight loss medication, currently following a diet (eg, plant-based diet, vegan, or medical weight loss program diet), celiac disease diagnosed within 6 months, end-stage hepatic disease or renal disease requiring dialysis, active cancer or receiving chemotherapy or radiation therapy, active alcohol or substance use disorder, history of eating disorders, fasting triglyceride level > 350 mg/dL, any psychological issues that prevent adherence, inability to speak English, limited mobility, and homeless or in housing with limited kitchen access. A baseline 3-day food record was collected from the participants and used in this secondary analysis.

Diet Analysis

Food records were analyzed using Esha Research Food Processor 4.0 to identify calorie and macronutrient information. To limit bias, food items were coded independently by 2 researchers into 4 food processing groups determined by the NOVA classification: MPF, CPF, PRF, and UPF.⁸ When possible, specific ingredient information was collected using internet searches for brand product websites. Initial coding had an 89% agreement rate for food item coding between the 2 researchers. As coding was done in duplicate, a third researcher resolved disagreements. The number of food items for each processing group was determined and the mean (SD) percentage of TEI for each NOVA group was provided across participants. A 1-way analysis of variance and Tukey Multiple Comparisons Test were used to determine significance between groups with an α = .05 using Prism V9.

Results

Of the 92 participants in the original study, only 79 met inclusion criteria and had baseline diet data. The 79 veterans had a mean (SD) age of 61 (13) years and 59 (75%) were male (Table 1). Mean (SD) TEI was 1921 (815) kcal. The mean (SD) percentage of calories from carbohydrate, fat, and protein were 46% (21%), 39% (20%), and 16% (6%), respectively (Table 2).

A mean (SD) of 36 (12) food items were analyzed from the 3-day food records. The majority of food items were UPFs (56%), 33% were MPFs, 8% were PRFs, and 3% were CPFs. In total, 75% of TEI came from UPFs (P < .001); only 14% of TEI came from minimally processed foods (Figure).

Discussion

To our knowledge, this is the first analysis of UPF consumption among US veterans. TEIs coming from UPFs appear to be about 20% higher among veterans compared to nonveterans: 75% vs 55%.¹¹ Coupled with high UPF consumption, MPFs (14%) and PRFs (9%) represent smaller sources of TEI among surveyed veterans. Top caloric sources of UPFs in the US include sandwiches (including burgers), sweet bakery products, savory snacks, pizza, sweetened beverages, and breads, rolls, and tortillas, and likely reflect the major sources of UPFs in the veteran diet.¹³ As the statistical comparison between the veteran data and nonveteran data is not feasible in the present study, a future study with a much larger sample size would be needed for a direct comparison.

While the exact cause of higher UPF consumption among sampled veterans remains unknown and likely multifactorial (eg, cost, food insecurity, access, cooking skills, nutrition knowledge), veterans can receive a consult or self-refer to a registered dietitian nutritionist (RDN) for nutrition education. Counseling has been shown to be an effective way to improve diet quality and increase daily fruit and vegetable intake.¹⁴ High consumption of UPFs, which are generally energy-dense and nutrient-poor, contributes to the low diet quality observed in veterans, and future research examining the relationship between UPF intake and overall diet quality among veterans is warranted.^2,15 As nutrition knowledge is associated with higher diet quality among veterans, increased use of nutrition services (ie, nutrition education or food supplement programs) has the potential to influence consumption of MPFs and decrease consumption of UPFs.¹⁶ Subsequently, UPF-targeted interventions developed by VA RDNs hold the promise to decrease consumption of UPFs and increase intake of MPFs and PRFs.

Veterans have a high prevalence of diabetes, hypertension, and obesity.⁹ The high UPF intake observed in this sample of veterans may increase the risk for these chronic diseases and overall mortality. The high percentage of TEI from UPFs among veterans is also of concern not only due to potential negative health outcomes, but also associated costs of treating veterans with multimorbidities.¹⁷ Targeting UPF intake via nutritional education may promote health and decrease the financial burden needed to support the health of veterans.

Improving veteran health and well-being, including enhancing health care accessibility in underserved areas, are pivotal objectives of the VA strategic plan for 2026 to 2030. Public policy aims to tackle food insecurity within the veteran population during the first 5 years of civilian life.¹⁸ In alignment with the White House Strategy on Hunger, Nutrition, and Health, VA established a Food Security Office (FSO) in 2023. The FSO mission is to use an interdisciplinary approach to provide resources to ensure veteran food security and create an environment where all veterans are food and nutrition secure.

Limitations

This study has several limitations. As the Food Processor software database does not include all brand items, similar brands were used to mirror the nutrient profile. While food records are common among veteran diet studies, accuracy may be reduced due to self-reporting bias.¹⁹ Different interpretation of the NOVA classification designation for various food items is possible, however, 89% of foods were coded the same by the research team which suggests high accuracy in food coding. Specific ingredient information was not collected from the 3-day food records; thus, these records were not produced in such a way to improve the accuracy of the NOVA classification designation. This study was limited by its small sample size (N = 79); although, this analysis is larger than other studies of UPF consumption in the US.^20,21 In addition, the generalizability of this study is limited as this population sample was from a single VA hospital and may not reflect the overall veteran population. Participants in this study were recruited only from those receiving VA care, thus their diet quality may not represent the quality consumed by veterans not participating in VA services. Further research on UPF consumption among veterans is warranted with a larger, more representative study sample size.

Conclusions

As this is the highest observed UPF intake documented in the US, these results should be of concern for the VA and its RDNs. More research is needed to better understand why UPF consumption is so high among veterans, what barriers veterans face to decreasing UPF consumption, and what intervention(s) veterans would welcome to improve their diet quality. Presently, veterans are provided with access to a variety of effective nutrition education and counseling options and should be encouraged to use these services. VA RDNs should be aware of the high intake of UPFs in the veteran population and familiarize themselves with education and counseling strategies that promote behavior change to replace UPFs with more nutrient-dense foods choices.

To the Editor:

Psoriasis is a chronic immune-mediated inflammatory skin disease that affects approximately 2% to 3% of the global population and an estimated 7.5 million adults in the United States.¹ The condition is characterized by recurrent episodes of erythematous scaly plaques driven by dysregulated immune responses, particularly involving the interleukin (IL) 23/T-helper (Th) 17 axis.² Although cutaneous symptoms are the most visible manifestation, psoriasis is a systemic disorder with broad multisystem involvement. Comorbidities include psoriatic arthritis, metabolic syndrome, cardiovascular disease, inflammatory bowel disease, depression, and anxiety.¹ These conditions contribute to a heightened risk for premature mortality, increased health care utilization, and an estimated direct cost burden exceeding $11 billion annually in the United States alone.³ Patients with moderate to severe disease frequently require systemic therapy, and long-term disease control is essential to prevent cumulative inflammatory damage and reduce associated morbidity.⁴

Globally, psoriasis prevalence and disease severity vary by geography, ethnicity, and environmental factors, with higher rates in Northern Europe and North America and lower reported prevalence in East Asia and ­sub-Saharan Africa.⁵ In lower-resource settings, access to advanced therapies is limited, and patients often are treated with less effective or more toxic systemic agents, such as methotrexate or cyclosporine.⁵ These disparities not only affect quality of life but also may influence comorbidity and malignancy patterns, underscoring the importance of studying biologic safety in diverse real-world populations.

Over the past decade, the therapeutic landscape for psoriasis has been transformed by biologic agents targeting specific immune pathways.⁶ Interleukin 17 inhibitors (eg, secukinumab, ixekizumab, brodalumab, bimekizumab) act by neutralizing IL-17A, IL-17F, or the IL-17 receptor, thereby reducing keratinocyte activation, neutrophil recruitment, and downstream cytokine production.⁶ Interleukin 23 inhibitors (eg, guselkumab, risankizumab, tildrakizumab) block the p19 subunit of IL-23, halting the expansion and maintenance of pathogenic Th17 cells.⁶ Ustekinumab, an IL-12/23 inhibitor, targets the shared p40 subunit of IL-12 and IL-23, attenuating both Th1 and Th17 signaling.⁶ These agents achieve rapid, durable skin clearance in a large proportion of patients, improve psoriatic arthritis symptoms, and generally are well tolerated, even with long-term use.⁶

Although efficacy is well established, the immunomodulatory nature of IL inhibitors raises theoretical concerns about malignancy risk. Immune surveillance plays a critical role in detecting and eliminating emerging tumor cells.⁷ Data from other systemic immunosuppressants, such as cyclosporine, show increased risks for certain cancers⁸; however, the IL-17 and IL-23 pathways have dual roles in cancer biology.⁷ In some tumor contexts, these cytokines promote carcinogenesis through angiogenesis, epithelial proliferation, and suppression of antitumor immunity; therefore, inhibiting these pathways could theoretically reduce cancer risk.⁷ The uncertainty around this risk-benefit balance has made malignancy a central consideration for dermatologists, particularly when initiating therapy in patients with a history of cancer or other risk factors.

The perception of malignancy risk can influence patient willingness to start biologics as well as physician prescribing patterns.⁹ Some clinicians opt for alternative therapies in individuals with a personal or family history of cancer despite limited direct evidence of harm from IL inhibitors. Conversely, a reassuring malignancy safety profile may support broader adoption of these therapies, especially in patients requiring lifelong disease control.⁹ Shared decision-making in this context requires robust, real-world evidence that accounts for both common and rare malignancy outcomes.

Randomized controlled trials of IL inhibitors have not demonstrated a consistent malignancy signal, but these studies often are underpowered for rare outcomes and limited by short follow-up durations, typically less than 1 year. They also frequently exclude high-risk populations, limiting generalizability.¹⁰ Observational studies using real-world data can address these gaps by including more diverse patient populations, longer observation windows, and larger sample sizes capable of detecting differences in uncommon outcomes.

The TriNetX Analytics Network (http://www.trinetx.com) offers a unique platform for large-scale, real-world pharmacoepidemiologic research. This federated database aggregates deidentified electronic health record data from more than 100 million patients across the United States and internationally, including at academic medical centers, integrated delivery networks, and community hospitals.⁴ Data contributors refresh their datasets regularly, ensuring near-contemporary representation of prescribing trends and clinical outcomes. Standardized terminology mapping, consistent International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) coding, and centralized data-quality checks enhance the reliability of analyses. Governance protocols and compliance with the Health Insurance Portability and Accountability Act deidentification standards further ensure ethical use of the data. The breadth and depth of the TriNetX network make it possible to evaluate not only common malignancies but also rare cancer types that smaller studies cannot assess with sufficient statistical power.

We performed a retrospective matched-cohort study, querying data from January 1, 2014, through December 31, 2024, using TriNetX to examine whether IL inhibitor exposure is associated with differences in incident malignancy risk among adults with psoriasis. Patients aged 18 years or older with a psoriasis diagnosis (ICD-10-CM code L40.x) and documented exposure to an IL-17, IL-23, or IL-12/23 inhibitor were eligible. Patients with a prior malignancy diagnosis were excluded to reduce prevalence bias. To ensure that malignancies were incident, we included only those diagnosed at least 1 day after initiation of an IL inhibitor.

The comparison cohort consisted of psoriasis patients without IL inhibitor exposure during their observation period. We used 1:1 propensity score matching based on age, sex, race, and ethnicity, applying a caliper of 0.1 to balance baseline characteristics and minimize demographic confounding. The index date for unexposed patients was randomly assigned within their observation period to align follow-up timing with exposed patients. Outcomes were identified by ICD-10-CM codes grouped by skin, hematologic, and solid-organ malignancies. Odds ratios (ORs) with 95% confidence intervals (CIs) were calculated, with statistical significance set at P<.05. Odds ratios were selected over hazard ratios due to variability in precise follow-up time capture and the primary analytic goal of comparing proportional odds within matched follow-up windows.

Propensity score matching was employed because it is a well-established technique in pharmacoepidemiology to mimic some of the balance achieved in randomized trials. By equating treatment and control groups on measured confounders, matching helps isolate the treatment effect, particularly important in nonrandomized datasets in which prescribing decisions may be influenced by baseline characteristics. Grouping cancers into clinically relevant categories allowed us to assess patterns of association, as some cancer types (eg, melanoma, lymphomas) may have pathophysiologic links to inflammatory pathways targeted by IL inhibitors.

The final cohort included 133,352 patients, with 66,676 in each group. The mean (SD) age was 49.3 (16.0) years, and demographic variables were well balanced after matching. The mean follow-up was approximately 3.8 years. Interleukin 17 inhibitors were the most frequently prescribed, followed by IL-23 inhibitors and ustekinumab. Baseline comorbidities such as cardiovascular disease, diabetes, and obesity were comparable between groups, reducing the likelihood of confounding from these factors.

Interleukin inhibitor exposure was associated with significantly reduced odds of several malignancies (eTable). Among skin cancers, melanoma risk was reduced by 36% (OR, 0.641; 95% CI, 0.534-0.77; P<.0001), basal cell carcinoma by 43% (OR, 0.565; 95% CI, 0.48-0.665; P<.0001), and squamous cell carcinoma by 18% (OR, 0.821; 95% CI, 0.676-0.996; P=.0452). Hematologic malignancies showed similar reductions, with non-Hodgkin lymphoma odds reduced by 35% (OR, 0.646; 95% CI, 0.512-0.815; P=.0002) and Hodgkin lymphoma by 50% (OR, 0.5; 95% CI, 0.292-0.855; P=.0098).

Protective associations also were observed for several solid tumors: lung (OR, 0.528; 95% CI, 0.452-0.617; P<.0001), liver (OR, 0.528; 95% CI, 0.399-0.698; P<.0001), pancreatic (OR, 0.65; 95% CI, 0.49-0.861; P=.0025), breast (OR, 0.663; 95% CI, 0.582-0.754; P<.0001), prostate (OR, 0.543; 95% CI, 0.468-0.629; P<.0001), colorectal (OR, 0.592; 95% CI, 0.414-0.846; P=.0036), colon (OR, 0.466; 95% CI, 0.375-0.579; P<.0001), and oropharyngeal (OR, 0.55; 95% CI, 0.327-0.925; P=.0222) cancers. Cervical cancer (OR, 0.604; 95% CI, 0.381-0.958; P=.0304) and anal cancer (OR, 0.4; 95% CI, 0.224-0.714; P=.0013) also showed significant reductions. Vaginal, vulvar, and penile cancers demonstrated no significant differences, likely due to their low incidence and limited statistical power.

The biological plausibility of these findings is supported by preclinical studies implicating IL-17 and IL-23 in tumor-promoting inflammation.¹¹ These cytokines can recruit myeloid-derived suppressor cells, promote angiogenesis, and facilitate tumor-immune evasion. Inhibition may shift the immune microenvironment toward enhanced tumor surveillance, reduce protumorigenic cytokine signaling, and normalize regulatory T-cell function.¹¹ These mechanisms could explain observed reductions in melanoma, lymphomas, and certain solid tumors.

Our results are consistent with several large registry studies showing no increased cancer incidence in IL inhibitor users and extend prior findings by demonstrating significant reductions in multiple cancer types.¹² The melanoma reduction contrasts with the findings in earlier biologic safety studies, possibly due to our larger sample size, broader geographic representation, and inclusion of multiple IL inhibitor classes.¹³ Similar reductions have not been consistently observed with tumor necrosis factor α inhibitors, which have different immunologic targets and a more complex malignancy safety history.¹⁴

Limitations of our study include the retrospective design, potential misclassification of cancer diagnoses, and lack of data on unmeasured confounders such as sun exposure, smoking, alcohol use, and family cancer history. Surveillance bias is possible, though it would likely bias toward higher, not lower, cancer detection in biologic users. Our mean follow-up period of 3.8 years may not be sufficient for cancers with long latency periods.

If replicated, our findings could have meaningful public health implications. Reassurance regarding malignancy safety may increase patient acceptance and physician confidence in prescribing IL inhibitors, particularly for patients requiring long-term therapy. From a payer perspective, the potential for reduced cancer incidence could translate into substantial cost savings over time, offsetting the high up-front cost of biologics. Additionally, these results may be relevant to other IL inhibitor indications, including psoriatic arthritis, ankylosing spondylitis, and inflammatory bowel disease, in which similar pathophysiologic mechanisms may be at play.

In conclusion, this large matched-cohort study found that IL inhibitor therapy in psoriasis was associated with significantly reduced odds of multiple malignancies, including melanoma, lymphomas, and several solid tumors. These findings contribute to the growing body of real-world evidence supporting the long-term safety of IL inhibitors and underscore the need for continued pharmacovigilance and mechanistic research.

To the Editor:

Psoriasis is a chronic immune-mediated inflammatory skin disease that affects approximately 2% to 3% of the global population and an estimated 7.5 million adults in the United States.¹ The condition is characterized by recurrent episodes of erythematous scaly plaques driven by dysregulated immune responses, particularly involving the interleukin (IL) 23/T-helper (Th) 17 axis.² Although cutaneous symptoms are the most visible manifestation, psoriasis is a systemic disorder with broad multisystem involvement. Comorbidities include psoriatic arthritis, metabolic syndrome, cardiovascular disease, inflammatory bowel disease, depression, and anxiety.¹ These conditions contribute to a heightened risk for premature mortality, increased health care utilization, and an estimated direct cost burden exceeding $11 billion annually in the United States alone.³ Patients with moderate to severe disease frequently require systemic therapy, and long-term disease control is essential to prevent cumulative inflammatory damage and reduce associated morbidity.⁴

Globally, psoriasis prevalence and disease severity vary by geography, ethnicity, and environmental factors, with higher rates in Northern Europe and North America and lower reported prevalence in East Asia and ­sub-Saharan Africa.⁵ In lower-resource settings, access to advanced therapies is limited, and patients often are treated with less effective or more toxic systemic agents, such as methotrexate or cyclosporine.⁵ These disparities not only affect quality of life but also may influence comorbidity and malignancy patterns, underscoring the importance of studying biologic safety in diverse real-world populations.

Over the past decade, the therapeutic landscape for psoriasis has been transformed by biologic agents targeting specific immune pathways.⁶ Interleukin 17 inhibitors (eg, secukinumab, ixekizumab, brodalumab, bimekizumab) act by neutralizing IL-17A, IL-17F, or the IL-17 receptor, thereby reducing keratinocyte activation, neutrophil recruitment, and downstream cytokine production.⁶ Interleukin 23 inhibitors (eg, guselkumab, risankizumab, tildrakizumab) block the p19 subunit of IL-23, halting the expansion and maintenance of pathogenic Th17 cells.⁶ Ustekinumab, an IL-12/23 inhibitor, targets the shared p40 subunit of IL-12 and IL-23, attenuating both Th1 and Th17 signaling.⁶ These agents achieve rapid, durable skin clearance in a large proportion of patients, improve psoriatic arthritis symptoms, and generally are well tolerated, even with long-term use.⁶

Although efficacy is well established, the immunomodulatory nature of IL inhibitors raises theoretical concerns about malignancy risk. Immune surveillance plays a critical role in detecting and eliminating emerging tumor cells.⁷ Data from other systemic immunosuppressants, such as cyclosporine, show increased risks for certain cancers⁸; however, the IL-17 and IL-23 pathways have dual roles in cancer biology.⁷ In some tumor contexts, these cytokines promote carcinogenesis through angiogenesis, epithelial proliferation, and suppression of antitumor immunity; therefore, inhibiting these pathways could theoretically reduce cancer risk.⁷ The uncertainty around this risk-benefit balance has made malignancy a central consideration for dermatologists, particularly when initiating therapy in patients with a history of cancer or other risk factors.

The perception of malignancy risk can influence patient willingness to start biologics as well as physician prescribing patterns.⁹ Some clinicians opt for alternative therapies in individuals with a personal or family history of cancer despite limited direct evidence of harm from IL inhibitors. Conversely, a reassuring malignancy safety profile may support broader adoption of these therapies, especially in patients requiring lifelong disease control.⁹ Shared decision-making in this context requires robust, real-world evidence that accounts for both common and rare malignancy outcomes.

Randomized controlled trials of IL inhibitors have not demonstrated a consistent malignancy signal, but these studies often are underpowered for rare outcomes and limited by short follow-up durations, typically less than 1 year. They also frequently exclude high-risk populations, limiting generalizability.¹⁰ Observational studies using real-world data can address these gaps by including more diverse patient populations, longer observation windows, and larger sample sizes capable of detecting differences in uncommon outcomes.

The TriNetX Analytics Network (http://www.trinetx.com) offers a unique platform for large-scale, real-world pharmacoepidemiologic research. This federated database aggregates deidentified electronic health record data from more than 100 million patients across the United States and internationally, including at academic medical centers, integrated delivery networks, and community hospitals.⁴ Data contributors refresh their datasets regularly, ensuring near-contemporary representation of prescribing trends and clinical outcomes. Standardized terminology mapping, consistent International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) coding, and centralized data-quality checks enhance the reliability of analyses. Governance protocols and compliance with the Health Insurance Portability and Accountability Act deidentification standards further ensure ethical use of the data. The breadth and depth of the TriNetX network make it possible to evaluate not only common malignancies but also rare cancer types that smaller studies cannot assess with sufficient statistical power.

We performed a retrospective matched-cohort study, querying data from January 1, 2014, through December 31, 2024, using TriNetX to examine whether IL inhibitor exposure is associated with differences in incident malignancy risk among adults with psoriasis. Patients aged 18 years or older with a psoriasis diagnosis (ICD-10-CM code L40.x) and documented exposure to an IL-17, IL-23, or IL-12/23 inhibitor were eligible. Patients with a prior malignancy diagnosis were excluded to reduce prevalence bias. To ensure that malignancies were incident, we included only those diagnosed at least 1 day after initiation of an IL inhibitor.

The comparison cohort consisted of psoriasis patients without IL inhibitor exposure during their observation period. We used 1:1 propensity score matching based on age, sex, race, and ethnicity, applying a caliper of 0.1 to balance baseline characteristics and minimize demographic confounding. The index date for unexposed patients was randomly assigned within their observation period to align follow-up timing with exposed patients. Outcomes were identified by ICD-10-CM codes grouped by skin, hematologic, and solid-organ malignancies. Odds ratios (ORs) with 95% confidence intervals (CIs) were calculated, with statistical significance set at P<.05. Odds ratios were selected over hazard ratios due to variability in precise follow-up time capture and the primary analytic goal of comparing proportional odds within matched follow-up windows.

Propensity score matching was employed because it is a well-established technique in pharmacoepidemiology to mimic some of the balance achieved in randomized trials. By equating treatment and control groups on measured confounders, matching helps isolate the treatment effect, particularly important in nonrandomized datasets in which prescribing decisions may be influenced by baseline characteristics. Grouping cancers into clinically relevant categories allowed us to assess patterns of association, as some cancer types (eg, melanoma, lymphomas) may have pathophysiologic links to inflammatory pathways targeted by IL inhibitors.

The final cohort included 133,352 patients, with 66,676 in each group. The mean (SD) age was 49.3 (16.0) years, and demographic variables were well balanced after matching. The mean follow-up was approximately 3.8 years. Interleukin 17 inhibitors were the most frequently prescribed, followed by IL-23 inhibitors and ustekinumab. Baseline comorbidities such as cardiovascular disease, diabetes, and obesity were comparable between groups, reducing the likelihood of confounding from these factors.

Interleukin inhibitor exposure was associated with significantly reduced odds of several malignancies (eTable). Among skin cancers, melanoma risk was reduced by 36% (OR, 0.641; 95% CI, 0.534-0.77; P<.0001), basal cell carcinoma by 43% (OR, 0.565; 95% CI, 0.48-0.665; P<.0001), and squamous cell carcinoma by 18% (OR, 0.821; 95% CI, 0.676-0.996; P=.0452). Hematologic malignancies showed similar reductions, with non-Hodgkin lymphoma odds reduced by 35% (OR, 0.646; 95% CI, 0.512-0.815; P=.0002) and Hodgkin lymphoma by 50% (OR, 0.5; 95% CI, 0.292-0.855; P=.0098).

Protective associations also were observed for several solid tumors: lung (OR, 0.528; 95% CI, 0.452-0.617; P<.0001), liver (OR, 0.528; 95% CI, 0.399-0.698; P<.0001), pancreatic (OR, 0.65; 95% CI, 0.49-0.861; P=.0025), breast (OR, 0.663; 95% CI, 0.582-0.754; P<.0001), prostate (OR, 0.543; 95% CI, 0.468-0.629; P<.0001), colorectal (OR, 0.592; 95% CI, 0.414-0.846; P=.0036), colon (OR, 0.466; 95% CI, 0.375-0.579; P<.0001), and oropharyngeal (OR, 0.55; 95% CI, 0.327-0.925; P=.0222) cancers. Cervical cancer (OR, 0.604; 95% CI, 0.381-0.958; P=.0304) and anal cancer (OR, 0.4; 95% CI, 0.224-0.714; P=.0013) also showed significant reductions. Vaginal, vulvar, and penile cancers demonstrated no significant differences, likely due to their low incidence and limited statistical power.

The biological plausibility of these findings is supported by preclinical studies implicating IL-17 and IL-23 in tumor-promoting inflammation.¹¹ These cytokines can recruit myeloid-derived suppressor cells, promote angiogenesis, and facilitate tumor-immune evasion. Inhibition may shift the immune microenvironment toward enhanced tumor surveillance, reduce protumorigenic cytokine signaling, and normalize regulatory T-cell function.¹¹ These mechanisms could explain observed reductions in melanoma, lymphomas, and certain solid tumors.

Our results are consistent with several large registry studies showing no increased cancer incidence in IL inhibitor users and extend prior findings by demonstrating significant reductions in multiple cancer types.¹² The melanoma reduction contrasts with the findings in earlier biologic safety studies, possibly due to our larger sample size, broader geographic representation, and inclusion of multiple IL inhibitor classes.¹³ Similar reductions have not been consistently observed with tumor necrosis factor α inhibitors, which have different immunologic targets and a more complex malignancy safety history.¹⁴

Limitations of our study include the retrospective design, potential misclassification of cancer diagnoses, and lack of data on unmeasured confounders such as sun exposure, smoking, alcohol use, and family cancer history. Surveillance bias is possible, though it would likely bias toward higher, not lower, cancer detection in biologic users. Our mean follow-up period of 3.8 years may not be sufficient for cancers with long latency periods.

If replicated, our findings could have meaningful public health implications. Reassurance regarding malignancy safety may increase patient acceptance and physician confidence in prescribing IL inhibitors, particularly for patients requiring long-term therapy. From a payer perspective, the potential for reduced cancer incidence could translate into substantial cost savings over time, offsetting the high up-front cost of biologics. Additionally, these results may be relevant to other IL inhibitor indications, including psoriatic arthritis, ankylosing spondylitis, and inflammatory bowel disease, in which similar pathophysiologic mechanisms may be at play.

In conclusion, this large matched-cohort study found that IL inhibitor therapy in psoriasis was associated with significantly reduced odds of multiple malignancies, including melanoma, lymphomas, and several solid tumors. These findings contribute to the growing body of real-world evidence supporting the long-term safety of IL inhibitors and underscore the need for continued pharmacovigilance and mechanistic research.

To the Editor:

Psoriasis is a chronic immune-mediated inflammatory skin disease that affects approximately 2% to 3% of the global population and an estimated 7.5 million adults in the United States.¹ The condition is characterized by recurrent episodes of erythematous scaly plaques driven by dysregulated immune responses, particularly involving the interleukin (IL) 23/T-helper (Th) 17 axis.² Although cutaneous symptoms are the most visible manifestation, psoriasis is a systemic disorder with broad multisystem involvement. Comorbidities include psoriatic arthritis, metabolic syndrome, cardiovascular disease, inflammatory bowel disease, depression, and anxiety.¹ These conditions contribute to a heightened risk for premature mortality, increased health care utilization, and an estimated direct cost burden exceeding $11 billion annually in the United States alone.³ Patients with moderate to severe disease frequently require systemic therapy, and long-term disease control is essential to prevent cumulative inflammatory damage and reduce associated morbidity.⁴

Globally, psoriasis prevalence and disease severity vary by geography, ethnicity, and environmental factors, with higher rates in Northern Europe and North America and lower reported prevalence in East Asia and ­sub-Saharan Africa.⁵ In lower-resource settings, access to advanced therapies is limited, and patients often are treated with less effective or more toxic systemic agents, such as methotrexate or cyclosporine.⁵ These disparities not only affect quality of life but also may influence comorbidity and malignancy patterns, underscoring the importance of studying biologic safety in diverse real-world populations.

Over the past decade, the therapeutic landscape for psoriasis has been transformed by biologic agents targeting specific immune pathways.⁶ Interleukin 17 inhibitors (eg, secukinumab, ixekizumab, brodalumab, bimekizumab) act by neutralizing IL-17A, IL-17F, or the IL-17 receptor, thereby reducing keratinocyte activation, neutrophil recruitment, and downstream cytokine production.⁶ Interleukin 23 inhibitors (eg, guselkumab, risankizumab, tildrakizumab) block the p19 subunit of IL-23, halting the expansion and maintenance of pathogenic Th17 cells.⁶ Ustekinumab, an IL-12/23 inhibitor, targets the shared p40 subunit of IL-12 and IL-23, attenuating both Th1 and Th17 signaling.⁶ These agents achieve rapid, durable skin clearance in a large proportion of patients, improve psoriatic arthritis symptoms, and generally are well tolerated, even with long-term use.⁶

Although efficacy is well established, the immunomodulatory nature of IL inhibitors raises theoretical concerns about malignancy risk. Immune surveillance plays a critical role in detecting and eliminating emerging tumor cells.⁷ Data from other systemic immunosuppressants, such as cyclosporine, show increased risks for certain cancers⁸; however, the IL-17 and IL-23 pathways have dual roles in cancer biology.⁷ In some tumor contexts, these cytokines promote carcinogenesis through angiogenesis, epithelial proliferation, and suppression of antitumor immunity; therefore, inhibiting these pathways could theoretically reduce cancer risk.⁷ The uncertainty around this risk-benefit balance has made malignancy a central consideration for dermatologists, particularly when initiating therapy in patients with a history of cancer or other risk factors.

The perception of malignancy risk can influence patient willingness to start biologics as well as physician prescribing patterns.⁹ Some clinicians opt for alternative therapies in individuals with a personal or family history of cancer despite limited direct evidence of harm from IL inhibitors. Conversely, a reassuring malignancy safety profile may support broader adoption of these therapies, especially in patients requiring lifelong disease control.⁹ Shared decision-making in this context requires robust, real-world evidence that accounts for both common and rare malignancy outcomes.

Randomized controlled trials of IL inhibitors have not demonstrated a consistent malignancy signal, but these studies often are underpowered for rare outcomes and limited by short follow-up durations, typically less than 1 year. They also frequently exclude high-risk populations, limiting generalizability.¹⁰ Observational studies using real-world data can address these gaps by including more diverse patient populations, longer observation windows, and larger sample sizes capable of detecting differences in uncommon outcomes.

The TriNetX Analytics Network (http://www.trinetx.com) offers a unique platform for large-scale, real-world pharmacoepidemiologic research. This federated database aggregates deidentified electronic health record data from more than 100 million patients across the United States and internationally, including at academic medical centers, integrated delivery networks, and community hospitals.⁴ Data contributors refresh their datasets regularly, ensuring near-contemporary representation of prescribing trends and clinical outcomes. Standardized terminology mapping, consistent International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) coding, and centralized data-quality checks enhance the reliability of analyses. Governance protocols and compliance with the Health Insurance Portability and Accountability Act deidentification standards further ensure ethical use of the data. The breadth and depth of the TriNetX network make it possible to evaluate not only common malignancies but also rare cancer types that smaller studies cannot assess with sufficient statistical power.

We performed a retrospective matched-cohort study, querying data from January 1, 2014, through December 31, 2024, using TriNetX to examine whether IL inhibitor exposure is associated with differences in incident malignancy risk among adults with psoriasis. Patients aged 18 years or older with a psoriasis diagnosis (ICD-10-CM code L40.x) and documented exposure to an IL-17, IL-23, or IL-12/23 inhibitor were eligible. Patients with a prior malignancy diagnosis were excluded to reduce prevalence bias. To ensure that malignancies were incident, we included only those diagnosed at least 1 day after initiation of an IL inhibitor.

The comparison cohort consisted of psoriasis patients without IL inhibitor exposure during their observation period. We used 1:1 propensity score matching based on age, sex, race, and ethnicity, applying a caliper of 0.1 to balance baseline characteristics and minimize demographic confounding. The index date for unexposed patients was randomly assigned within their observation period to align follow-up timing with exposed patients. Outcomes were identified by ICD-10-CM codes grouped by skin, hematologic, and solid-organ malignancies. Odds ratios (ORs) with 95% confidence intervals (CIs) were calculated, with statistical significance set at P<.05. Odds ratios were selected over hazard ratios due to variability in precise follow-up time capture and the primary analytic goal of comparing proportional odds within matched follow-up windows.

Propensity score matching was employed because it is a well-established technique in pharmacoepidemiology to mimic some of the balance achieved in randomized trials. By equating treatment and control groups on measured confounders, matching helps isolate the treatment effect, particularly important in nonrandomized datasets in which prescribing decisions may be influenced by baseline characteristics. Grouping cancers into clinically relevant categories allowed us to assess patterns of association, as some cancer types (eg, melanoma, lymphomas) may have pathophysiologic links to inflammatory pathways targeted by IL inhibitors.

The final cohort included 133,352 patients, with 66,676 in each group. The mean (SD) age was 49.3 (16.0) years, and demographic variables were well balanced after matching. The mean follow-up was approximately 3.8 years. Interleukin 17 inhibitors were the most frequently prescribed, followed by IL-23 inhibitors and ustekinumab. Baseline comorbidities such as cardiovascular disease, diabetes, and obesity were comparable between groups, reducing the likelihood of confounding from these factors.

Interleukin inhibitor exposure was associated with significantly reduced odds of several malignancies (eTable). Among skin cancers, melanoma risk was reduced by 36% (OR, 0.641; 95% CI, 0.534-0.77; P<.0001), basal cell carcinoma by 43% (OR, 0.565; 95% CI, 0.48-0.665; P<.0001), and squamous cell carcinoma by 18% (OR, 0.821; 95% CI, 0.676-0.996; P=.0452). Hematologic malignancies showed similar reductions, with non-Hodgkin lymphoma odds reduced by 35% (OR, 0.646; 95% CI, 0.512-0.815; P=.0002) and Hodgkin lymphoma by 50% (OR, 0.5; 95% CI, 0.292-0.855; P=.0098).

Protective associations also were observed for several solid tumors: lung (OR, 0.528; 95% CI, 0.452-0.617; P<.0001), liver (OR, 0.528; 95% CI, 0.399-0.698; P<.0001), pancreatic (OR, 0.65; 95% CI, 0.49-0.861; P=.0025), breast (OR, 0.663; 95% CI, 0.582-0.754; P<.0001), prostate (OR, 0.543; 95% CI, 0.468-0.629; P<.0001), colorectal (OR, 0.592; 95% CI, 0.414-0.846; P=.0036), colon (OR, 0.466; 95% CI, 0.375-0.579; P<.0001), and oropharyngeal (OR, 0.55; 95% CI, 0.327-0.925; P=.0222) cancers. Cervical cancer (OR, 0.604; 95% CI, 0.381-0.958; P=.0304) and anal cancer (OR, 0.4; 95% CI, 0.224-0.714; P=.0013) also showed significant reductions. Vaginal, vulvar, and penile cancers demonstrated no significant differences, likely due to their low incidence and limited statistical power.

The biological plausibility of these findings is supported by preclinical studies implicating IL-17 and IL-23 in tumor-promoting inflammation.¹¹ These cytokines can recruit myeloid-derived suppressor cells, promote angiogenesis, and facilitate tumor-immune evasion. Inhibition may shift the immune microenvironment toward enhanced tumor surveillance, reduce protumorigenic cytokine signaling, and normalize regulatory T-cell function.¹¹ These mechanisms could explain observed reductions in melanoma, lymphomas, and certain solid tumors.

Our results are consistent with several large registry studies showing no increased cancer incidence in IL inhibitor users and extend prior findings by demonstrating significant reductions in multiple cancer types.¹² The melanoma reduction contrasts with the findings in earlier biologic safety studies, possibly due to our larger sample size, broader geographic representation, and inclusion of multiple IL inhibitor classes.¹³ Similar reductions have not been consistently observed with tumor necrosis factor α inhibitors, which have different immunologic targets and a more complex malignancy safety history.¹⁴

Limitations of our study include the retrospective design, potential misclassification of cancer diagnoses, and lack of data on unmeasured confounders such as sun exposure, smoking, alcohol use, and family cancer history. Surveillance bias is possible, though it would likely bias toward higher, not lower, cancer detection in biologic users. Our mean follow-up period of 3.8 years may not be sufficient for cancers with long latency periods.

If replicated, our findings could have meaningful public health implications. Reassurance regarding malignancy safety may increase patient acceptance and physician confidence in prescribing IL inhibitors, particularly for patients requiring long-term therapy. From a payer perspective, the potential for reduced cancer incidence could translate into substantial cost savings over time, offsetting the high up-front cost of biologics. Additionally, these results may be relevant to other IL inhibitor indications, including psoriatic arthritis, ankylosing spondylitis, and inflammatory bowel disease, in which similar pathophysiologic mechanisms may be at play.

In conclusion, this large matched-cohort study found that IL inhibitor therapy in psoriasis was associated with significantly reduced odds of multiple malignancies, including melanoma, lymphomas, and several solid tumors. These findings contribute to the growing body of real-world evidence supporting the long-term safety of IL inhibitors and underscore the need for continued pharmacovigilance and mechanistic research.

To the Editor:

The widespread availability and popularity of ChatGPT (OpenAI) have sparked interest in its potential applications within various fields, including medical diagnostics.¹ In dermatology, large language models (LLMs) already are being cited as a possible way to reliably respond to common patient queries and produce concise patient education materials.^2,3 That being said, there is skepticism regarding the technology’s efficacy and reliability in producing accurate treatment plans, with variability among popular LLMs; for example, a recent study by Chau et al⁴ demonstrated that ChatGPT was best at providing specific and accurate information regarding patient-facing responses to questions about 5 dermatologic diagnoses compared to Google Bard (now rebranded as Google Gemini) and Bing AI (now rebranded as Microsoft Copilot), which more often produced inaccurate or nonspecific responses. Google Bard also declined to answer one prompt.⁴ Large language models also have been evaluated in diagnosing skin lesions. In 2024, SkinGPT-4 (a pretrained multimodel LLM developed by Zhou et al⁵) achieved just over 80% accuracy in interpreting images of skin lesions and was considered informative by 82.5% of board-certified dermatologists, demonstrating that LLMs may have the potential to become integrated into clinical practice.⁵

Our study aimed to evaluate the performance of GPT-4o (OpenAI)—a widely accessible, low-cost LLM—in diagnosing dermatologic conditions using the HAM10000 dataset, a well-curated collection of dermatoscopic images developed for training and benchmarking artificial intelligence (AI) algorithms.⁶ HAM10000 comprises images representing 7 distinct skin conditions: actinic keratoses (ak), basal cell carcinoma (bcc), benign keratosis (bk), dermatofibroma (df), melanoma (mel), melanocytic nevi (nv), and vascular skin lesions (vsl), providing a robust platform for multiclass classification assessment. We evaluated GPT-4o using 100 dermatoscopic images per condition to assess diagnostic accuracy, potential biases, and limitations in skin lesion identification. The HAM10000 dataset was selected because it offers a large standardized reference set of dermatoscopic (rather than conventional clinical) images commonly used in dermatologic AI research. GPT-4o was chosen due to its patient-friendly interface, widespread use, and prior reports suggesting greater reliability in skin lesion assessment compared with other LLMs.

One hundred images from each of the 7 dermatologic categories were randomly selected for use in our analysis in 2024. The images were selected by our data scientist (J.C.) through random sampling from the dataset. Each image was separately presented to GPT-4o without any preprocessing or modification alongside 2 prompts designed to evaluate the diagnostic capabilities of GPT-4o. Both prompts included the same list of 7 dermatologic conditions for answer choices but differed in contextual information, where prompt 1 provided patient demographic information and localization of the dermatological condition but prompt 2 did not provide these details (Table). No follow-up questions were presented.

For prompt 1, the confusion matrix showed a strong bias toward detecting mel and bcc, with high true positives (mel, 83%; bcc, 37%)(eFigure 1). This pattern possibly suggests a tendency to favor malignant labels (eg, mel, BCC) when uncertainty is present. Interestingly, df and vsl also had notable true positives (46% and 37%, respectively), which is unexpected for less critical conditions because the model’s correct classifications were uneven across benign lesions. Actinic keratoses and nv showed higher misclassification rates, suggesting the model struggled to distinguish them from other lesions.

As shown in eTable 1, prompt 1 exhibited the highest recall for mel at 0.83 but performed worse in precision (0.242) and specificity (0.567) compared to ak, which had an extremely low recall (0.03) but very high specificity (0.992) and moderate precision score (0.375). The highest precision score was seen with vsl (0.738), which also achieved high scores in specificity (0.982) and accuracy (0.88) and performed moderately well in recall (0.31). All performance metrics are reported as proportions (0-1.0), wherein 1.0 indicates 100.

For prompt 2, the second confusion matrix followed similar trends as prompt 1 but still differed in key areas (eFigure 2). Melanoma detection remained strong (true positives, 95%), while bcc shows slightly fewer true positives (24%). Vascular skin lesions improve in true positives (40%), and df dropped slightly (33%). The model continues to struggle with ak and nv, with notable misclassifications observed across other categories

Similar to prompt 1, prompt 2 achieved its highest recall for mel (0.95%), but demonstrated lower precision (0.223%) and specificity (0.488%) for this class. Prompt 2 also produced the highest accuracy for vascular skin lesions (0.90%). The highest specificity was observed for both bk and ak (0.992% each); however, ak again demonstrated the lowest recall, with a value of 0.01%.

A previous study utilizing a model of binary classification to distinguish between mel and benign dermatologic conditions demonstrated poor performance.¹ Additionally, prior studies have employed a less-strict, open-ended style question approach to examine ChatGPT’s ability to diagnose mel with limited efficacy.⁷ The HAM10000 dataset was specifically selected despite its limitations (including the absence of clinical images and limited diversity in skin tones) due to its comprehensive nature, robust annotation standards, and widespread acceptance in dermatologic AI research. Compared to the Diverse Dermatology Images dataset, which notably lacks skin tone diversity, HAM10000 provides a balanced representation of several dermatologic conditions crucial for multiclass classification tasks, making it suitable for benchmarking AI performance. This study aimed to eliminate these limitations by employing a multiclass classification approach; however, despite this switch, our results indicate continued and major limitations of the diagnostic capabilities of GPT-4o.

In its current form, GPT-4o appeared to demonstrate a clear accuracy bias toward correctly identifying specific and severe dermatologic conditions (eg, mel, bcc) but showed low and variable class-level performance for other categories (eg, ak, nv, df, vsl), with frequent misclassification into melanoma or basal cell carcinoma and low recall for some classes (eTables 1 and 2). This finding emphasized that GPT-4o currently lacks the reliability needed for real-life clinical applications in dermatology, as both binary and multiclass models fail to achieve consistent accurate performance across all skin conditions. Notably, GPT-4o may generate false-positive malignant classifications among patients due to its skew in predicted labels toward labeling benign lesions as malignant.

From the patient perspective, younger individuals may upload images of benign nevi only to unnecessarily fear a mel diagnosis after receiving GPT-4o results. Statistically, younger patients are less likely than older patients to have malignant lesions and more likely to instead present with common vsl or df—lesions that GPT-4o appears likely to identify correctly.⁸ For older users, however, the situation may differ. Beyond ak being misclassified as bcc, older patients also may encounter GPT-4o outputs that mislabel lesions as mel, raising concerns and heightening anxiety. Given the technology’s tendency to overestimate the risk of serious dermatologic conditions, this behavior poses a considerable challenge in its current state and may inadvertently intensify public anxiety around mel.

A notable limitation of our study was that, compared to publicly available datasets, the HAM10000 dataset includes only dermatoscopic images rather than a combination of clinical and dermatoscopic images. Furthermore, the HAM10000 dataset comprises images primarily from White patients, whereas other diverse databases (eg, the Diverse Dermatology Images dataset) may be more suitable for training AI algorithms to accurately diagnose skin lesions in individuals with a variety of skin tones.⁹

Ultimately, our results signal that major advancements in the design and training of LLMs such as GPT-4o are necessary before these systems can be integrated into dermatologic diagnostic decision-making to offer benefit rather than cause harm. Consulting a health care professional rather than relying solely on AI, which might otherwise lead to avoidable stress, unnecessary alarm, and potentially increased health care costs due to unwarranted follow-up and testing, should remain the recommended standard of care for patients suspecting a skin lesion.

To the Editor:

The widespread availability and popularity of ChatGPT (OpenAI) have sparked interest in its potential applications within various fields, including medical diagnostics.¹ In dermatology, large language models (LLMs) already are being cited as a possible way to reliably respond to common patient queries and produce concise patient education materials.^2,3 That being said, there is skepticism regarding the technology’s efficacy and reliability in producing accurate treatment plans, with variability among popular LLMs; for example, a recent study by Chau et al⁴ demonstrated that ChatGPT was best at providing specific and accurate information regarding patient-facing responses to questions about 5 dermatologic diagnoses compared to Google Bard (now rebranded as Google Gemini) and Bing AI (now rebranded as Microsoft Copilot), which more often produced inaccurate or nonspecific responses. Google Bard also declined to answer one prompt.⁴ Large language models also have been evaluated in diagnosing skin lesions. In 2024, SkinGPT-4 (a pretrained multimodel LLM developed by Zhou et al⁵) achieved just over 80% accuracy in interpreting images of skin lesions and was considered informative by 82.5% of board-certified dermatologists, demonstrating that LLMs may have the potential to become integrated into clinical practice.⁵

Our study aimed to evaluate the performance of GPT-4o (OpenAI)—a widely accessible, low-cost LLM—in diagnosing dermatologic conditions using the HAM10000 dataset, a well-curated collection of dermatoscopic images developed for training and benchmarking artificial intelligence (AI) algorithms.⁶ HAM10000 comprises images representing 7 distinct skin conditions: actinic keratoses (ak), basal cell carcinoma (bcc), benign keratosis (bk), dermatofibroma (df), melanoma (mel), melanocytic nevi (nv), and vascular skin lesions (vsl), providing a robust platform for multiclass classification assessment. We evaluated GPT-4o using 100 dermatoscopic images per condition to assess diagnostic accuracy, potential biases, and limitations in skin lesion identification. The HAM10000 dataset was selected because it offers a large standardized reference set of dermatoscopic (rather than conventional clinical) images commonly used in dermatologic AI research. GPT-4o was chosen due to its patient-friendly interface, widespread use, and prior reports suggesting greater reliability in skin lesion assessment compared with other LLMs.

One hundred images from each of the 7 dermatologic categories were randomly selected for use in our analysis in 2024. The images were selected by our data scientist (J.C.) through random sampling from the dataset. Each image was separately presented to GPT-4o without any preprocessing or modification alongside 2 prompts designed to evaluate the diagnostic capabilities of GPT-4o. Both prompts included the same list of 7 dermatologic conditions for answer choices but differed in contextual information, where prompt 1 provided patient demographic information and localization of the dermatological condition but prompt 2 did not provide these details (Table). No follow-up questions were presented.

For prompt 1, the confusion matrix showed a strong bias toward detecting mel and bcc, with high true positives (mel, 83%; bcc, 37%)(eFigure 1). This pattern possibly suggests a tendency to favor malignant labels (eg, mel, BCC) when uncertainty is present. Interestingly, df and vsl also had notable true positives (46% and 37%, respectively), which is unexpected for less critical conditions because the model’s correct classifications were uneven across benign lesions. Actinic keratoses and nv showed higher misclassification rates, suggesting the model struggled to distinguish them from other lesions.

As shown in eTable 1, prompt 1 exhibited the highest recall for mel at 0.83 but performed worse in precision (0.242) and specificity (0.567) compared to ak, which had an extremely low recall (0.03) but very high specificity (0.992) and moderate precision score (0.375). The highest precision score was seen with vsl (0.738), which also achieved high scores in specificity (0.982) and accuracy (0.88) and performed moderately well in recall (0.31). All performance metrics are reported as proportions (0-1.0), wherein 1.0 indicates 100.

For prompt 2, the second confusion matrix followed similar trends as prompt 1 but still differed in key areas (eFigure 2). Melanoma detection remained strong (true positives, 95%), while bcc shows slightly fewer true positives (24%). Vascular skin lesions improve in true positives (40%), and df dropped slightly (33%). The model continues to struggle with ak and nv, with notable misclassifications observed across other categories

Similar to prompt 1, prompt 2 achieved its highest recall for mel (0.95%), but demonstrated lower precision (0.223%) and specificity (0.488%) for this class. Prompt 2 also produced the highest accuracy for vascular skin lesions (0.90%). The highest specificity was observed for both bk and ak (0.992% each); however, ak again demonstrated the lowest recall, with a value of 0.01%.

A previous study utilizing a model of binary classification to distinguish between mel and benign dermatologic conditions demonstrated poor performance.¹ Additionally, prior studies have employed a less-strict, open-ended style question approach to examine ChatGPT’s ability to diagnose mel with limited efficacy.⁷ The HAM10000 dataset was specifically selected despite its limitations (including the absence of clinical images and limited diversity in skin tones) due to its comprehensive nature, robust annotation standards, and widespread acceptance in dermatologic AI research. Compared to the Diverse Dermatology Images dataset, which notably lacks skin tone diversity, HAM10000 provides a balanced representation of several dermatologic conditions crucial for multiclass classification tasks, making it suitable for benchmarking AI performance. This study aimed to eliminate these limitations by employing a multiclass classification approach; however, despite this switch, our results indicate continued and major limitations of the diagnostic capabilities of GPT-4o.

In its current form, GPT-4o appeared to demonstrate a clear accuracy bias toward correctly identifying specific and severe dermatologic conditions (eg, mel, bcc) but showed low and variable class-level performance for other categories (eg, ak, nv, df, vsl), with frequent misclassification into melanoma or basal cell carcinoma and low recall for some classes (eTables 1 and 2). This finding emphasized that GPT-4o currently lacks the reliability needed for real-life clinical applications in dermatology, as both binary and multiclass models fail to achieve consistent accurate performance across all skin conditions. Notably, GPT-4o may generate false-positive malignant classifications among patients due to its skew in predicted labels toward labeling benign lesions as malignant.

From the patient perspective, younger individuals may upload images of benign nevi only to unnecessarily fear a mel diagnosis after receiving GPT-4o results. Statistically, younger patients are less likely than older patients to have malignant lesions and more likely to instead present with common vsl or df—lesions that GPT-4o appears likely to identify correctly.⁸ For older users, however, the situation may differ. Beyond ak being misclassified as bcc, older patients also may encounter GPT-4o outputs that mislabel lesions as mel, raising concerns and heightening anxiety. Given the technology’s tendency to overestimate the risk of serious dermatologic conditions, this behavior poses a considerable challenge in its current state and may inadvertently intensify public anxiety around mel.

A notable limitation of our study was that, compared to publicly available datasets, the HAM10000 dataset includes only dermatoscopic images rather than a combination of clinical and dermatoscopic images. Furthermore, the HAM10000 dataset comprises images primarily from White patients, whereas other diverse databases (eg, the Diverse Dermatology Images dataset) may be more suitable for training AI algorithms to accurately diagnose skin lesions in individuals with a variety of skin tones.⁹

Ultimately, our results signal that major advancements in the design and training of LLMs such as GPT-4o are necessary before these systems can be integrated into dermatologic diagnostic decision-making to offer benefit rather than cause harm. Consulting a health care professional rather than relying solely on AI, which might otherwise lead to avoidable stress, unnecessary alarm, and potentially increased health care costs due to unwarranted follow-up and testing, should remain the recommended standard of care for patients suspecting a skin lesion.

To the Editor:

The widespread availability and popularity of ChatGPT (OpenAI) have sparked interest in its potential applications within various fields, including medical diagnostics.¹ In dermatology, large language models (LLMs) already are being cited as a possible way to reliably respond to common patient queries and produce concise patient education materials.^2,3 That being said, there is skepticism regarding the technology’s efficacy and reliability in producing accurate treatment plans, with variability among popular LLMs; for example, a recent study by Chau et al⁴ demonstrated that ChatGPT was best at providing specific and accurate information regarding patient-facing responses to questions about 5 dermatologic diagnoses compared to Google Bard (now rebranded as Google Gemini) and Bing AI (now rebranded as Microsoft Copilot), which more often produced inaccurate or nonspecific responses. Google Bard also declined to answer one prompt.⁴ Large language models also have been evaluated in diagnosing skin lesions. In 2024, SkinGPT-4 (a pretrained multimodel LLM developed by Zhou et al⁵) achieved just over 80% accuracy in interpreting images of skin lesions and was considered informative by 82.5% of board-certified dermatologists, demonstrating that LLMs may have the potential to become integrated into clinical practice.⁵

Our study aimed to evaluate the performance of GPT-4o (OpenAI)—a widely accessible, low-cost LLM—in diagnosing dermatologic conditions using the HAM10000 dataset, a well-curated collection of dermatoscopic images developed for training and benchmarking artificial intelligence (AI) algorithms.⁶ HAM10000 comprises images representing 7 distinct skin conditions: actinic keratoses (ak), basal cell carcinoma (bcc), benign keratosis (bk), dermatofibroma (df), melanoma (mel), melanocytic nevi (nv), and vascular skin lesions (vsl), providing a robust platform for multiclass classification assessment. We evaluated GPT-4o using 100 dermatoscopic images per condition to assess diagnostic accuracy, potential biases, and limitations in skin lesion identification. The HAM10000 dataset was selected because it offers a large standardized reference set of dermatoscopic (rather than conventional clinical) images commonly used in dermatologic AI research. GPT-4o was chosen due to its patient-friendly interface, widespread use, and prior reports suggesting greater reliability in skin lesion assessment compared with other LLMs.

One hundred images from each of the 7 dermatologic categories were randomly selected for use in our analysis in 2024. The images were selected by our data scientist (J.C.) through random sampling from the dataset. Each image was separately presented to GPT-4o without any preprocessing or modification alongside 2 prompts designed to evaluate the diagnostic capabilities of GPT-4o. Both prompts included the same list of 7 dermatologic conditions for answer choices but differed in contextual information, where prompt 1 provided patient demographic information and localization of the dermatological condition but prompt 2 did not provide these details (Table). No follow-up questions were presented.

For prompt 1, the confusion matrix showed a strong bias toward detecting mel and bcc, with high true positives (mel, 83%; bcc, 37%)(eFigure 1). This pattern possibly suggests a tendency to favor malignant labels (eg, mel, BCC) when uncertainty is present. Interestingly, df and vsl also had notable true positives (46% and 37%, respectively), which is unexpected for less critical conditions because the model’s correct classifications were uneven across benign lesions. Actinic keratoses and nv showed higher misclassification rates, suggesting the model struggled to distinguish them from other lesions.

As shown in eTable 1, prompt 1 exhibited the highest recall for mel at 0.83 but performed worse in precision (0.242) and specificity (0.567) compared to ak, which had an extremely low recall (0.03) but very high specificity (0.992) and moderate precision score (0.375). The highest precision score was seen with vsl (0.738), which also achieved high scores in specificity (0.982) and accuracy (0.88) and performed moderately well in recall (0.31). All performance metrics are reported as proportions (0-1.0), wherein 1.0 indicates 100.

For prompt 2, the second confusion matrix followed similar trends as prompt 1 but still differed in key areas (eFigure 2). Melanoma detection remained strong (true positives, 95%), while bcc shows slightly fewer true positives (24%). Vascular skin lesions improve in true positives (40%), and df dropped slightly (33%). The model continues to struggle with ak and nv, with notable misclassifications observed across other categories

Similar to prompt 1, prompt 2 achieved its highest recall for mel (0.95%), but demonstrated lower precision (0.223%) and specificity (0.488%) for this class. Prompt 2 also produced the highest accuracy for vascular skin lesions (0.90%). The highest specificity was observed for both bk and ak (0.992% each); however, ak again demonstrated the lowest recall, with a value of 0.01%.

A previous study utilizing a model of binary classification to distinguish between mel and benign dermatologic conditions demonstrated poor performance.¹ Additionally, prior studies have employed a less-strict, open-ended style question approach to examine ChatGPT’s ability to diagnose mel with limited efficacy.⁷ The HAM10000 dataset was specifically selected despite its limitations (including the absence of clinical images and limited diversity in skin tones) due to its comprehensive nature, robust annotation standards, and widespread acceptance in dermatologic AI research. Compared to the Diverse Dermatology Images dataset, which notably lacks skin tone diversity, HAM10000 provides a balanced representation of several dermatologic conditions crucial for multiclass classification tasks, making it suitable for benchmarking AI performance. This study aimed to eliminate these limitations by employing a multiclass classification approach; however, despite this switch, our results indicate continued and major limitations of the diagnostic capabilities of GPT-4o.

In its current form, GPT-4o appeared to demonstrate a clear accuracy bias toward correctly identifying specific and severe dermatologic conditions (eg, mel, bcc) but showed low and variable class-level performance for other categories (eg, ak, nv, df, vsl), with frequent misclassification into melanoma or basal cell carcinoma and low recall for some classes (eTables 1 and 2). This finding emphasized that GPT-4o currently lacks the reliability needed for real-life clinical applications in dermatology, as both binary and multiclass models fail to achieve consistent accurate performance across all skin conditions. Notably, GPT-4o may generate false-positive malignant classifications among patients due to its skew in predicted labels toward labeling benign lesions as malignant.

From the patient perspective, younger individuals may upload images of benign nevi only to unnecessarily fear a mel diagnosis after receiving GPT-4o results. Statistically, younger patients are less likely than older patients to have malignant lesions and more likely to instead present with common vsl or df—lesions that GPT-4o appears likely to identify correctly.⁸ For older users, however, the situation may differ. Beyond ak being misclassified as bcc, older patients also may encounter GPT-4o outputs that mislabel lesions as mel, raising concerns and heightening anxiety. Given the technology’s tendency to overestimate the risk of serious dermatologic conditions, this behavior poses a considerable challenge in its current state and may inadvertently intensify public anxiety around mel.

A notable limitation of our study was that, compared to publicly available datasets, the HAM10000 dataset includes only dermatoscopic images rather than a combination of clinical and dermatoscopic images. Furthermore, the HAM10000 dataset comprises images primarily from White patients, whereas other diverse databases (eg, the Diverse Dermatology Images dataset) may be more suitable for training AI algorithms to accurately diagnose skin lesions in individuals with a variety of skin tones.⁹

Ultimately, our results signal that major advancements in the design and training of LLMs such as GPT-4o are necessary before these systems can be integrated into dermatologic diagnostic decision-making to offer benefit rather than cause harm. Consulting a health care professional rather than relying solely on AI, which might otherwise lead to avoidable stress, unnecessary alarm, and potentially increased health care costs due to unwarranted follow-up and testing, should remain the recommended standard of care for patients suspecting a skin lesion.

To the Editor:

It is known that dermatologist evaluation of skin conditions in hospitalized patients confers enhanced diagnostic accuracy, timely and appropriate treatment, and an overall reduction in readmissions compared to assessments by nondermatology hospitalists.¹ Dermatology consultations have been shown to alter diagnoses in up to 50% of cases and lead to changes in management in nearly 75% of cases, even for prevalent dermatologic conditions such as drug rashes, cellulitis, and stasis dermatitis.^1,2 Previous studies have observed a multiday reduction in length of hospital stay, a 10-fold reduction in readmission rate, and lower 30-day mortality, all leading to a reduction in patient morbidity and costs to both the patient and the health care system.^3,4 Despite these benefits, there has been a decrease in the number of dermatologists providing inpatient services and a reduction in medical centers offering dermatology consultations over the past several years.⁵ To better appreciate current trends of declining dermatology inpatient and consultative services within our region, we evaluated the availability of dermatology care at hospitals across Virginia.

A simple telephone survey was conducted across community hospitals in Virginia wherein medical staff administrators were asked to provide details regarding their dermatology staffing. The following figures were collected: number of dermatologists on staff, number of dermatologists with consulting privileges, number of affiliated dermatologists, and number of advanced-practice dermatology providers. Follow-up calls were carried out to elaborate on how dermatologists (when available) were integrated into inpatient care workflow and made accessible to hospitalists and emergency medicine departments. Academic centers, military hospitals, and specialty hospitals were excluded from the survey.

To better appreciate the relationships between hospital and population characteristics and the availability of dermatology care, publicly available data were collected on hospital bed counts and regional population density for each facility.^6-9 Spearman rank correlation analyses were conducted in Microsoft Excel to evaluate the association between the number of dermatologists on staff, number of consulting dermatologists, staffed inpatient beds, and population size.

Sixty-four hospitals—more than 70% of the 90 eligible community hospitals—responded to the survey between May and August 2024 and were included in the study. On-staff dermatologists were present at 8 (12.5%) of the hospitals surveyed; of these, 4 (50.0%) hospitals had between 1 and 5 dermatologists, 3 (37.5%) had between 6 and 10 dermatologists, and 1 (12.5%) had between 11 and 15 dermatologists. An additional 4 (6.3%) hospitals provided consultative dermatology services from outside dermatology clinics. Urban hospitals accounted for 9 of 12 (75%) hospitals offering in-house dermatology services, either through on-staff physicians or consultations with clinic-based providers.

Based on Spearman rank correlation analysis, there was a positive correlation between the number of dermatologists on staff and the number of staffed hospital beds (r=0.61; P <.001). Similarly, there was a positive correlation between the number of dermatologists on staff and the population density of the affiliated region (r=0.58; P <.001). Finally, there was a positive correlation between the number of dermatologists on staff and the number of available consulting dermatologists (r=0.89; P <.001).

At facilities with only consultative dermatology services accessible, there often was no formal dermatology team or department present. Rather, the hospitals relied on a loosely affiliated network of dermatology providers or navigated inpatient dermatology needs almost exclusively via internal medicine hospitalists or emergency medicine physicians. When available, dermatology support from dermatology physicians often was provided through teledermatology platforms. Although teledermatology has a large role in increasing access to care within underserved areas, its reliance on images and second-hand case descriptions can limit the provider’s ability to perform a comprehensive examination and assessment. Moreover, it was noted that few hospital representatives could offer clarity on how dermatologists were integrated into the inpatient setting. It remained unclear whether dermatologists were practically accessible to the inpatient care teams in a structured manner.

The uneven distribution and limited availability of dermatology inpatient care in Virginia reflect national trends and underscore ongoing access issues for patients. Without intentional intervention, these trends are expected to continue, contributing to a glaring gap in hospital services as well as to patient morbidity and mortality. The correlation data obtained in our study further qualify these disparities. The positive correlations between dermatologist availability and hospital size and population density suggest that larger, more urban facilities are more likely to offer inpatient dermatology care, whether through staffing or consultation. This relationship is not unexpected, given the greater financial resources and specialist networks available to facilities with large patient volumes. This suggests that dermatology care is shaped by institutional capacity and geographic leverage rather than clinical need, reinforcing existing disparities.

Importantly, it should be noted that the data may overestimate the true availability of dermatologists to these patients. As revealed via follow-up survey calls, respondent facilities that provided dermatology via consultative services often did not have a defined structure for integrating this care into the inpatient workflow. In some instances, dermatologists were technically affiliated with the hospital but had varying levels of practical interaction with the hospital providers and their patients. Administrative staff's differing awareness regarding dermatology interaction with the hospital facility may reveal systemic underutilization and opportunities to improve coordination to achieve the greatest benefit from dermatology services. These observations are further informed by the scope of our study, which focused specifically on community hospitals. The exclusion of academic and military institutions—and the tendency of these to exist in more densely populated areas—may have limited how broadly our findings reflect nationwide dermatology access by omitting more established dermatology departments and specialty care. As a result, regional variations in predominant facility type should be considered when interpreting the implications of these results beyond Virginia’s community hospital system.

In response to access limitations and differences in availability, facilities are turning to integrated teledermatology as a valuable tool to expand the reach of specialist care, particularly in rural or resource-limited settings. This modality acts as an important step toward improving equity in care by beginning to bridge geographic gaps; however, along with these logistical advantages, teledermatology also confers diagnostic limitations and clinical trade-offs that should be thoughtfully considered. Our findings highlight the need to expand access in a way that integrates technological advances with in-person care to build a sustainable and effective path forward without compromising the quality of care patients receive. We present these outcomes to emphasize the importance of increasing dermatology involvement in the care of hospitalized patients, which is a promising strategy to improve patient outcomes and reduce existing disparities in Virginia and nationwide.

To the Editor:

It is known that dermatologist evaluation of skin conditions in hospitalized patients confers enhanced diagnostic accuracy, timely and appropriate treatment, and an overall reduction in readmissions compared to assessments by nondermatology hospitalists.¹ Dermatology consultations have been shown to alter diagnoses in up to 50% of cases and lead to changes in management in nearly 75% of cases, even for prevalent dermatologic conditions such as drug rashes, cellulitis, and stasis dermatitis.^1,2 Previous studies have observed a multiday reduction in length of hospital stay, a 10-fold reduction in readmission rate, and lower 30-day mortality, all leading to a reduction in patient morbidity and costs to both the patient and the health care system.^3,4 Despite these benefits, there has been a decrease in the number of dermatologists providing inpatient services and a reduction in medical centers offering dermatology consultations over the past several years.⁵ To better appreciate current trends of declining dermatology inpatient and consultative services within our region, we evaluated the availability of dermatology care at hospitals across Virginia.

A simple telephone survey was conducted across community hospitals in Virginia wherein medical staff administrators were asked to provide details regarding their dermatology staffing. The following figures were collected: number of dermatologists on staff, number of dermatologists with consulting privileges, number of affiliated dermatologists, and number of advanced-practice dermatology providers. Follow-up calls were carried out to elaborate on how dermatologists (when available) were integrated into inpatient care workflow and made accessible to hospitalists and emergency medicine departments. Academic centers, military hospitals, and specialty hospitals were excluded from the survey.

To better appreciate the relationships between hospital and population characteristics and the availability of dermatology care, publicly available data were collected on hospital bed counts and regional population density for each facility.^6-9 Spearman rank correlation analyses were conducted in Microsoft Excel to evaluate the association between the number of dermatologists on staff, number of consulting dermatologists, staffed inpatient beds, and population size.

Sixty-four hospitals—more than 70% of the 90 eligible community hospitals—responded to the survey between May and August 2024 and were included in the study. On-staff dermatologists were present at 8 (12.5%) of the hospitals surveyed; of these, 4 (50.0%) hospitals had between 1 and 5 dermatologists, 3 (37.5%) had between 6 and 10 dermatologists, and 1 (12.5%) had between 11 and 15 dermatologists. An additional 4 (6.3%) hospitals provided consultative dermatology services from outside dermatology clinics. Urban hospitals accounted for 9 of 12 (75%) hospitals offering in-house dermatology services, either through on-staff physicians or consultations with clinic-based providers.

Based on Spearman rank correlation analysis, there was a positive correlation between the number of dermatologists on staff and the number of staffed hospital beds (r=0.61; P <.001). Similarly, there was a positive correlation between the number of dermatologists on staff and the population density of the affiliated region (r=0.58; P <.001). Finally, there was a positive correlation between the number of dermatologists on staff and the number of available consulting dermatologists (r=0.89; P <.001).

At facilities with only consultative dermatology services accessible, there often was no formal dermatology team or department present. Rather, the hospitals relied on a loosely affiliated network of dermatology providers or navigated inpatient dermatology needs almost exclusively via internal medicine hospitalists or emergency medicine physicians. When available, dermatology support from dermatology physicians often was provided through teledermatology platforms. Although teledermatology has a large role in increasing access to care within underserved areas, its reliance on images and second-hand case descriptions can limit the provider’s ability to perform a comprehensive examination and assessment. Moreover, it was noted that few hospital representatives could offer clarity on how dermatologists were integrated into the inpatient setting. It remained unclear whether dermatologists were practically accessible to the inpatient care teams in a structured manner.

The uneven distribution and limited availability of dermatology inpatient care in Virginia reflect national trends and underscore ongoing access issues for patients. Without intentional intervention, these trends are expected to continue, contributing to a glaring gap in hospital services as well as to patient morbidity and mortality. The correlation data obtained in our study further qualify these disparities. The positive correlations between dermatologist availability and hospital size and population density suggest that larger, more urban facilities are more likely to offer inpatient dermatology care, whether through staffing or consultation. This relationship is not unexpected, given the greater financial resources and specialist networks available to facilities with large patient volumes. This suggests that dermatology care is shaped by institutional capacity and geographic leverage rather than clinical need, reinforcing existing disparities.

Importantly, it should be noted that the data may overestimate the true availability of dermatologists to these patients. As revealed via follow-up survey calls, respondent facilities that provided dermatology via consultative services often did not have a defined structure for integrating this care into the inpatient workflow. In some instances, dermatologists were technically affiliated with the hospital but had varying levels of practical interaction with the hospital providers and their patients. Administrative staff's differing awareness regarding dermatology interaction with the hospital facility may reveal systemic underutilization and opportunities to improve coordination to achieve the greatest benefit from dermatology services. These observations are further informed by the scope of our study, which focused specifically on community hospitals. The exclusion of academic and military institutions—and the tendency of these to exist in more densely populated areas—may have limited how broadly our findings reflect nationwide dermatology access by omitting more established dermatology departments and specialty care. As a result, regional variations in predominant facility type should be considered when interpreting the implications of these results beyond Virginia’s community hospital system.

In response to access limitations and differences in availability, facilities are turning to integrated teledermatology as a valuable tool to expand the reach of specialist care, particularly in rural or resource-limited settings. This modality acts as an important step toward improving equity in care by beginning to bridge geographic gaps; however, along with these logistical advantages, teledermatology also confers diagnostic limitations and clinical trade-offs that should be thoughtfully considered. Our findings highlight the need to expand access in a way that integrates technological advances with in-person care to build a sustainable and effective path forward without compromising the quality of care patients receive. We present these outcomes to emphasize the importance of increasing dermatology involvement in the care of hospitalized patients, which is a promising strategy to improve patient outcomes and reduce existing disparities in Virginia and nationwide.

To the Editor:

It is known that dermatologist evaluation of skin conditions in hospitalized patients confers enhanced diagnostic accuracy, timely and appropriate treatment, and an overall reduction in readmissions compared to assessments by nondermatology hospitalists.¹ Dermatology consultations have been shown to alter diagnoses in up to 50% of cases and lead to changes in management in nearly 75% of cases, even for prevalent dermatologic conditions such as drug rashes, cellulitis, and stasis dermatitis.^1,2 Previous studies have observed a multiday reduction in length of hospital stay, a 10-fold reduction in readmission rate, and lower 30-day mortality, all leading to a reduction in patient morbidity and costs to both the patient and the health care system.^3,4 Despite these benefits, there has been a decrease in the number of dermatologists providing inpatient services and a reduction in medical centers offering dermatology consultations over the past several years.⁵ To better appreciate current trends of declining dermatology inpatient and consultative services within our region, we evaluated the availability of dermatology care at hospitals across Virginia.

A simple telephone survey was conducted across community hospitals in Virginia wherein medical staff administrators were asked to provide details regarding their dermatology staffing. The following figures were collected: number of dermatologists on staff, number of dermatologists with consulting privileges, number of affiliated dermatologists, and number of advanced-practice dermatology providers. Follow-up calls were carried out to elaborate on how dermatologists (when available) were integrated into inpatient care workflow and made accessible to hospitalists and emergency medicine departments. Academic centers, military hospitals, and specialty hospitals were excluded from the survey.

To better appreciate the relationships between hospital and population characteristics and the availability of dermatology care, publicly available data were collected on hospital bed counts and regional population density for each facility.^6-9 Spearman rank correlation analyses were conducted in Microsoft Excel to evaluate the association between the number of dermatologists on staff, number of consulting dermatologists, staffed inpatient beds, and population size.

Sixty-four hospitals—more than 70% of the 90 eligible community hospitals—responded to the survey between May and August 2024 and were included in the study. On-staff dermatologists were present at 8 (12.5%) of the hospitals surveyed; of these, 4 (50.0%) hospitals had between 1 and 5 dermatologists, 3 (37.5%) had between 6 and 10 dermatologists, and 1 (12.5%) had between 11 and 15 dermatologists. An additional 4 (6.3%) hospitals provided consultative dermatology services from outside dermatology clinics. Urban hospitals accounted for 9 of 12 (75%) hospitals offering in-house dermatology services, either through on-staff physicians or consultations with clinic-based providers.

Based on Spearman rank correlation analysis, there was a positive correlation between the number of dermatologists on staff and the number of staffed hospital beds (r=0.61; P <.001). Similarly, there was a positive correlation between the number of dermatologists on staff and the population density of the affiliated region (r=0.58; P <.001). Finally, there was a positive correlation between the number of dermatologists on staff and the number of available consulting dermatologists (r=0.89; P <.001).

At facilities with only consultative dermatology services accessible, there often was no formal dermatology team or department present. Rather, the hospitals relied on a loosely affiliated network of dermatology providers or navigated inpatient dermatology needs almost exclusively via internal medicine hospitalists or emergency medicine physicians. When available, dermatology support from dermatology physicians often was provided through teledermatology platforms. Although teledermatology has a large role in increasing access to care within underserved areas, its reliance on images and second-hand case descriptions can limit the provider’s ability to perform a comprehensive examination and assessment. Moreover, it was noted that few hospital representatives could offer clarity on how dermatologists were integrated into the inpatient setting. It remained unclear whether dermatologists were practically accessible to the inpatient care teams in a structured manner.

The uneven distribution and limited availability of dermatology inpatient care in Virginia reflect national trends and underscore ongoing access issues for patients. Without intentional intervention, these trends are expected to continue, contributing to a glaring gap in hospital services as well as to patient morbidity and mortality. The correlation data obtained in our study further qualify these disparities. The positive correlations between dermatologist availability and hospital size and population density suggest that larger, more urban facilities are more likely to offer inpatient dermatology care, whether through staffing or consultation. This relationship is not unexpected, given the greater financial resources and specialist networks available to facilities with large patient volumes. This suggests that dermatology care is shaped by institutional capacity and geographic leverage rather than clinical need, reinforcing existing disparities.

Importantly, it should be noted that the data may overestimate the true availability of dermatologists to these patients. As revealed via follow-up survey calls, respondent facilities that provided dermatology via consultative services often did not have a defined structure for integrating this care into the inpatient workflow. In some instances, dermatologists were technically affiliated with the hospital but had varying levels of practical interaction with the hospital providers and their patients. Administrative staff's differing awareness regarding dermatology interaction with the hospital facility may reveal systemic underutilization and opportunities to improve coordination to achieve the greatest benefit from dermatology services. These observations are further informed by the scope of our study, which focused specifically on community hospitals. The exclusion of academic and military institutions—and the tendency of these to exist in more densely populated areas—may have limited how broadly our findings reflect nationwide dermatology access by omitting more established dermatology departments and specialty care. As a result, regional variations in predominant facility type should be considered when interpreting the implications of these results beyond Virginia’s community hospital system.

In response to access limitations and differences in availability, facilities are turning to integrated teledermatology as a valuable tool to expand the reach of specialist care, particularly in rural or resource-limited settings. This modality acts as an important step toward improving equity in care by beginning to bridge geographic gaps; however, along with these logistical advantages, teledermatology also confers diagnostic limitations and clinical trade-offs that should be thoughtfully considered. Our findings highlight the need to expand access in a way that integrates technological advances with in-person care to build a sustainable and effective path forward without compromising the quality of care patients receive. We present these outcomes to emphasize the importance of increasing dermatology involvement in the care of hospitalized patients, which is a promising strategy to improve patient outcomes and reduce existing disparities in Virginia and nationwide.

Squamous cell carcinoma (SCC) is the malignant proliferation of keratinocytes in the epidermis of the skin. Most SCCs are caused by UV light exposure, with sex and increased age acting as the primary known risk factors: SCCs are nearly twice as prevalent in men vs women, and the average age of presentation is the middle of the seventh decade of life.¹ In the United States, there are an estimated 1.8 million new SCC cases annually.² Although not usually life threatening, if left untreated, SCC can metastasize, thereby reducing the 10-year survival rate from above 90% with treatment to 16%.^3-6

Most invasive SCC lesions are treated surgically, but intralesional methotrexate (IL-MTX) has emerged as an alternative treatment for cutaneous SCC. It offers the potential for lower-cost, efficacious outpatient treatment.^7-12 Methotrexate competitively inhibits the enzyme dihydrofolate reductase, which converts dihydrofolate into tetrahydrofolate.¹³ In doing so, MTX indirectly prevents the synthesis of thymine, a nucleotide required for DNA synthesis. Thus, MTX can halt DNA synthesis and consequently, cell division. Intralesional MTX has been shown to successfully treat keratoacanthomas, lymphomas, and various inflammatory dermatologic conditions.^8-12 

Surgical options include standard excision, Mohs micrographic surgery, or electrodesiccation and curettage. Surgical treatment has high (92% to 99%) cure rates and typically requires only 1 or 2 appointments.^14,15 Although costs can vary, one 2012 study using Medicare fee schedules found that total costs (including primary procedure, biopsy, follow-up appointments through 2 months, and other associated costs) for cutaneous SCC were $475 for electrodesiccation and curettage, $1302.92 for excision, and $2093.14 for Mohs micrographic surgery.¹⁶ For some patients, surgery is not an ideal option due to the tumor location, poor wound healing, anticoagulation, and cost. In these patients, photodynamic therapy, topical therapy with 5-fluorouracil or imiquimod, radiation, and cryotherapy are options listed in the American Academy of Dermatology guidelines.¹⁵ Compared with surgery, radiation is more demanding on the patient, often requiring multiple visits a week and including common undesirable adverse effects such as radiation dermatitis and prolonged wounds on the lower legs.¹⁷ Radiation also can be costly, with one study reporting costs between $2559 and $3431 for SCC of the forearm.¹⁸ Furthermore, in young patients, radiotherapy can increase the risk for developing nonmelanoma skin cancer later in life.¹⁶ 

Intralesional MTX is a localized treatment option that avoids the high costs of surgery, the side effects of radiotherapy, prolonged healing, and the systemic effects of chemotherapy. Treatment with IL-MTX can vary depending on the number of treatments necessary but usually only costs a few hundred dollars, rarely costing more than $1000.⁷ Although IL-MTX is less expensive, it typically requires several follow-up visits, whereas surgical removal may only require 1 visit.

Prior research has noted the efficacy of IL-MTX as a neoadjuvant therapy, with one study finding that IL-MTX can reduce the size of SCC lesions by an average of 0.52 cm² prior to surgery.¹⁹ Several case studies also have documented the effectiveness of IL-MTX as a treatment for SCC.^20-22 However, larger studies involving multiple patients to evaluate the efficacy of IL-MTX as a sole treatment for SCC are lacking. Gualdi et al²³ looked at the outcomes (complete resolution, partial response, or no response) for SCC treated with IL-MTX and found that 62% (13/21) of patients experienced improvement, with 48% (10/21) experiencing at least 50% improvement. Although these results are promising, further research is needed.

Our study sought to examine IL-MTX efficacy as well as evaluate the dosage and number of appointments/­sessions needed to achieve resolution of the lesions.

Methods

We conducted a retrospective chart review of patients who received only IL-MTX for clinically evident or biopsy-proven SCC at US Dermatology Partners clinics in Phoenix, Arizona, from January 1, 2022, to June 30, 2023. Patients aged 18 to 89 years were included, and they had not received other treatment for their SCC lesions such as radiation or systemic chemotherapy. Each patient received at least 1 dose of IL-MTX, beginning with a concentration of 12.5 mg/mL and with all subsequent doses at a concentration of 25 mg/mL (low dose vs high dose). Lesion resolution was categorized as no gross clinical tumor on follow-up. Patients received additional doses of IL-MTX based on the clinical appearance of their lesion(s).

Patient-level descriptive statistics are reported as mean (SD) or median (interquartile range [IQR]) for continuous variables as well as frequency and percentage for categorical variables. To account for the correlation of multiple lesions within individual patients, marginal Cox proportional hazard models were used. Time as well as cumulative dose to lesion resolution were evaluated and presented via the cumulative hazard function, while differences in resolution were estimated using separate Cox models for age, sex, and initial dose.

Results

In total, 107 different lesions from 21 patients were included in the analysis. The median number of lesions was 4 per patient (range, 1-15; IQR, 2-7), with a mean (SD) age of 80 (6) years. Patients were primarily female (81% [17/21]). From the data provided, the majority of lesions (83% [89/107]) resolved with IL-MTX. Of the 18 unresolved lesions, 5 (5%) were referred for a different procedure, and the remaining 13 (12%) were censored (lost to follow-up). Figure 1 provides the cumulative incidence function for lesion resolution. Approximately 50% of patient lesions resolved by the second appointment. Similarly, Figure 2 provides the cumulative dose function for lesion resolution; the median cumulative total dose for resolution was 5 mg (IQR, 2.5–12.5). Finally, concerning the ratio for case resolution, no difference in hazard ratio (HR) was observed for age (female vs male, HR: 1.01; 95% CI: 0.96-1.06), biological sex (HR, 1.01; 95% CI, 0.63-1.63), or initial dose (high vs low, HR: 1.13; 95% CI: 0.77-1.65).

Comment

Results of this study demonstrate the efficacy of IL-MTX for the treatment of cutaneous SCC. More than 80% of the lesions resolved by IL-MTX alone. This treatment approach is more cost-effective with fewer adverse effects when compared to other options. In our study, treatment with IL-MTX also proved to be reasonable in terms of the number of appointments and total dose required, with more than 50% of lesions resolving within 2 appointments and a median cumulative total dose of 5 mg. Intralesional MTX appears to be similarly efficacious in men and women, and the concentration of the initial dose (12.5 mg/mL vs 25 mg/mL) does not change the treatment outcome.

Although these data are encouraging for the use of IL-MTX in the treatment of SCC, future work should consider the relationships between lesion characteristics (such as size and location) and case resolution with IL-MTX as well as recurrence rates with lesions treated by IL-MTX compared to other treatment options.

Conclusion

This study demonstrated the efficacy of IL-MTX as a treatment for SCC that is cost-effective, avoids bothersome side effects, and can be accomplished in relatively few appointments. However, more data are needed to characterize the lesion type best suited to this treatment.

Squamous cell carcinoma (SCC) is the malignant proliferation of keratinocytes in the epidermis of the skin. Most SCCs are caused by UV light exposure, with sex and increased age acting as the primary known risk factors: SCCs are nearly twice as prevalent in men vs women, and the average age of presentation is the middle of the seventh decade of life.¹ In the United States, there are an estimated 1.8 million new SCC cases annually.² Although not usually life threatening, if left untreated, SCC can metastasize, thereby reducing the 10-year survival rate from above 90% with treatment to 16%.^3-6

Most invasive SCC lesions are treated surgically, but intralesional methotrexate (IL-MTX) has emerged as an alternative treatment for cutaneous SCC. It offers the potential for lower-cost, efficacious outpatient treatment.^7-12 Methotrexate competitively inhibits the enzyme dihydrofolate reductase, which converts dihydrofolate into tetrahydrofolate.¹³ In doing so, MTX indirectly prevents the synthesis of thymine, a nucleotide required for DNA synthesis. Thus, MTX can halt DNA synthesis and consequently, cell division. Intralesional MTX has been shown to successfully treat keratoacanthomas, lymphomas, and various inflammatory dermatologic conditions.^8-12 

Surgical options include standard excision, Mohs micrographic surgery, or electrodesiccation and curettage. Surgical treatment has high (92% to 99%) cure rates and typically requires only 1 or 2 appointments.^14,15 Although costs can vary, one 2012 study using Medicare fee schedules found that total costs (including primary procedure, biopsy, follow-up appointments through 2 months, and other associated costs) for cutaneous SCC were $475 for electrodesiccation and curettage, $1302.92 for excision, and $2093.14 for Mohs micrographic surgery.¹⁶ For some patients, surgery is not an ideal option due to the tumor location, poor wound healing, anticoagulation, and cost. In these patients, photodynamic therapy, topical therapy with 5-fluorouracil or imiquimod, radiation, and cryotherapy are options listed in the American Academy of Dermatology guidelines.¹⁵ Compared with surgery, radiation is more demanding on the patient, often requiring multiple visits a week and including common undesirable adverse effects such as radiation dermatitis and prolonged wounds on the lower legs.¹⁷ Radiation also can be costly, with one study reporting costs between $2559 and $3431 for SCC of the forearm.¹⁸ Furthermore, in young patients, radiotherapy can increase the risk for developing nonmelanoma skin cancer later in life.¹⁶ 

Intralesional MTX is a localized treatment option that avoids the high costs of surgery, the side effects of radiotherapy, prolonged healing, and the systemic effects of chemotherapy. Treatment with IL-MTX can vary depending on the number of treatments necessary but usually only costs a few hundred dollars, rarely costing more than $1000.⁷ Although IL-MTX is less expensive, it typically requires several follow-up visits, whereas surgical removal may only require 1 visit.

Prior research has noted the efficacy of IL-MTX as a neoadjuvant therapy, with one study finding that IL-MTX can reduce the size of SCC lesions by an average of 0.52 cm² prior to surgery.¹⁹ Several case studies also have documented the effectiveness of IL-MTX as a treatment for SCC.^20-22 However, larger studies involving multiple patients to evaluate the efficacy of IL-MTX as a sole treatment for SCC are lacking. Gualdi et al²³ looked at the outcomes (complete resolution, partial response, or no response) for SCC treated with IL-MTX and found that 62% (13/21) of patients experienced improvement, with 48% (10/21) experiencing at least 50% improvement. Although these results are promising, further research is needed.

Our study sought to examine IL-MTX efficacy as well as evaluate the dosage and number of appointments/­sessions needed to achieve resolution of the lesions.

Methods

We conducted a retrospective chart review of patients who received only IL-MTX for clinically evident or biopsy-proven SCC at US Dermatology Partners clinics in Phoenix, Arizona, from January 1, 2022, to June 30, 2023. Patients aged 18 to 89 years were included, and they had not received other treatment for their SCC lesions such as radiation or systemic chemotherapy. Each patient received at least 1 dose of IL-MTX, beginning with a concentration of 12.5 mg/mL and with all subsequent doses at a concentration of 25 mg/mL (low dose vs high dose). Lesion resolution was categorized as no gross clinical tumor on follow-up. Patients received additional doses of IL-MTX based on the clinical appearance of their lesion(s).

Patient-level descriptive statistics are reported as mean (SD) or median (interquartile range [IQR]) for continuous variables as well as frequency and percentage for categorical variables. To account for the correlation of multiple lesions within individual patients, marginal Cox proportional hazard models were used. Time as well as cumulative dose to lesion resolution were evaluated and presented via the cumulative hazard function, while differences in resolution were estimated using separate Cox models for age, sex, and initial dose.

Results

In total, 107 different lesions from 21 patients were included in the analysis. The median number of lesions was 4 per patient (range, 1-15; IQR, 2-7), with a mean (SD) age of 80 (6) years. Patients were primarily female (81% [17/21]). From the data provided, the majority of lesions (83% [89/107]) resolved with IL-MTX. Of the 18 unresolved lesions, 5 (5%) were referred for a different procedure, and the remaining 13 (12%) were censored (lost to follow-up). Figure 1 provides the cumulative incidence function for lesion resolution. Approximately 50% of patient lesions resolved by the second appointment. Similarly, Figure 2 provides the cumulative dose function for lesion resolution; the median cumulative total dose for resolution was 5 mg (IQR, 2.5–12.5). Finally, concerning the ratio for case resolution, no difference in hazard ratio (HR) was observed for age (female vs male, HR: 1.01; 95% CI: 0.96-1.06), biological sex (HR, 1.01; 95% CI, 0.63-1.63), or initial dose (high vs low, HR: 1.13; 95% CI: 0.77-1.65).

Comment

Results of this study demonstrate the efficacy of IL-MTX for the treatment of cutaneous SCC. More than 80% of the lesions resolved by IL-MTX alone. This treatment approach is more cost-effective with fewer adverse effects when compared to other options. In our study, treatment with IL-MTX also proved to be reasonable in terms of the number of appointments and total dose required, with more than 50% of lesions resolving within 2 appointments and a median cumulative total dose of 5 mg. Intralesional MTX appears to be similarly efficacious in men and women, and the concentration of the initial dose (12.5 mg/mL vs 25 mg/mL) does not change the treatment outcome.

Although these data are encouraging for the use of IL-MTX in the treatment of SCC, future work should consider the relationships between lesion characteristics (such as size and location) and case resolution with IL-MTX as well as recurrence rates with lesions treated by IL-MTX compared to other treatment options.

Conclusion

This study demonstrated the efficacy of IL-MTX as a treatment for SCC that is cost-effective, avoids bothersome side effects, and can be accomplished in relatively few appointments. However, more data are needed to characterize the lesion type best suited to this treatment.

Squamous cell carcinoma (SCC) is the malignant proliferation of keratinocytes in the epidermis of the skin. Most SCCs are caused by UV light exposure, with sex and increased age acting as the primary known risk factors: SCCs are nearly twice as prevalent in men vs women, and the average age of presentation is the middle of the seventh decade of life.¹ In the United States, there are an estimated 1.8 million new SCC cases annually.² Although not usually life threatening, if left untreated, SCC can metastasize, thereby reducing the 10-year survival rate from above 90% with treatment to 16%.^3-6

Most invasive SCC lesions are treated surgically, but intralesional methotrexate (IL-MTX) has emerged as an alternative treatment for cutaneous SCC. It offers the potential for lower-cost, efficacious outpatient treatment.^7-12 Methotrexate competitively inhibits the enzyme dihydrofolate reductase, which converts dihydrofolate into tetrahydrofolate.¹³ In doing so, MTX indirectly prevents the synthesis of thymine, a nucleotide required for DNA synthesis. Thus, MTX can halt DNA synthesis and consequently, cell division. Intralesional MTX has been shown to successfully treat keratoacanthomas, lymphomas, and various inflammatory dermatologic conditions.^8-12 

Surgical options include standard excision, Mohs micrographic surgery, or electrodesiccation and curettage. Surgical treatment has high (92% to 99%) cure rates and typically requires only 1 or 2 appointments.^14,15 Although costs can vary, one 2012 study using Medicare fee schedules found that total costs (including primary procedure, biopsy, follow-up appointments through 2 months, and other associated costs) for cutaneous SCC were $475 for electrodesiccation and curettage, $1302.92 for excision, and $2093.14 for Mohs micrographic surgery.¹⁶ For some patients, surgery is not an ideal option due to the tumor location, poor wound healing, anticoagulation, and cost. In these patients, photodynamic therapy, topical therapy with 5-fluorouracil or imiquimod, radiation, and cryotherapy are options listed in the American Academy of Dermatology guidelines.¹⁵ Compared with surgery, radiation is more demanding on the patient, often requiring multiple visits a week and including common undesirable adverse effects such as radiation dermatitis and prolonged wounds on the lower legs.¹⁷ Radiation also can be costly, with one study reporting costs between $2559 and $3431 for SCC of the forearm.¹⁸ Furthermore, in young patients, radiotherapy can increase the risk for developing nonmelanoma skin cancer later in life.¹⁶ 

Intralesional MTX is a localized treatment option that avoids the high costs of surgery, the side effects of radiotherapy, prolonged healing, and the systemic effects of chemotherapy. Treatment with IL-MTX can vary depending on the number of treatments necessary but usually only costs a few hundred dollars, rarely costing more than $1000.⁷ Although IL-MTX is less expensive, it typically requires several follow-up visits, whereas surgical removal may only require 1 visit.

Prior research has noted the efficacy of IL-MTX as a neoadjuvant therapy, with one study finding that IL-MTX can reduce the size of SCC lesions by an average of 0.52 cm² prior to surgery.¹⁹ Several case studies also have documented the effectiveness of IL-MTX as a treatment for SCC.^20-22 However, larger studies involving multiple patients to evaluate the efficacy of IL-MTX as a sole treatment for SCC are lacking. Gualdi et al²³ looked at the outcomes (complete resolution, partial response, or no response) for SCC treated with IL-MTX and found that 62% (13/21) of patients experienced improvement, with 48% (10/21) experiencing at least 50% improvement. Although these results are promising, further research is needed.

Our study sought to examine IL-MTX efficacy as well as evaluate the dosage and number of appointments/­sessions needed to achieve resolution of the lesions.

Methods

We conducted a retrospective chart review of patients who received only IL-MTX for clinically evident or biopsy-proven SCC at US Dermatology Partners clinics in Phoenix, Arizona, from January 1, 2022, to June 30, 2023. Patients aged 18 to 89 years were included, and they had not received other treatment for their SCC lesions such as radiation or systemic chemotherapy. Each patient received at least 1 dose of IL-MTX, beginning with a concentration of 12.5 mg/mL and with all subsequent doses at a concentration of 25 mg/mL (low dose vs high dose). Lesion resolution was categorized as no gross clinical tumor on follow-up. Patients received additional doses of IL-MTX based on the clinical appearance of their lesion(s).

Patient-level descriptive statistics are reported as mean (SD) or median (interquartile range [IQR]) for continuous variables as well as frequency and percentage for categorical variables. To account for the correlation of multiple lesions within individual patients, marginal Cox proportional hazard models were used. Time as well as cumulative dose to lesion resolution were evaluated and presented via the cumulative hazard function, while differences in resolution were estimated using separate Cox models for age, sex, and initial dose.

Results

In total, 107 different lesions from 21 patients were included in the analysis. The median number of lesions was 4 per patient (range, 1-15; IQR, 2-7), with a mean (SD) age of 80 (6) years. Patients were primarily female (81% [17/21]). From the data provided, the majority of lesions (83% [89/107]) resolved with IL-MTX. Of the 18 unresolved lesions, 5 (5%) were referred for a different procedure, and the remaining 13 (12%) were censored (lost to follow-up). Figure 1 provides the cumulative incidence function for lesion resolution. Approximately 50% of patient lesions resolved by the second appointment. Similarly, Figure 2 provides the cumulative dose function for lesion resolution; the median cumulative total dose for resolution was 5 mg (IQR, 2.5–12.5). Finally, concerning the ratio for case resolution, no difference in hazard ratio (HR) was observed for age (female vs male, HR: 1.01; 95% CI: 0.96-1.06), biological sex (HR, 1.01; 95% CI, 0.63-1.63), or initial dose (high vs low, HR: 1.13; 95% CI: 0.77-1.65).

Comment

Results of this study demonstrate the efficacy of IL-MTX for the treatment of cutaneous SCC. More than 80% of the lesions resolved by IL-MTX alone. This treatment approach is more cost-effective with fewer adverse effects when compared to other options. In our study, treatment with IL-MTX also proved to be reasonable in terms of the number of appointments and total dose required, with more than 50% of lesions resolving within 2 appointments and a median cumulative total dose of 5 mg. Intralesional MTX appears to be similarly efficacious in men and women, and the concentration of the initial dose (12.5 mg/mL vs 25 mg/mL) does not change the treatment outcome.

Although these data are encouraging for the use of IL-MTX in the treatment of SCC, future work should consider the relationships between lesion characteristics (such as size and location) and case resolution with IL-MTX as well as recurrence rates with lesions treated by IL-MTX compared to other treatment options.

Conclusion

This study demonstrated the efficacy of IL-MTX as a treatment for SCC that is cost-effective, avoids bothersome side effects, and can be accomplished in relatively few appointments. However, more data are needed to characterize the lesion type best suited to this treatment.

Large epidemiologic studies have shown disparities in COVID-19 outcomes by race, ethnicity, and socioeconomic status (SES). Racial and ethnic minorities and individuals of lower SES have experienced disproportionately higher rates of intensive care unit (ICU) admission and death. In Washington, DC, Black individuals (47% of the population) accounted for 51% of COVID-19 cases and 75% of deaths. In comparison, White individuals (41% of the population) accounted for 21% of cases and 11% of deaths.¹ Place of residence, such as living in socially vulnerable communities, has also been shown to be associated with higher rates of COVID-19 mortality and lower vaccination rates.^2-4 Social and structural inequities, such as limited access to health care services and mistrust of the health care system, may explain some of the observed disparities.⁵ However, data are limited regarding COVID-19 outcomes for individuals with equal access to care.

The Veterans Health Administration (VHA) is the largest integrated US health care system and operates 123 acute care hospitals. Previous research has demonstrated that disparities in outcomes for other diseases are attenuated or erased among veterans receiving VHA care.^6,7 Based on literature from the pandemic, markers of health care inequity relating to SES (eg, place of residence, median income) are expected to impact the outcomes of patients acutely hospitalized with COVID-19.⁴ We hypothesized that the impact on clinical outcomes of infection would be mitigated for veterans receiving VHA care.

This retrospective cohort study included veterans who presented to Washington Veterans Affairs Medical Center (WVAMC) with the goal of determining whether place of residence as a marker of SES, health care access, and median income were predictive of COVID-19 disease severity.

Methods

The WVAMC serves about 125,000 veterans across the metropolitan area, including parts of Maryland and Virginia. It is a high-complexity hospital with 164 acute care beds, 30 psychosocial residential rehabilitation beds, and an adjacent 120-bed community living center providing long-term, hospice, and palliative care.⁸

The WVAMC developed a dashboard that tracked patients with COVID-19 through on-site testing by admission date, ward, and other key demographics (PowerBi, Corporate Data Warehouse). All patients admitted to WVAMC with a diagnosis of COVID-19 between March 1, 2020, and June 30, 2021, were included in this retrospective review. Using the Computerized Patient Record System (CPRS) and the dashboard, we collected demographic information, baseline clinical diagnoses, laboratory results, and clinical interventions for all patients with documented COVID-19 infection as established by laboratory testing methods available at the time of diagnosis. Veterans treated exclusively outside the WVAMC were excluded. Hospitalization was defined as any acute inpatient admission or transfer recorded within 5 days before and 30 days after the laboratory collection of a positive COVID-19 test. Home testing kits were not widely available during the study period. An ICU stay was defined as any inpatient admission or transfer recorded within 5 days before or 30 days after the laboratory collection of a positive COVID-19 test for which the ward location had the specialty of medical or surgical ICU. Death due to COVID-19 was defined as occurring within 42 days (6 weeks) of a positive COVID-19 test.⁹ This definition assumed that during the peak of the pandemic, COVID-19 was the attributable cause of death, despite the possible contribution of underlying health conditions.

Patients’ admission periods were based on US Centers for Disease Control and Prevention (CDC) national data and classified as early 2020 (January 2020–April 2020), mid-2020 (May 2020–August 2020), late 2020 (September 2020–December 2020), and early 2021 (January 2021–April 2021).¹⁰ We chose to use these time periods as surrogates for the frequent changes in circulating COVID-19 variants, surges in case numbers, therapies and interventions available during the pandemic. The dominant COVID-19 variant during the study period was Alpha (B.1.17). Beta (B.1.351) variants were circulating infrequently, and Delta and Omicron appeared after the study period.¹¹ Treatment strategies evolved rapidly with emerging evidence, including the use of dexamethasone, beginning in June 2020.¹² WVAMC followed the Advisory Committee on Immunization Practices guidance on vaccination rollout beginning in December 2020.¹³

Patients' income was estimated by the median household income of the zip code residence based on US Census Bureau 2021 estimates and was assessed as both a continuous and categorical variable.¹⁴ The Charlson Comorbidity Index (CCI) was included in models as a continuous variable.¹⁵ Variables contributing to the CCI include myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, hemiplegia or paraplegia, ulcer disease, hepatic disease, diabetes (with or without end-organ damage), chronic obstructive pulmonary disease (COPD), connective tissue disease, leukemia, lymphoma, moderate or severe renal disease, solid tumor (with or without metastases), and HIV/AIDS. The WVAMC Institutional Review Board approved this study (IRB #1573071).

Variables

This study assessed 3 primary outcomes as indicators of disease severity during hospitalization: need for high-flow oxygen (HFO), intubation, and presumed mortality at any time during hospitalization. The following variables were collected as potential social determinants or clinical risk-adjustment predictors of disease severity outcomes: age; sex; race and ethnicity; median income for patient’s zip code residence, state, and county; wards within Washington, DC; comorbidities, CCI; tobacco use; and body mass index.¹⁵ Although medications at baseline, treatments during hospitalization for COVID-19, and laboratory parameters during hospitalization are shown in eAppendices 1 and 2, they are beyond the scope of this analysis.

Statistical Analysis

Three types of logistic regression models were calculated for predicting the disease severity outcomes: (1) simple unadjusted models; (2) models predicting from single variables plus age (age-adjusted); and (3) multivariable models using all nonredundant potential predictors with adequate sample sizes (multivariable). Variables were considered to have inadequate sample sizes if there was nontrivial missing data or small numbers within categories, (eg, AIDS, connective tissue disease). Potential predictors for the multivariable model included age, sex, race, median income by zip code residence, CCI, CDC admission period, obesity, hypertension, chronic kidney disease, obstructive sleep apnea (OSA), diabetes, COPD or asthma, liver disease, antibiotics, and acute kidney injury.

For the multivariable models, the following modifications were made to avoid unreliable parameter estimation and computation problems (quasi-separation): age and CCI were included as continuous rather than categorical variables. Race was recoded as a 2-category variable (Black vs other [White, Hispanic, American Indian, Alaska Native, Asian, Native Hawaiian, and Pacific Islander]), and ethnicity was excluded because of the small number of patients in this group (n = 16). Admission period was included. Predicted probability plots were generated for each outcome with continuous independent predictors (income and CCI), both unadjusted and adjusted for age as a continuous covariate. All analyses were performed using SAS version 9.4.

Heat Maps

Heat maps were generated to visualize the geospatial distribution of COVID-19 cases and median incomes across zip codes in the greater Washington, DC area. Patient case data and median income, aggregated by zip code, were imported using ArcGIS Online. A zip code boundary layer from Esri (United States Zip Code Boundaries) was used to spatially align the case data. Data were joined by matching zip codes or median incomes in the patient dataset to those in the boundary layer. The resulting polygon layer was styled using the Counts and Amounts (Color) symbology in ArcGIS Online, with case counts or median income determining the intensity of the color gradient.

Results

Between March 1, 2020, and June 30, 2021, 348 patients were hospitalized with COVID-19 (Table 1). The mean (SD) age was 68.4 (13.9) years, 313 patients (90.2%) were male, 281 patients (83.4%) were Black, 47 patients (13.6%) were White, and 16 patients (4.8%) were Hispanic. One hundred forty patients (40.2%) resided in Washington, DC, 151 (43.4%) in Maryland, and 19 (5.5%) in Virginia. HFO was received by 86 patients (24.7%), 33 (9.5%) required intubation and mechanical ventilation, and 57 (16.4%) died. All intubations and deaths occurred among patients aged > 50 years, with death occurring in 17.8% of patients aged > 50 years.

Demographic characteristics and baseline comorbidities associated with COVID-19 disease severity can be found in eAppendix 2. In unadjusted analyses, age was significantly associated with the risk of HFO, with a mean (SD) age of 72.5 (11.7) years among those requiring HFO and 67.1 (14.4) years among patients without HFO (odds ratio [OR], 1.03; 95% CI, 1.01-1.05; P = .002). Although age was not associated with the risk of intubation, it was significantly associated with mortality. Patients who died had a mean (SD) age of 76.8 (11.8) years compared with 66.8 (13.7) years among survivors (OR, 1.06; 95% CI, 1.04-1.09; P < .001).

Compared with patients with no comorbidities, CCI categories of mild, moderate, and severe were associated with increased risk of requiring HFO (eAppendix 3). The adjusted OR (aOR) was highest among patients with severe CCI (aOR, 7.00; 95% CI, 2.42-20.32; P = .0007). In age-adjusted analyses, CCI was not associated with intubation or mortality.

Geospatial Analyses

State of residence, county of residence, and geographic area (including Washington, DC wards, and geographic divisions within counties of residence in Maryland and Virginia) were not associated with the clinical outcomes studied (eAppendix 4). However, zip code-based median income, analyzed as a continuous variable, was associated with a reduced likelihood of receiving HFO (aOR, 0.91; 95% CI, 0.84-0.99; P = .03). Income was not significantly associated with intubation or mortality.

The majority of patients hospitalized for COVID-19 at WVAMC resided in zip codes in eastern Washington, DC, inclusive of wards 7 and 8, and Prince George’s County, Maryland (Figure 1). These areas also corresponded to the lowest median household income by zip code (Figure 2).

Multivariable Analysis

Significant predictors of HFO requirement included comorbid diabetes (OR, 2.42; 95% CI, 1.27-4.61; P = .006) and liver disease or cirrhosis (OR, 2.19; 95% CI, 1.09-4.39; P = .02) (Table 2). CDC admission period was also associated with HFO need. Patients admitted after early 2020 had lower odds of receiving HFO. Race and median income based on zip code residence were not associated with HFO requirement.

Comorbid liver disease or cirrhosis was a significant predictor of intubation (OR, 2.81; 95% CI, 1.07-7.40; P = .03). CDC admission period was associated with intubation with lower odds of intubation for patients admitted after early 2020. Race and median income by zip code were not associated with intubation.

Significant predictors of mortality included age (OR, 2.20; 95% CI, 1.55-3.14; P = .0001), comorbid liver disease or cirrhosis (OR, 2.97; 95% CI, 1.31-6.74; P = .008), and OSA (OR, 3.45; 95% CI, 1.49-7.97; P = .003). CDC admission period was associated with mortality, with lower odds of intubation for patients admitted in mid- and late 2020. Race and median income by zip code residence were not associated with intubation.

Discussion

In this study of COVID-19 disease severity at a large integrated health care system that provides equal access to care, race, ethnicity, and geographic location were not associated with the need for HFO, intubation, or presumed mortality. Median income by zip code residence was associated with reduced HFO use in univariable analyses but not in multivariable models.

These findings support existing literature suggesting that race and ethnicity alone do not explain disparities in COVID-19 outcomes. Multiple studies have demonstrated that disparities in health outcomes have been reduced for patients receiving VHA care.^6,16-19 However, even within a health care system with assumed equal access, the finding of an association between income and need for HFO in the univariable analysis may reflect a greater likelihood of delays in care due to structural barriers. Multiple studies suggest low SES may be an independent risk factor for severe COVID-19 disease. Individuals with low SES have higher rates of chronic diseases of obesity, diabetes, heart disease, and lung disease; thus, they are also at greater risk of serious illness with COVID-19.^20-24 Socioeconomic disadvantage may also have limited individuals’ ability to engage in protective behaviors to reduce COVID-19 infection risk, including food stockpiling, social distancing, avoidance of public transportation, and refraining from working in “essential jobs.”²¹

Beyond SES, place of residence also influences health outcomes. Prior literature supports using zip codes to assess area-based SES status and monitor health disparities.²⁵ The Social Vulnerability Index incorporates SES factors for communities and measures social determinates of health at a zip code level exclusive of race and ethnicity.²⁶ Socially vulnerable communities are known to have higher rates of chronic diseases, COVID-19 mortality, and lower vaccination rates.³ Within a defined geographic area, an individual’s outcome for COVID-19 can be influenced by individual resources such as access to care and median income. Disposable income may mitigate COVID-19 risk by facilitating timely care, reducing occupational exposure, improving housing stability, and supporting health-promoting behaviors.²¹

Limitations

Due to the evolving nature of the COVID-19 pandemic, variants, treatments, and interventions varied throughout the study period and are not included in this analysis. In late December 2020, COVID-19 vaccination was approved with a tiered allocation for at-risk patients and direct health care professionals. Three of the 4 study periods analyzed in this study were prior to vaccine rollout and therefore vaccination history was not assessed. However, we tried to capture the evolving changes in COVID-19 variants, treatments and interventions, and skill in treating the disease through use of CDC-defined time frames. Another limitation is that some studies have shown that use of median income by zip code residence can underestimate mortality.²⁷ Also, shared resources and access to other sources of disposable income can impact the immediate attainment of social needs. For example, during the COVID-19 pandemic, health care systems in Washington, DC assisted vulnerable individuals by providing food, housing, and other resources.^28,29 Finally, the modest sample size limits generalizability and power to detect differences for certain variables, including Hispanic ethnicity.

Conclusions

There have been widely described disparities in disease severity and death during the COVID-19 pandemic. In this urban veteran cohort of hospitalized patients, there was no difference in the need for intubation or mortality associated with race. The findings suggest that a lower median income by zip code residence may be associated with greater disease severity at presentation, but do not predict severe outcomes and mortality overall. VHA care, which provides equal access to care, may mitigate the disparities seen in the private sector.

Large epidemiologic studies have shown disparities in COVID-19 outcomes by race, ethnicity, and socioeconomic status (SES). Racial and ethnic minorities and individuals of lower SES have experienced disproportionately higher rates of intensive care unit (ICU) admission and death. In Washington, DC, Black individuals (47% of the population) accounted for 51% of COVID-19 cases and 75% of deaths. In comparison, White individuals (41% of the population) accounted for 21% of cases and 11% of deaths.¹ Place of residence, such as living in socially vulnerable communities, has also been shown to be associated with higher rates of COVID-19 mortality and lower vaccination rates.^2-4 Social and structural inequities, such as limited access to health care services and mistrust of the health care system, may explain some of the observed disparities.⁵ However, data are limited regarding COVID-19 outcomes for individuals with equal access to care.

The Veterans Health Administration (VHA) is the largest integrated US health care system and operates 123 acute care hospitals. Previous research has demonstrated that disparities in outcomes for other diseases are attenuated or erased among veterans receiving VHA care.^6,7 Based on literature from the pandemic, markers of health care inequity relating to SES (eg, place of residence, median income) are expected to impact the outcomes of patients acutely hospitalized with COVID-19.⁴ We hypothesized that the impact on clinical outcomes of infection would be mitigated for veterans receiving VHA care.

This retrospective cohort study included veterans who presented to Washington Veterans Affairs Medical Center (WVAMC) with the goal of determining whether place of residence as a marker of SES, health care access, and median income were predictive of COVID-19 disease severity.

Methods

The WVAMC serves about 125,000 veterans across the metropolitan area, including parts of Maryland and Virginia. It is a high-complexity hospital with 164 acute care beds, 30 psychosocial residential rehabilitation beds, and an adjacent 120-bed community living center providing long-term, hospice, and palliative care.⁸

The WVAMC developed a dashboard that tracked patients with COVID-19 through on-site testing by admission date, ward, and other key demographics (PowerBi, Corporate Data Warehouse). All patients admitted to WVAMC with a diagnosis of COVID-19 between March 1, 2020, and June 30, 2021, were included in this retrospective review. Using the Computerized Patient Record System (CPRS) and the dashboard, we collected demographic information, baseline clinical diagnoses, laboratory results, and clinical interventions for all patients with documented COVID-19 infection as established by laboratory testing methods available at the time of diagnosis. Veterans treated exclusively outside the WVAMC were excluded. Hospitalization was defined as any acute inpatient admission or transfer recorded within 5 days before and 30 days after the laboratory collection of a positive COVID-19 test. Home testing kits were not widely available during the study period. An ICU stay was defined as any inpatient admission or transfer recorded within 5 days before or 30 days after the laboratory collection of a positive COVID-19 test for which the ward location had the specialty of medical or surgical ICU. Death due to COVID-19 was defined as occurring within 42 days (6 weeks) of a positive COVID-19 test.⁹ This definition assumed that during the peak of the pandemic, COVID-19 was the attributable cause of death, despite the possible contribution of underlying health conditions.

Patients’ admission periods were based on US Centers for Disease Control and Prevention (CDC) national data and classified as early 2020 (January 2020–April 2020), mid-2020 (May 2020–August 2020), late 2020 (September 2020–December 2020), and early 2021 (January 2021–April 2021).¹⁰ We chose to use these time periods as surrogates for the frequent changes in circulating COVID-19 variants, surges in case numbers, therapies and interventions available during the pandemic. The dominant COVID-19 variant during the study period was Alpha (B.1.17). Beta (B.1.351) variants were circulating infrequently, and Delta and Omicron appeared after the study period.¹¹ Treatment strategies evolved rapidly with emerging evidence, including the use of dexamethasone, beginning in June 2020.¹² WVAMC followed the Advisory Committee on Immunization Practices guidance on vaccination rollout beginning in December 2020.¹³

Patients' income was estimated by the median household income of the zip code residence based on US Census Bureau 2021 estimates and was assessed as both a continuous and categorical variable.¹⁴ The Charlson Comorbidity Index (CCI) was included in models as a continuous variable.¹⁵ Variables contributing to the CCI include myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, hemiplegia or paraplegia, ulcer disease, hepatic disease, diabetes (with or without end-organ damage), chronic obstructive pulmonary disease (COPD), connective tissue disease, leukemia, lymphoma, moderate or severe renal disease, solid tumor (with or without metastases), and HIV/AIDS. The WVAMC Institutional Review Board approved this study (IRB #1573071).

Variables

This study assessed 3 primary outcomes as indicators of disease severity during hospitalization: need for high-flow oxygen (HFO), intubation, and presumed mortality at any time during hospitalization. The following variables were collected as potential social determinants or clinical risk-adjustment predictors of disease severity outcomes: age; sex; race and ethnicity; median income for patient’s zip code residence, state, and county; wards within Washington, DC; comorbidities, CCI; tobacco use; and body mass index.¹⁵ Although medications at baseline, treatments during hospitalization for COVID-19, and laboratory parameters during hospitalization are shown in eAppendices 1 and 2, they are beyond the scope of this analysis.

Statistical Analysis

Three types of logistic regression models were calculated for predicting the disease severity outcomes: (1) simple unadjusted models; (2) models predicting from single variables plus age (age-adjusted); and (3) multivariable models using all nonredundant potential predictors with adequate sample sizes (multivariable). Variables were considered to have inadequate sample sizes if there was nontrivial missing data or small numbers within categories, (eg, AIDS, connective tissue disease). Potential predictors for the multivariable model included age, sex, race, median income by zip code residence, CCI, CDC admission period, obesity, hypertension, chronic kidney disease, obstructive sleep apnea (OSA), diabetes, COPD or asthma, liver disease, antibiotics, and acute kidney injury.

For the multivariable models, the following modifications were made to avoid unreliable parameter estimation and computation problems (quasi-separation): age and CCI were included as continuous rather than categorical variables. Race was recoded as a 2-category variable (Black vs other [White, Hispanic, American Indian, Alaska Native, Asian, Native Hawaiian, and Pacific Islander]), and ethnicity was excluded because of the small number of patients in this group (n = 16). Admission period was included. Predicted probability plots were generated for each outcome with continuous independent predictors (income and CCI), both unadjusted and adjusted for age as a continuous covariate. All analyses were performed using SAS version 9.4.

Heat Maps

Heat maps were generated to visualize the geospatial distribution of COVID-19 cases and median incomes across zip codes in the greater Washington, DC area. Patient case data and median income, aggregated by zip code, were imported using ArcGIS Online. A zip code boundary layer from Esri (United States Zip Code Boundaries) was used to spatially align the case data. Data were joined by matching zip codes or median incomes in the patient dataset to those in the boundary layer. The resulting polygon layer was styled using the Counts and Amounts (Color) symbology in ArcGIS Online, with case counts or median income determining the intensity of the color gradient.

Results

Between March 1, 2020, and June 30, 2021, 348 patients were hospitalized with COVID-19 (Table 1). The mean (SD) age was 68.4 (13.9) years, 313 patients (90.2%) were male, 281 patients (83.4%) were Black, 47 patients (13.6%) were White, and 16 patients (4.8%) were Hispanic. One hundred forty patients (40.2%) resided in Washington, DC, 151 (43.4%) in Maryland, and 19 (5.5%) in Virginia. HFO was received by 86 patients (24.7%), 33 (9.5%) required intubation and mechanical ventilation, and 57 (16.4%) died. All intubations and deaths occurred among patients aged > 50 years, with death occurring in 17.8% of patients aged > 50 years.

Demographic characteristics and baseline comorbidities associated with COVID-19 disease severity can be found in eAppendix 2. In unadjusted analyses, age was significantly associated with the risk of HFO, with a mean (SD) age of 72.5 (11.7) years among those requiring HFO and 67.1 (14.4) years among patients without HFO (odds ratio [OR], 1.03; 95% CI, 1.01-1.05; P = .002). Although age was not associated with the risk of intubation, it was significantly associated with mortality. Patients who died had a mean (SD) age of 76.8 (11.8) years compared with 66.8 (13.7) years among survivors (OR, 1.06; 95% CI, 1.04-1.09; P < .001).

Compared with patients with no comorbidities, CCI categories of mild, moderate, and severe were associated with increased risk of requiring HFO (eAppendix 3). The adjusted OR (aOR) was highest among patients with severe CCI (aOR, 7.00; 95% CI, 2.42-20.32; P = .0007). In age-adjusted analyses, CCI was not associated with intubation or mortality.

Geospatial Analyses

State of residence, county of residence, and geographic area (including Washington, DC wards, and geographic divisions within counties of residence in Maryland and Virginia) were not associated with the clinical outcomes studied (eAppendix 4). However, zip code-based median income, analyzed as a continuous variable, was associated with a reduced likelihood of receiving HFO (aOR, 0.91; 95% CI, 0.84-0.99; P = .03). Income was not significantly associated with intubation or mortality.

The majority of patients hospitalized for COVID-19 at WVAMC resided in zip codes in eastern Washington, DC, inclusive of wards 7 and 8, and Prince George’s County, Maryland (Figure 1). These areas also corresponded to the lowest median household income by zip code (Figure 2).

Multivariable Analysis

Significant predictors of HFO requirement included comorbid diabetes (OR, 2.42; 95% CI, 1.27-4.61; P = .006) and liver disease or cirrhosis (OR, 2.19; 95% CI, 1.09-4.39; P = .02) (Table 2). CDC admission period was also associated with HFO need. Patients admitted after early 2020 had lower odds of receiving HFO. Race and median income based on zip code residence were not associated with HFO requirement.

Comorbid liver disease or cirrhosis was a significant predictor of intubation (OR, 2.81; 95% CI, 1.07-7.40; P = .03). CDC admission period was associated with intubation with lower odds of intubation for patients admitted after early 2020. Race and median income by zip code were not associated with intubation.

Significant predictors of mortality included age (OR, 2.20; 95% CI, 1.55-3.14; P = .0001), comorbid liver disease or cirrhosis (OR, 2.97; 95% CI, 1.31-6.74; P = .008), and OSA (OR, 3.45; 95% CI, 1.49-7.97; P = .003). CDC admission period was associated with mortality, with lower odds of intubation for patients admitted in mid- and late 2020. Race and median income by zip code residence were not associated with intubation.

Discussion

In this study of COVID-19 disease severity at a large integrated health care system that provides equal access to care, race, ethnicity, and geographic location were not associated with the need for HFO, intubation, or presumed mortality. Median income by zip code residence was associated with reduced HFO use in univariable analyses but not in multivariable models.

These findings support existing literature suggesting that race and ethnicity alone do not explain disparities in COVID-19 outcomes. Multiple studies have demonstrated that disparities in health outcomes have been reduced for patients receiving VHA care.^6,16-19 However, even within a health care system with assumed equal access, the finding of an association between income and need for HFO in the univariable analysis may reflect a greater likelihood of delays in care due to structural barriers. Multiple studies suggest low SES may be an independent risk factor for severe COVID-19 disease. Individuals with low SES have higher rates of chronic diseases of obesity, diabetes, heart disease, and lung disease; thus, they are also at greater risk of serious illness with COVID-19.^20-24 Socioeconomic disadvantage may also have limited individuals’ ability to engage in protective behaviors to reduce COVID-19 infection risk, including food stockpiling, social distancing, avoidance of public transportation, and refraining from working in “essential jobs.”²¹

Beyond SES, place of residence also influences health outcomes. Prior literature supports using zip codes to assess area-based SES status and monitor health disparities.²⁵ The Social Vulnerability Index incorporates SES factors for communities and measures social determinates of health at a zip code level exclusive of race and ethnicity.²⁶ Socially vulnerable communities are known to have higher rates of chronic diseases, COVID-19 mortality, and lower vaccination rates.³ Within a defined geographic area, an individual’s outcome for COVID-19 can be influenced by individual resources such as access to care and median income. Disposable income may mitigate COVID-19 risk by facilitating timely care, reducing occupational exposure, improving housing stability, and supporting health-promoting behaviors.²¹

Limitations

Due to the evolving nature of the COVID-19 pandemic, variants, treatments, and interventions varied throughout the study period and are not included in this analysis. In late December 2020, COVID-19 vaccination was approved with a tiered allocation for at-risk patients and direct health care professionals. Three of the 4 study periods analyzed in this study were prior to vaccine rollout and therefore vaccination history was not assessed. However, we tried to capture the evolving changes in COVID-19 variants, treatments and interventions, and skill in treating the disease through use of CDC-defined time frames. Another limitation is that some studies have shown that use of median income by zip code residence can underestimate mortality.²⁷ Also, shared resources and access to other sources of disposable income can impact the immediate attainment of social needs. For example, during the COVID-19 pandemic, health care systems in Washington, DC assisted vulnerable individuals by providing food, housing, and other resources.^28,29 Finally, the modest sample size limits generalizability and power to detect differences for certain variables, including Hispanic ethnicity.

Conclusions

There have been widely described disparities in disease severity and death during the COVID-19 pandemic. In this urban veteran cohort of hospitalized patients, there was no difference in the need for intubation or mortality associated with race. The findings suggest that a lower median income by zip code residence may be associated with greater disease severity at presentation, but do not predict severe outcomes and mortality overall. VHA care, which provides equal access to care, may mitigate the disparities seen in the private sector.

Large epidemiologic studies have shown disparities in COVID-19 outcomes by race, ethnicity, and socioeconomic status (SES). Racial and ethnic minorities and individuals of lower SES have experienced disproportionately higher rates of intensive care unit (ICU) admission and death. In Washington, DC, Black individuals (47% of the population) accounted for 51% of COVID-19 cases and 75% of deaths. In comparison, White individuals (41% of the population) accounted for 21% of cases and 11% of deaths.¹ Place of residence, such as living in socially vulnerable communities, has also been shown to be associated with higher rates of COVID-19 mortality and lower vaccination rates.^2-4 Social and structural inequities, such as limited access to health care services and mistrust of the health care system, may explain some of the observed disparities.⁵ However, data are limited regarding COVID-19 outcomes for individuals with equal access to care.

The Veterans Health Administration (VHA) is the largest integrated US health care system and operates 123 acute care hospitals. Previous research has demonstrated that disparities in outcomes for other diseases are attenuated or erased among veterans receiving VHA care.^6,7 Based on literature from the pandemic, markers of health care inequity relating to SES (eg, place of residence, median income) are expected to impact the outcomes of patients acutely hospitalized with COVID-19.⁴ We hypothesized that the impact on clinical outcomes of infection would be mitigated for veterans receiving VHA care.

This retrospective cohort study included veterans who presented to Washington Veterans Affairs Medical Center (WVAMC) with the goal of determining whether place of residence as a marker of SES, health care access, and median income were predictive of COVID-19 disease severity.

Methods

The WVAMC serves about 125,000 veterans across the metropolitan area, including parts of Maryland and Virginia. It is a high-complexity hospital with 164 acute care beds, 30 psychosocial residential rehabilitation beds, and an adjacent 120-bed community living center providing long-term, hospice, and palliative care.⁸

The WVAMC developed a dashboard that tracked patients with COVID-19 through on-site testing by admission date, ward, and other key demographics (PowerBi, Corporate Data Warehouse). All patients admitted to WVAMC with a diagnosis of COVID-19 between March 1, 2020, and June 30, 2021, were included in this retrospective review. Using the Computerized Patient Record System (CPRS) and the dashboard, we collected demographic information, baseline clinical diagnoses, laboratory results, and clinical interventions for all patients with documented COVID-19 infection as established by laboratory testing methods available at the time of diagnosis. Veterans treated exclusively outside the WVAMC were excluded. Hospitalization was defined as any acute inpatient admission or transfer recorded within 5 days before and 30 days after the laboratory collection of a positive COVID-19 test. Home testing kits were not widely available during the study period. An ICU stay was defined as any inpatient admission or transfer recorded within 5 days before or 30 days after the laboratory collection of a positive COVID-19 test for which the ward location had the specialty of medical or surgical ICU. Death due to COVID-19 was defined as occurring within 42 days (6 weeks) of a positive COVID-19 test.⁹ This definition assumed that during the peak of the pandemic, COVID-19 was the attributable cause of death, despite the possible contribution of underlying health conditions.

Patients’ admission periods were based on US Centers for Disease Control and Prevention (CDC) national data and classified as early 2020 (January 2020–April 2020), mid-2020 (May 2020–August 2020), late 2020 (September 2020–December 2020), and early 2021 (January 2021–April 2021).¹⁰ We chose to use these time periods as surrogates for the frequent changes in circulating COVID-19 variants, surges in case numbers, therapies and interventions available during the pandemic. The dominant COVID-19 variant during the study period was Alpha (B.1.17). Beta (B.1.351) variants were circulating infrequently, and Delta and Omicron appeared after the study period.¹¹ Treatment strategies evolved rapidly with emerging evidence, including the use of dexamethasone, beginning in June 2020.¹² WVAMC followed the Advisory Committee on Immunization Practices guidance on vaccination rollout beginning in December 2020.¹³

Patients' income was estimated by the median household income of the zip code residence based on US Census Bureau 2021 estimates and was assessed as both a continuous and categorical variable.¹⁴ The Charlson Comorbidity Index (CCI) was included in models as a continuous variable.¹⁵ Variables contributing to the CCI include myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, hemiplegia or paraplegia, ulcer disease, hepatic disease, diabetes (with or without end-organ damage), chronic obstructive pulmonary disease (COPD), connective tissue disease, leukemia, lymphoma, moderate or severe renal disease, solid tumor (with or without metastases), and HIV/AIDS. The WVAMC Institutional Review Board approved this study (IRB #1573071).

Variables

This study assessed 3 primary outcomes as indicators of disease severity during hospitalization: need for high-flow oxygen (HFO), intubation, and presumed mortality at any time during hospitalization. The following variables were collected as potential social determinants or clinical risk-adjustment predictors of disease severity outcomes: age; sex; race and ethnicity; median income for patient’s zip code residence, state, and county; wards within Washington, DC; comorbidities, CCI; tobacco use; and body mass index.¹⁵ Although medications at baseline, treatments during hospitalization for COVID-19, and laboratory parameters during hospitalization are shown in eAppendices 1 and 2, they are beyond the scope of this analysis.

Statistical Analysis

Three types of logistic regression models were calculated for predicting the disease severity outcomes: (1) simple unadjusted models; (2) models predicting from single variables plus age (age-adjusted); and (3) multivariable models using all nonredundant potential predictors with adequate sample sizes (multivariable). Variables were considered to have inadequate sample sizes if there was nontrivial missing data or small numbers within categories, (eg, AIDS, connective tissue disease). Potential predictors for the multivariable model included age, sex, race, median income by zip code residence, CCI, CDC admission period, obesity, hypertension, chronic kidney disease, obstructive sleep apnea (OSA), diabetes, COPD or asthma, liver disease, antibiotics, and acute kidney injury.

For the multivariable models, the following modifications were made to avoid unreliable parameter estimation and computation problems (quasi-separation): age and CCI were included as continuous rather than categorical variables. Race was recoded as a 2-category variable (Black vs other [White, Hispanic, American Indian, Alaska Native, Asian, Native Hawaiian, and Pacific Islander]), and ethnicity was excluded because of the small number of patients in this group (n = 16). Admission period was included. Predicted probability plots were generated for each outcome with continuous independent predictors (income and CCI), both unadjusted and adjusted for age as a continuous covariate. All analyses were performed using SAS version 9.4.

Heat Maps

Heat maps were generated to visualize the geospatial distribution of COVID-19 cases and median incomes across zip codes in the greater Washington, DC area. Patient case data and median income, aggregated by zip code, were imported using ArcGIS Online. A zip code boundary layer from Esri (United States Zip Code Boundaries) was used to spatially align the case data. Data were joined by matching zip codes or median incomes in the patient dataset to those in the boundary layer. The resulting polygon layer was styled using the Counts and Amounts (Color) symbology in ArcGIS Online, with case counts or median income determining the intensity of the color gradient.

Results

Between March 1, 2020, and June 30, 2021, 348 patients were hospitalized with COVID-19 (Table 1). The mean (SD) age was 68.4 (13.9) years, 313 patients (90.2%) were male, 281 patients (83.4%) were Black, 47 patients (13.6%) were White, and 16 patients (4.8%) were Hispanic. One hundred forty patients (40.2%) resided in Washington, DC, 151 (43.4%) in Maryland, and 19 (5.5%) in Virginia. HFO was received by 86 patients (24.7%), 33 (9.5%) required intubation and mechanical ventilation, and 57 (16.4%) died. All intubations and deaths occurred among patients aged > 50 years, with death occurring in 17.8% of patients aged > 50 years.

Demographic characteristics and baseline comorbidities associated with COVID-19 disease severity can be found in eAppendix 2. In unadjusted analyses, age was significantly associated with the risk of HFO, with a mean (SD) age of 72.5 (11.7) years among those requiring HFO and 67.1 (14.4) years among patients without HFO (odds ratio [OR], 1.03; 95% CI, 1.01-1.05; P = .002). Although age was not associated with the risk of intubation, it was significantly associated with mortality. Patients who died had a mean (SD) age of 76.8 (11.8) years compared with 66.8 (13.7) years among survivors (OR, 1.06; 95% CI, 1.04-1.09; P < .001).

Compared with patients with no comorbidities, CCI categories of mild, moderate, and severe were associated with increased risk of requiring HFO (eAppendix 3). The adjusted OR (aOR) was highest among patients with severe CCI (aOR, 7.00; 95% CI, 2.42-20.32; P = .0007). In age-adjusted analyses, CCI was not associated with intubation or mortality.

Geospatial Analyses

State of residence, county of residence, and geographic area (including Washington, DC wards, and geographic divisions within counties of residence in Maryland and Virginia) were not associated with the clinical outcomes studied (eAppendix 4). However, zip code-based median income, analyzed as a continuous variable, was associated with a reduced likelihood of receiving HFO (aOR, 0.91; 95% CI, 0.84-0.99; P = .03). Income was not significantly associated with intubation or mortality.

The majority of patients hospitalized for COVID-19 at WVAMC resided in zip codes in eastern Washington, DC, inclusive of wards 7 and 8, and Prince George’s County, Maryland (Figure 1). These areas also corresponded to the lowest median household income by zip code (Figure 2).

Multivariable Analysis

Significant predictors of HFO requirement included comorbid diabetes (OR, 2.42; 95% CI, 1.27-4.61; P = .006) and liver disease or cirrhosis (OR, 2.19; 95% CI, 1.09-4.39; P = .02) (Table 2). CDC admission period was also associated with HFO need. Patients admitted after early 2020 had lower odds of receiving HFO. Race and median income based on zip code residence were not associated with HFO requirement.

Comorbid liver disease or cirrhosis was a significant predictor of intubation (OR, 2.81; 95% CI, 1.07-7.40; P = .03). CDC admission period was associated with intubation with lower odds of intubation for patients admitted after early 2020. Race and median income by zip code were not associated with intubation.

Significant predictors of mortality included age (OR, 2.20; 95% CI, 1.55-3.14; P = .0001), comorbid liver disease or cirrhosis (OR, 2.97; 95% CI, 1.31-6.74; P = .008), and OSA (OR, 3.45; 95% CI, 1.49-7.97; P = .003). CDC admission period was associated with mortality, with lower odds of intubation for patients admitted in mid- and late 2020. Race and median income by zip code residence were not associated with intubation.

Discussion

In this study of COVID-19 disease severity at a large integrated health care system that provides equal access to care, race, ethnicity, and geographic location were not associated with the need for HFO, intubation, or presumed mortality. Median income by zip code residence was associated with reduced HFO use in univariable analyses but not in multivariable models.

These findings support existing literature suggesting that race and ethnicity alone do not explain disparities in COVID-19 outcomes. Multiple studies have demonstrated that disparities in health outcomes have been reduced for patients receiving VHA care.^6,16-19 However, even within a health care system with assumed equal access, the finding of an association between income and need for HFO in the univariable analysis may reflect a greater likelihood of delays in care due to structural barriers. Multiple studies suggest low SES may be an independent risk factor for severe COVID-19 disease. Individuals with low SES have higher rates of chronic diseases of obesity, diabetes, heart disease, and lung disease; thus, they are also at greater risk of serious illness with COVID-19.^20-24 Socioeconomic disadvantage may also have limited individuals’ ability to engage in protective behaviors to reduce COVID-19 infection risk, including food stockpiling, social distancing, avoidance of public transportation, and refraining from working in “essential jobs.”²¹

Beyond SES, place of residence also influences health outcomes. Prior literature supports using zip codes to assess area-based SES status and monitor health disparities.²⁵ The Social Vulnerability Index incorporates SES factors for communities and measures social determinates of health at a zip code level exclusive of race and ethnicity.²⁶ Socially vulnerable communities are known to have higher rates of chronic diseases, COVID-19 mortality, and lower vaccination rates.³ Within a defined geographic area, an individual’s outcome for COVID-19 can be influenced by individual resources such as access to care and median income. Disposable income may mitigate COVID-19 risk by facilitating timely care, reducing occupational exposure, improving housing stability, and supporting health-promoting behaviors.²¹

Limitations

Due to the evolving nature of the COVID-19 pandemic, variants, treatments, and interventions varied throughout the study period and are not included in this analysis. In late December 2020, COVID-19 vaccination was approved with a tiered allocation for at-risk patients and direct health care professionals. Three of the 4 study periods analyzed in this study were prior to vaccine rollout and therefore vaccination history was not assessed. However, we tried to capture the evolving changes in COVID-19 variants, treatments and interventions, and skill in treating the disease through use of CDC-defined time frames. Another limitation is that some studies have shown that use of median income by zip code residence can underestimate mortality.²⁷ Also, shared resources and access to other sources of disposable income can impact the immediate attainment of social needs. For example, during the COVID-19 pandemic, health care systems in Washington, DC assisted vulnerable individuals by providing food, housing, and other resources.^28,29 Finally, the modest sample size limits generalizability and power to detect differences for certain variables, including Hispanic ethnicity.

Conclusions

There have been widely described disparities in disease severity and death during the COVID-19 pandemic. In this urban veteran cohort of hospitalized patients, there was no difference in the need for intubation or mortality associated with race. The findings suggest that a lower median income by zip code residence may be associated with greater disease severity at presentation, but do not predict severe outcomes and mortality overall. VHA care, which provides equal access to care, may mitigate the disparities seen in the private sector.

Acne keloidalis nuchae (AKN) classically presents as chronic inflammation of the hair follicles on the occipital scalp/nape of the neck manifesting as papules and pustules that may progress to keloidlike scarring.¹ Photographs depicting the typical clinical presentation of AKN are shown in the Figure. In the literature, AKN has been described as primarily occurring in postpubertal males of African descent.² Despite its similar name, AKN is not related to acne vulgaris.³ The underlying cause of AKN is hypothesized to be multifactorial, including inflammation, infection, and trauma.² Acne keloidalis nuchae is most common in males aged 14 to 50 years, which may indicate that increased androgens contribute to its development.³ In some cases, patients have reported developing AKN lesions after receiving a haircut or shaving, suggesting a potential role of trauma to the hair follicles and secondary infection.² Histopathology typically shows a perifollicular inflammatory infiltrate that obscures the hair follicles with associated proximal fibrosis.⁴ On physical examination, dermoscopy can be used to visualize perifollicular pustules and fibrosis, which appears white, in the early stages of AKN. Patients may present with tufted hairs in more advanced stages.⁵ Patients with AKN often describe the lesions as pruritic and painful.²

In this study, we evaluated the most common treatment regimens used over a 6-year period by patients in the Los Angeles County hospital system in California and their efficacy on AKN lesions. Our study includes one of the largest cohorts of patients reported to date and as such demonstrates the real-world effects that current treatment regimens for AKN have on patient outcomes nationwide.

Methods

We performed a retrospective cross-sectional analysis of patient medical records from the Los Angeles County hospital system i2b2 (i2b2 tranSMART Foundation) clinical data warehouse over a 6-year period (January 2017–January 2023). We used the International Statistical Classification of Diseases, Tenth Revision codes L73.0 (acne keloid) and L73.1 (pseudofolliculitis barbae) to conduct our search in order to identify as many patients with follicular disorders as possible to include in the study. Of the 478 total medical records we reviewed, 183 patients were included based on a diagnosis of AKN by a dermatologist.

We then collected data on patient demographics and treatments received, including whether patients had received monotherapy or combination therapy. Of the 183 patients we initially identified, 4 were excluded from the study because they had not received any treatment, and 78 were excluded because no treatment outcomes were documented. The 101 patients who were included had received either monotherapy or a combination of treatments. Treatment outcomes were categorized as either improvement in the number and appearance of papules and/or keloidlike plaques, maintenance of stable lesions (ie, well controlled), and/or resolution of lesions as documented by the treating physician. No patients had overall worsening of their disease.

Results

Of the 101 patients included in the study, 34 (33.7%) received a combination of topical, systemic, and procedural treatments; 34 (33.7%) received a combination of topical and procedural treatments; 17 (16.8%) were treated with topicals only; 13 (12.9%) were treated with a combination of topical and systemic treatments; and 3 (3.0%) were treated with monotherapy of either a topical, systemic, or procedural therapy. Systemic and/or procedural therapy combined with topicals was provided as a first-line treatment for 63 (62.4%) patients. Treatment escalation to systemic or procedural therapy for those who did not respond to topical treatment was observed in 23 (22.8%) patients. The average number of unique treatments received per patient was 3.67.

Clindamycin and clobetasol were the most prescribed topical treatments, doxycycline was the most prescribed systemic therapy, and intralesional (IL) triamcinolone was the most performed procedural therapy. The most common treatment regimens were topical clindamycin and clobetasol, topical clindamycin and clobetasol with IL triamcinolone, and topical clindamycin and clobetasol with both IL triamcinolone and doxycycline.

Improvement in AKN lesions was reported for the majority of patients with known treatment outcomes across all types of regimens. Ninety-eight percent (99/101) of patients had improvement in lesions, 55.5% (56/101) had well-controlled lesions, and 20.8% (21/101) achieved resolution of disease. The treatment outcomes are outlined in eTables 1 and 2.

Comment

Most clinicians opted for a multitherapy treatment regimen, and improvement was noted in most patients regardless of which regimen was chosen. As expected, patients who had mild or early disease generally received topical agents first, including most commonly a mid- to high-potency steroid, antibiotic, retinoid, and/or antifungal; specifically, clindamycin, clobetasol, and fluocinolone were the most common agents chosen. Patients with severe disease were more likely to receive systemic and/or procedural treatments, including oral antibiotics or IL steroid injections most commonly. Improvement was documented in the majority of patients using these treatment regimens, and some patients did achieve full resolution of disease.

Our data cannot be used to determine which treatment alone is most effective for patients with AKN, as the patients in our study had varying levels of disease activity and types of lesions, and most received combination therapy. What our data do show is that combination therapies often work well to control or improve disease, but also that current therapeutic options only rarely lead to full resolution of disease.

Limitations of our study included an inability to stratify disease, an inability to rigorously analyze specific treatment outcomes since most patients did not receive monotherapy. The strength of our study is its size, which allows us to show that many different treatment regimens currently are being employed by dermatologists to treat AKN, and most of these seem to be somewhat effective.

Conclusion

Acne keloidalis nuchae is difficult to treat due to a lack of understanding of which pathophysiologic mechanisms dominate in any given patient, a lack of good data on treatment outcomes, and the variability of ways that the disease manifests. Thus far, as shown by the patients described in this study, the most efficacious treatment regimens seem to be combination therapies that target the multifactorial causes of this disease. Physicians should continue to choose treatments based on disease severity and cutaneous manifestations, tailor their approach by accounting for patient preferences, and consider a multimodal approach to treatment.

Acne keloidalis nuchae (AKN) classically presents as chronic inflammation of the hair follicles on the occipital scalp/nape of the neck manifesting as papules and pustules that may progress to keloidlike scarring.¹ Photographs depicting the typical clinical presentation of AKN are shown in the Figure. In the literature, AKN has been described as primarily occurring in postpubertal males of African descent.² Despite its similar name, AKN is not related to acne vulgaris.³ The underlying cause of AKN is hypothesized to be multifactorial, including inflammation, infection, and trauma.² Acne keloidalis nuchae is most common in males aged 14 to 50 years, which may indicate that increased androgens contribute to its development.³ In some cases, patients have reported developing AKN lesions after receiving a haircut or shaving, suggesting a potential role of trauma to the hair follicles and secondary infection.² Histopathology typically shows a perifollicular inflammatory infiltrate that obscures the hair follicles with associated proximal fibrosis.⁴ On physical examination, dermoscopy can be used to visualize perifollicular pustules and fibrosis, which appears white, in the early stages of AKN. Patients may present with tufted hairs in more advanced stages.⁵ Patients with AKN often describe the lesions as pruritic and painful.²

In this study, we evaluated the most common treatment regimens used over a 6-year period by patients in the Los Angeles County hospital system in California and their efficacy on AKN lesions. Our study includes one of the largest cohorts of patients reported to date and as such demonstrates the real-world effects that current treatment regimens for AKN have on patient outcomes nationwide.

Methods

We performed a retrospective cross-sectional analysis of patient medical records from the Los Angeles County hospital system i2b2 (i2b2 tranSMART Foundation) clinical data warehouse over a 6-year period (January 2017–January 2023). We used the International Statistical Classification of Diseases, Tenth Revision codes L73.0 (acne keloid) and L73.1 (pseudofolliculitis barbae) to conduct our search in order to identify as many patients with follicular disorders as possible to include in the study. Of the 478 total medical records we reviewed, 183 patients were included based on a diagnosis of AKN by a dermatologist.

We then collected data on patient demographics and treatments received, including whether patients had received monotherapy or combination therapy. Of the 183 patients we initially identified, 4 were excluded from the study because they had not received any treatment, and 78 were excluded because no treatment outcomes were documented. The 101 patients who were included had received either monotherapy or a combination of treatments. Treatment outcomes were categorized as either improvement in the number and appearance of papules and/or keloidlike plaques, maintenance of stable lesions (ie, well controlled), and/or resolution of lesions as documented by the treating physician. No patients had overall worsening of their disease.

Results

Of the 101 patients included in the study, 34 (33.7%) received a combination of topical, systemic, and procedural treatments; 34 (33.7%) received a combination of topical and procedural treatments; 17 (16.8%) were treated with topicals only; 13 (12.9%) were treated with a combination of topical and systemic treatments; and 3 (3.0%) were treated with monotherapy of either a topical, systemic, or procedural therapy. Systemic and/or procedural therapy combined with topicals was provided as a first-line treatment for 63 (62.4%) patients. Treatment escalation to systemic or procedural therapy for those who did not respond to topical treatment was observed in 23 (22.8%) patients. The average number of unique treatments received per patient was 3.67.

Clindamycin and clobetasol were the most prescribed topical treatments, doxycycline was the most prescribed systemic therapy, and intralesional (IL) triamcinolone was the most performed procedural therapy. The most common treatment regimens were topical clindamycin and clobetasol, topical clindamycin and clobetasol with IL triamcinolone, and topical clindamycin and clobetasol with both IL triamcinolone and doxycycline.

Improvement in AKN lesions was reported for the majority of patients with known treatment outcomes across all types of regimens. Ninety-eight percent (99/101) of patients had improvement in lesions, 55.5% (56/101) had well-controlled lesions, and 20.8% (21/101) achieved resolution of disease. The treatment outcomes are outlined in eTables 1 and 2.

Comment

Most clinicians opted for a multitherapy treatment regimen, and improvement was noted in most patients regardless of which regimen was chosen. As expected, patients who had mild or early disease generally received topical agents first, including most commonly a mid- to high-potency steroid, antibiotic, retinoid, and/or antifungal; specifically, clindamycin, clobetasol, and fluocinolone were the most common agents chosen. Patients with severe disease were more likely to receive systemic and/or procedural treatments, including oral antibiotics or IL steroid injections most commonly. Improvement was documented in the majority of patients using these treatment regimens, and some patients did achieve full resolution of disease.

Our data cannot be used to determine which treatment alone is most effective for patients with AKN, as the patients in our study had varying levels of disease activity and types of lesions, and most received combination therapy. What our data do show is that combination therapies often work well to control or improve disease, but also that current therapeutic options only rarely lead to full resolution of disease.

Limitations of our study included an inability to stratify disease, an inability to rigorously analyze specific treatment outcomes since most patients did not receive monotherapy. The strength of our study is its size, which allows us to show that many different treatment regimens currently are being employed by dermatologists to treat AKN, and most of these seem to be somewhat effective.

Conclusion

Acne keloidalis nuchae is difficult to treat due to a lack of understanding of which pathophysiologic mechanisms dominate in any given patient, a lack of good data on treatment outcomes, and the variability of ways that the disease manifests. Thus far, as shown by the patients described in this study, the most efficacious treatment regimens seem to be combination therapies that target the multifactorial causes of this disease. Physicians should continue to choose treatments based on disease severity and cutaneous manifestations, tailor their approach by accounting for patient preferences, and consider a multimodal approach to treatment.

Acne keloidalis nuchae (AKN) classically presents as chronic inflammation of the hair follicles on the occipital scalp/nape of the neck manifesting as papules and pustules that may progress to keloidlike scarring.¹ Photographs depicting the typical clinical presentation of AKN are shown in the Figure. In the literature, AKN has been described as primarily occurring in postpubertal males of African descent.² Despite its similar name, AKN is not related to acne vulgaris.³ The underlying cause of AKN is hypothesized to be multifactorial, including inflammation, infection, and trauma.² Acne keloidalis nuchae is most common in males aged 14 to 50 years, which may indicate that increased androgens contribute to its development.³ In some cases, patients have reported developing AKN lesions after receiving a haircut or shaving, suggesting a potential role of trauma to the hair follicles and secondary infection.² Histopathology typically shows a perifollicular inflammatory infiltrate that obscures the hair follicles with associated proximal fibrosis.⁴ On physical examination, dermoscopy can be used to visualize perifollicular pustules and fibrosis, which appears white, in the early stages of AKN. Patients may present with tufted hairs in more advanced stages.⁵ Patients with AKN often describe the lesions as pruritic and painful.²

In this study, we evaluated the most common treatment regimens used over a 6-year period by patients in the Los Angeles County hospital system in California and their efficacy on AKN lesions. Our study includes one of the largest cohorts of patients reported to date and as such demonstrates the real-world effects that current treatment regimens for AKN have on patient outcomes nationwide.

Methods

We performed a retrospective cross-sectional analysis of patient medical records from the Los Angeles County hospital system i2b2 (i2b2 tranSMART Foundation) clinical data warehouse over a 6-year period (January 2017–January 2023). We used the International Statistical Classification of Diseases, Tenth Revision codes L73.0 (acne keloid) and L73.1 (pseudofolliculitis barbae) to conduct our search in order to identify as many patients with follicular disorders as possible to include in the study. Of the 478 total medical records we reviewed, 183 patients were included based on a diagnosis of AKN by a dermatologist.

We then collected data on patient demographics and treatments received, including whether patients had received monotherapy or combination therapy. Of the 183 patients we initially identified, 4 were excluded from the study because they had not received any treatment, and 78 were excluded because no treatment outcomes were documented. The 101 patients who were included had received either monotherapy or a combination of treatments. Treatment outcomes were categorized as either improvement in the number and appearance of papules and/or keloidlike plaques, maintenance of stable lesions (ie, well controlled), and/or resolution of lesions as documented by the treating physician. No patients had overall worsening of their disease.

Results

Of the 101 patients included in the study, 34 (33.7%) received a combination of topical, systemic, and procedural treatments; 34 (33.7%) received a combination of topical and procedural treatments; 17 (16.8%) were treated with topicals only; 13 (12.9%) were treated with a combination of topical and systemic treatments; and 3 (3.0%) were treated with monotherapy of either a topical, systemic, or procedural therapy. Systemic and/or procedural therapy combined with topicals was provided as a first-line treatment for 63 (62.4%) patients. Treatment escalation to systemic or procedural therapy for those who did not respond to topical treatment was observed in 23 (22.8%) patients. The average number of unique treatments received per patient was 3.67.

Clindamycin and clobetasol were the most prescribed topical treatments, doxycycline was the most prescribed systemic therapy, and intralesional (IL) triamcinolone was the most performed procedural therapy. The most common treatment regimens were topical clindamycin and clobetasol, topical clindamycin and clobetasol with IL triamcinolone, and topical clindamycin and clobetasol with both IL triamcinolone and doxycycline.

Improvement in AKN lesions was reported for the majority of patients with known treatment outcomes across all types of regimens. Ninety-eight percent (99/101) of patients had improvement in lesions, 55.5% (56/101) had well-controlled lesions, and 20.8% (21/101) achieved resolution of disease. The treatment outcomes are outlined in eTables 1 and 2.

Comment

Most clinicians opted for a multitherapy treatment regimen, and improvement was noted in most patients regardless of which regimen was chosen. As expected, patients who had mild or early disease generally received topical agents first, including most commonly a mid- to high-potency steroid, antibiotic, retinoid, and/or antifungal; specifically, clindamycin, clobetasol, and fluocinolone were the most common agents chosen. Patients with severe disease were more likely to receive systemic and/or procedural treatments, including oral antibiotics or IL steroid injections most commonly. Improvement was documented in the majority of patients using these treatment regimens, and some patients did achieve full resolution of disease.

Our data cannot be used to determine which treatment alone is most effective for patients with AKN, as the patients in our study had varying levels of disease activity and types of lesions, and most received combination therapy. What our data do show is that combination therapies often work well to control or improve disease, but also that current therapeutic options only rarely lead to full resolution of disease.

Limitations of our study included an inability to stratify disease, an inability to rigorously analyze specific treatment outcomes since most patients did not receive monotherapy. The strength of our study is its size, which allows us to show that many different treatment regimens currently are being employed by dermatologists to treat AKN, and most of these seem to be somewhat effective.

Conclusion

Acne keloidalis nuchae is difficult to treat due to a lack of understanding of which pathophysiologic mechanisms dominate in any given patient, a lack of good data on treatment outcomes, and the variability of ways that the disease manifests. Thus far, as shown by the patients described in this study, the most efficacious treatment regimens seem to be combination therapies that target the multifactorial causes of this disease. Physicians should continue to choose treatments based on disease severity and cutaneous manifestations, tailor their approach by accounting for patient preferences, and consider a multimodal approach to treatment.

The estimated prevalence of psoriasis in individuals older than 20 years in the United States has been reported at approximately 3%, or more than 7.5 million people.¹ There currently is no cure for psoriasis, and available therapeutics, including phototherapy,² topical therapies,³ systemic medications,⁴ and biologic agents,⁵ are focused only on controlling symptoms. The National Psoriasis Foundation defines an acceptable treatment response for plaque psoriasis as 3% or lower body surface area (BSA) involvement after 3 months of therapy, with a treat-to-target (TTT) goal of 1% or less BSA involvement.⁶

Cytokines are known to mediate psoriasis pathology, and biologic therapies target the signaling cascade of various cytokines. Biologics approved to treat moderate to severe plaque psoriasis include IgG monoclonal antibodies binding and inhibiting the activity of interleukin (IL)-17 (ixekizumab,⁷ secukinumab⁸), IL-23 (guselkumab,⁹ risankizumab,¹⁰ tildrakizumab¹¹), and IL-12/23 (ustekinumab¹²). Despite targeting these cytokines, biologics may not sufficiently suppress the symptoms of psoriatic disease and their severity in all patients. Adding a topical treatment to biologic therapy can augment clinical response without increasing the incidence of adverse effects^13-15 and may reduce the need to switch biologics due to ineffectiveness. Switching biologics likely would increase cost burden to the health care system and/or patient depending on their insurance plan and possibly introduce new safety and/or tolerability issues.^16,17

In patients who do not adequately respond to biologics, better responses were reported when topical medications including halobetasol propionate–tazarotene lotion¹⁶ or calcipotriene/betamethasone dipropionate foam^17,18 were administered. In randomized or open-label, real-world studies, patients with psoriasis responded well when topical medications were added to a biologic, such as tildrakizumab combined with halcinonide ointment 0.1%,¹⁹ etanercept combined with topical clobetasol propionate foam,²⁰ or adalimumab combined with calcipotriene/betamethasone dipropionate foam.²¹ No additional safety concerns were observed with the topical add-ons in any of these studies.

Tapinarof is an aryl hydrocarbon receptor agonist approved by the US Food and Drug Administration for topical treatment of plaque psoriasis in adults.²² It is a first-in-class small molecule with a novel mechanism of action that downregulates IL-17A and IL-17F and normalizes the skin barrier through expression of filaggrin, loricrin, and involucrin; it also has antioxidant activity.²³ In the phase 3 PSOARING 1 and 2 trials, daily application of tapinarof cream was safe and efficacious in patients with plaque psoriasis,^24,25 with a remittive (maintenance) effect of a median of approximately 4 months after discontinuation.²⁵ In these 2 phase 3 studies, tapinarof significantly (P<0.01 at week 12) relieved itch, which was seen rapidly (P<0.05 at week 2),²⁶ improved quality of life,²⁷ and led to high patient satisfaction.²⁷ When tapinarof cream was combined with deucravacitinib in a patient with severe plaque psoriasis, symptoms rapidly cleared, with a 75% decrease in disease severity after 4 weeks.²⁸

The objective of this prospective, open-label, real-world, single-center study was to assess the effectiveness, safety, and remittive (or maintenance) effect of nonsteroidal tapinarof cream 1% added to ongoing biologic therapy in patients with plaque psoriasis who were not adequately responding to a biologic alone.

Methods

Study Design and Participants—This prospective, open-label, real-world, single-center study assessed the safety and effectiveness of once-daily application of tapinarof cream 1% added to ongoing biologic therapy to help reach the National Psoriasis Foundation’s TTT goal of 1% or less BSA involvement at week 12 in patients with 3% or greater BSA involvement after taking the biologic for 24 weeks or more. The study protocol was approved by an institutional review board, and the study was conducted according to the ethical guidelines of the Declaration of Helsinki and Good Clinical Practice. All participants provided written informed consent before the study began.

Eligible participants were otherwise healthy males and females aged 18 years and older with moderate to severe plaque psoriasis (BSA involvement ≥3%) who had been treated with a biologic for 24 weeks or more. Patients were recruited from the Psoriasis Treatment Center of New Jersey (East Windsor, New Jersey). Exclusion criteria were recent use of oral systemic therapies (within 4 weeks of baseline) or topical therapies (within 2 weeks) to treat psoriasis, recent use of UVB (within 2 weeks) or psoralen plus UVA (within 4 weeks) phototherapy, or use of any investigational drug within 4 weeks of baseline (or within 5 pharmacokinetic/pharmacodynamic half-lives, whichever was longer). Patients who were pregnant or breastfeeding or who had any known hypersensitivity to the excipients of tapinarof cream also were excluded from the study.

Eligible participants received tapinarof cream 1% once daily plus their ongoing biologic for 12 weeks, after which tapinarof was discontinued and the biologic was continued for an additional 4 weeks. A remittive (maintenance) effect was assessed at week 16.

Study Outcomes—Safety and efficacy were evaluated at baseline and weeks 2, 4, 8, 12, and 16. The primary end point was the proportion of patients who reached the TTT goal of 1% or less BSA involvement at week 12. Secondary end points included the proportion of patients with 1% or less BSA involvement at weeks 2, 4, 8, and 16; and PGA scores, composite PGA multiplied by mean percentage of BSA involvement (PGA×BSA), and PASI scores at baseline and weeks 2, 4, 8, 12, and 16. The patient-reported outcomes of Dermatology Life Quality Index (DLQI) and Worst Itch Numeric Rating Scale (WI-NRS) scores also were evaluated at baseline and weeks 2, 4, 8, 12, and 16. In patients who had disease involvement on the scalp or genital region at baseline, Psoriasis Scalp Severity Index (PSSI) and Static Physician’s Global Assessment of Genitalia scores, respectively, were assessed at baseline and weeks 2, 4, 8, 12, and 16. Safety was determined by the incidence, severity, and relatedness of adverse events (AEs) and serious AEs.

Statistical Analysis—Approximately 30 participants were planned for enrollment and recruited consecutively as they were identified during screening against inclusion and exclusion criteria. Changes from baseline in all outcomes were summarized descriptively. Missing data were not imputed. Given the sample size, no formal statistical analyses were conducted. Safety was summarized by descriptively collating AEs and serious AEs, including their frequency, severity, and treatment relatedness.

Results

Thirty participants were enrolled in the study, and 20 fully completed the study. Nine discontinued treatment before week 12 (6 were lost to follow-up, 2 were terminated early by the investigators, and 1 voluntarily withdrew); 1 additional participant was lost to follow-up after week 12. Patients were predominantly male (20/30 [66.7%]) and White (21/30 [70.0%]); the mean age of all participants was 55.4 years, and the mean (SD) duration of psoriasis was 21.4 (15.0) years (Table 1). The mean baseline percentage of BSA involvement and mean baseline PGA, PASI, and DLQI scores are shown in Table 1. Most (19/30 [63.3%]) patients received biologics that inhibited IL-23 activity (guselkumab, risankizumab, tildrakizumab), approximately one-third (9/30 [30.0%]) received biologics that inhibited IL-17 activity (ixekizumab, secukinumab), and 2 (6.7%) received biologics that inhibited IL-12/IL-23 activity (ustekinumab)(Table 1).

For the primary end point, 52.4% (11/21) of patients reached the TTT goal (BSA involvement ≤1% after 12 weeks of treatment with tapinarof cream added to a prescribed biologic). The proportion of patients reaching the TTT goal increased over time with the combined treatment (eFigure 1). Additionally, the mean percentage of BSA involvement (eFigure 2) as well as the mean values for PGA (eFigure 3) and PGA×BSA decreased over time. The mean percentage of BSA involvement was 5.0% at baseline and dropped to 2.0% by week 12. Similar reductions were observed for PGA and PGA×BSA scores at week 12.

After discontinuing tapinarof cream at week 12 and receiving only the biologic for 4 weeks, the proportion of patients maintaining 1% or less BSA involvement fell to 40.0% (8/20) at week 16, which was closer to that observed at week 8 (36% [9/25]) than at week 12 (52.4% [11/21])(eFigure 1).

The mean PASI score was 5.5 at baseline, then decreased over time when tapinarof cream was combined with a biologic (eFigure 4), falling to 3.1 by week 2 and 1.6 by week 12; it was maintained at 1.7 at week 16. Nine (30.0%) participants had psoriasis on the scalp at baseline with a mean PSSI score of 2.6, which decreased to 0.83 by week 2. By week 12, the mean PSSI score remained stable at 0.95 in the 2 (9.5%) participants who still had scalp involvement. The mean PSSI score increased slightly to 1.45 after patients received only the biologic for 4 weeks. At baseline, 3 (10.0%) patients had genital involvement (mean Static Physician’s Global Assessment of Genitalia score, 0.27). Symptoms resolved in 2 (66.7%) of these patients at week 2 and stayed consistent until week 16; the third patient withdrew at week 2.

Both DLQI and WI-NRS scores decreased with use of tapinarof cream added to a biologic up to week 12 (eFigures 5 and 6). Mean DLQI scores were 5.3 at baseline and 3.1 at week 12. At week 16, the mean DLQI score remained stable at 2.8. Mean WI-NRS scores decreased from 4.0 at baseline to 2.7 at week 12 with the therapy combination; at week 16, the mean WI-NRS score fell further to 1.8.

A total of 6 AEs were reported in 5 (16.7%) patients (Table 2). The majority (4/6 [67.0%]) of AEs were considered mild. Two reported cases of COVID-19 were both considered mild and unrelated. Mild folliculitis and moderate worsening of psoriasis in 2 (6.7%) different patients were the only AEs considered related to treatment. No serious AEs were reported, and no patient withdrew from the study due to an AE.

Comment

In this prospective, open-label, real-world study, we found that when a patient’s response to a biologic is unsatisfactory, adding nonsteroidal tapinarof cream can enhance the treatment effect of the biologic. More than half of patients who were not adequately responding to a biologic reached the TTT goal at week 12, which was maintained in most of the responders for 4 weeks after tapinarof discontinuation. All measures of disease activity and quality of life (measured by the DLQI) improved when tapinarof was introduced, and the combination was well tolerated and without any unexpected safety signals.

Disease activity improvements we observed with the nonsteroidal tapinarof cream were consistent with those reported when topical steroidal therapies were given to patients responding poorly to their current biologic. Our primary end point (proportion of patients with BSA involvement ≤1% after 12 weeks) showed that half (52% [11/21]) of patients whose BSA involvement was 3% or greater with a biologic for 24 weeks or more reached the TTT goal after 12 weeks of tapinarof-biologic treatment. Other studies of halobetasol propionate–tazarotene lotion¹⁶ and calcipotriene/betamethasone dipropionate foam^17,18 added to the current biologic of poor responders found 60% to 68% of patients had reductions in their percentage BSA to 1% or lower at 12 to 16 weeks of treatment. Randomized studies showed etanercept plus topical clobetasol propionate foam²⁰ or adalimumab plus calcipotriene/betamethasone dipropionate foam²¹ similarly enhanced treatment effects vs biologic alone.

A phase 3 PSOARING trial demonstrated benefit from treatment with tapinarof alone, with a remittive effect of approximately 4 months after discontinuation.²⁵ Our data are consistent with these findings, with 40% (8/20) of patients demonstrating a remittive effect 4 weeks after discontinuing tapinarof while receiving a biologic. A similar maintenance effect was reported in another study in 50% (9/18) of patients treated with a biologic plus halobetasol propionate–tazarotene lotion.¹⁶ Additionally, when halcinonide ointment was given to patients receiving tildrakizumab, mean percentage of BSA involvement, PGA scores, PGA×BSA, and DLQI scores improved and were maintained 4 weeks after halcinonide ointment was stopped.¹⁹ Thus, topical therapy can augment and extend a biologic’s effect for up to 4 weeks. While a longer remittive effect is possible in our patients based on the 4-month remittive rate observed in the PSOARING trials,²⁵ further patient observation would be needed to confirm.

In our study, tapinarof cream added to a biologic had a good safety and tolerability profile. Few AEs were recorded, with most being mild in nature, and no serious AEs or discontinuations due to AEs were reported. Only 1 case of mild folliculitis and 1 case of moderate worsening of psoriasis were considered treatment related. Further, no unexpected or new safety signals with the tapinarof-biologic combination were observed compared with tapinarof alone.²⁷Prior studies have found that supplementing a biologic with topical therapy can reduce the probability of patients switching to another biologic.^16,19 We previously found that adding halobetasol propionate–tazarotene lotion¹⁶ or calcipotriene/betamethasone dipropionate foam¹⁷ to a biologic helped reduce the probability of switching biologics from 88% to 90% at baseline to 12% to 24% after 12 weeks of combined therapy. Such combinations also could prevent a less responsive patient from being prescribed a higher biologic dose.¹⁹ These are important research findings, as patients—even when not responding well to their current biologic—are more likely to be tolerating that biologic well, and switching to a new biologic may introduce new safety or tolerability concerns. Thus, by enhancing the effect of a biologic with a topical therapy, one can avoid increasing the dose of the current biologic or switching to a new biologic, either of which may increase safety and/or tolerability risks. Switching biologics also has increased cost implications to the health care system and/or the patient. When comparing the cost of adding halobetasol propionate–tazarotene lotion to a biologic compared with switching to another biologic, the cost was 1.2 to 2.9 times higher to switch, depending on the biologic, compared with a smaller incremental cost increase to add a topical to the current biologic.¹⁶ Similar observations were reported with calcipotriene/betamethasone dipropionate foam plus a biologic.¹⁷ Although we did not evaluate biologic switching here, we anticipate a similar clinical scenario with a tapinarof-biologic combination.

Limitations of our study included the open-label design, lack of a control arm, and the relatively small study population; however, for studies investigating the safety and effectiveness of a treatment in a real-world setting, these limitations are common and are not unexpected. Our results also are consistent with the overall improvement seen in other studies^16-21 examining the effects of adding a topical to a biologic. Future research is warranted to investigate a longer remittive effect and potential health care system and patient cost savings without having to switch biologics due to lack of effectiveness.

Conclusion

This study demonstrated that adjunctive use of nonsteroidal tapinarof cream 1% may enhance a biologic treatment effect in patients with moderate to severe plaque psoriasis, providing an adequate response for many patients who were not responding well to a biologic alone. Clinical outcomes improved with the tapinarof-biologic combination, and a remittive effect was noted 4 weeks after tapinarof discontinuation without any new safety signals. Adding tapinarof cream to a biologic also may prevent the need to switch biologics when patients do not sufficiently respond, preserving the safety and cost associated with a patient’s current biologic.

The estimated prevalence of psoriasis in individuals older than 20 years in the United States has been reported at approximately 3%, or more than 7.5 million people.¹ There currently is no cure for psoriasis, and available therapeutics, including phototherapy,² topical therapies,³ systemic medications,⁴ and biologic agents,⁵ are focused only on controlling symptoms. The National Psoriasis Foundation defines an acceptable treatment response for plaque psoriasis as 3% or lower body surface area (BSA) involvement after 3 months of therapy, with a treat-to-target (TTT) goal of 1% or less BSA involvement.⁶

Cytokines are known to mediate psoriasis pathology, and biologic therapies target the signaling cascade of various cytokines. Biologics approved to treat moderate to severe plaque psoriasis include IgG monoclonal antibodies binding and inhibiting the activity of interleukin (IL)-17 (ixekizumab,⁷ secukinumab⁸), IL-23 (guselkumab,⁹ risankizumab,¹⁰ tildrakizumab¹¹), and IL-12/23 (ustekinumab¹²). Despite targeting these cytokines, biologics may not sufficiently suppress the symptoms of psoriatic disease and their severity in all patients. Adding a topical treatment to biologic therapy can augment clinical response without increasing the incidence of adverse effects^13-15 and may reduce the need to switch biologics due to ineffectiveness. Switching biologics likely would increase cost burden to the health care system and/or patient depending on their insurance plan and possibly introduce new safety and/or tolerability issues.^16,17

In patients who do not adequately respond to biologics, better responses were reported when topical medications including halobetasol propionate–tazarotene lotion¹⁶ or calcipotriene/betamethasone dipropionate foam^17,18 were administered. In randomized or open-label, real-world studies, patients with psoriasis responded well when topical medications were added to a biologic, such as tildrakizumab combined with halcinonide ointment 0.1%,¹⁹ etanercept combined with topical clobetasol propionate foam,²⁰ or adalimumab combined with calcipotriene/betamethasone dipropionate foam.²¹ No additional safety concerns were observed with the topical add-ons in any of these studies.

Tapinarof is an aryl hydrocarbon receptor agonist approved by the US Food and Drug Administration for topical treatment of plaque psoriasis in adults.²² It is a first-in-class small molecule with a novel mechanism of action that downregulates IL-17A and IL-17F and normalizes the skin barrier through expression of filaggrin, loricrin, and involucrin; it also has antioxidant activity.²³ In the phase 3 PSOARING 1 and 2 trials, daily application of tapinarof cream was safe and efficacious in patients with plaque psoriasis,^24,25 with a remittive (maintenance) effect of a median of approximately 4 months after discontinuation.²⁵ In these 2 phase 3 studies, tapinarof significantly (P<0.01 at week 12) relieved itch, which was seen rapidly (P<0.05 at week 2),²⁶ improved quality of life,²⁷ and led to high patient satisfaction.²⁷ When tapinarof cream was combined with deucravacitinib in a patient with severe plaque psoriasis, symptoms rapidly cleared, with a 75% decrease in disease severity after 4 weeks.²⁸

The objective of this prospective, open-label, real-world, single-center study was to assess the effectiveness, safety, and remittive (or maintenance) effect of nonsteroidal tapinarof cream 1% added to ongoing biologic therapy in patients with plaque psoriasis who were not adequately responding to a biologic alone.

Methods

Study Design and Participants—This prospective, open-label, real-world, single-center study assessed the safety and effectiveness of once-daily application of tapinarof cream 1% added to ongoing biologic therapy to help reach the National Psoriasis Foundation’s TTT goal of 1% or less BSA involvement at week 12 in patients with 3% or greater BSA involvement after taking the biologic for 24 weeks or more. The study protocol was approved by an institutional review board, and the study was conducted according to the ethical guidelines of the Declaration of Helsinki and Good Clinical Practice. All participants provided written informed consent before the study began.

Eligible participants were otherwise healthy males and females aged 18 years and older with moderate to severe plaque psoriasis (BSA involvement ≥3%) who had been treated with a biologic for 24 weeks or more. Patients were recruited from the Psoriasis Treatment Center of New Jersey (East Windsor, New Jersey). Exclusion criteria were recent use of oral systemic therapies (within 4 weeks of baseline) or topical therapies (within 2 weeks) to treat psoriasis, recent use of UVB (within 2 weeks) or psoralen plus UVA (within 4 weeks) phototherapy, or use of any investigational drug within 4 weeks of baseline (or within 5 pharmacokinetic/pharmacodynamic half-lives, whichever was longer). Patients who were pregnant or breastfeeding or who had any known hypersensitivity to the excipients of tapinarof cream also were excluded from the study.

Eligible participants received tapinarof cream 1% once daily plus their ongoing biologic for 12 weeks, after which tapinarof was discontinued and the biologic was continued for an additional 4 weeks. A remittive (maintenance) effect was assessed at week 16.

Study Outcomes—Safety and efficacy were evaluated at baseline and weeks 2, 4, 8, 12, and 16. The primary end point was the proportion of patients who reached the TTT goal of 1% or less BSA involvement at week 12. Secondary end points included the proportion of patients with 1% or less BSA involvement at weeks 2, 4, 8, and 16; and PGA scores, composite PGA multiplied by mean percentage of BSA involvement (PGA×BSA), and PASI scores at baseline and weeks 2, 4, 8, 12, and 16. The patient-reported outcomes of Dermatology Life Quality Index (DLQI) and Worst Itch Numeric Rating Scale (WI-NRS) scores also were evaluated at baseline and weeks 2, 4, 8, 12, and 16. In patients who had disease involvement on the scalp or genital region at baseline, Psoriasis Scalp Severity Index (PSSI) and Static Physician’s Global Assessment of Genitalia scores, respectively, were assessed at baseline and weeks 2, 4, 8, 12, and 16. Safety was determined by the incidence, severity, and relatedness of adverse events (AEs) and serious AEs.

Statistical Analysis—Approximately 30 participants were planned for enrollment and recruited consecutively as they were identified during screening against inclusion and exclusion criteria. Changes from baseline in all outcomes were summarized descriptively. Missing data were not imputed. Given the sample size, no formal statistical analyses were conducted. Safety was summarized by descriptively collating AEs and serious AEs, including their frequency, severity, and treatment relatedness.

Results

Thirty participants were enrolled in the study, and 20 fully completed the study. Nine discontinued treatment before week 12 (6 were lost to follow-up, 2 were terminated early by the investigators, and 1 voluntarily withdrew); 1 additional participant was lost to follow-up after week 12. Patients were predominantly male (20/30 [66.7%]) and White (21/30 [70.0%]); the mean age of all participants was 55.4 years, and the mean (SD) duration of psoriasis was 21.4 (15.0) years (Table 1). The mean baseline percentage of BSA involvement and mean baseline PGA, PASI, and DLQI scores are shown in Table 1. Most (19/30 [63.3%]) patients received biologics that inhibited IL-23 activity (guselkumab, risankizumab, tildrakizumab), approximately one-third (9/30 [30.0%]) received biologics that inhibited IL-17 activity (ixekizumab, secukinumab), and 2 (6.7%) received biologics that inhibited IL-12/IL-23 activity (ustekinumab)(Table 1).

For the primary end point, 52.4% (11/21) of patients reached the TTT goal (BSA involvement ≤1% after 12 weeks of treatment with tapinarof cream added to a prescribed biologic). The proportion of patients reaching the TTT goal increased over time with the combined treatment (eFigure 1). Additionally, the mean percentage of BSA involvement (eFigure 2) as well as the mean values for PGA (eFigure 3) and PGA×BSA decreased over time. The mean percentage of BSA involvement was 5.0% at baseline and dropped to 2.0% by week 12. Similar reductions were observed for PGA and PGA×BSA scores at week 12.

After discontinuing tapinarof cream at week 12 and receiving only the biologic for 4 weeks, the proportion of patients maintaining 1% or less BSA involvement fell to 40.0% (8/20) at week 16, which was closer to that observed at week 8 (36% [9/25]) than at week 12 (52.4% [11/21])(eFigure 1).

The mean PASI score was 5.5 at baseline, then decreased over time when tapinarof cream was combined with a biologic (eFigure 4), falling to 3.1 by week 2 and 1.6 by week 12; it was maintained at 1.7 at week 16. Nine (30.0%) participants had psoriasis on the scalp at baseline with a mean PSSI score of 2.6, which decreased to 0.83 by week 2. By week 12, the mean PSSI score remained stable at 0.95 in the 2 (9.5%) participants who still had scalp involvement. The mean PSSI score increased slightly to 1.45 after patients received only the biologic for 4 weeks. At baseline, 3 (10.0%) patients had genital involvement (mean Static Physician’s Global Assessment of Genitalia score, 0.27). Symptoms resolved in 2 (66.7%) of these patients at week 2 and stayed consistent until week 16; the third patient withdrew at week 2.

Both DLQI and WI-NRS scores decreased with use of tapinarof cream added to a biologic up to week 12 (eFigures 5 and 6). Mean DLQI scores were 5.3 at baseline and 3.1 at week 12. At week 16, the mean DLQI score remained stable at 2.8. Mean WI-NRS scores decreased from 4.0 at baseline to 2.7 at week 12 with the therapy combination; at week 16, the mean WI-NRS score fell further to 1.8.

A total of 6 AEs were reported in 5 (16.7%) patients (Table 2). The majority (4/6 [67.0%]) of AEs were considered mild. Two reported cases of COVID-19 were both considered mild and unrelated. Mild folliculitis and moderate worsening of psoriasis in 2 (6.7%) different patients were the only AEs considered related to treatment. No serious AEs were reported, and no patient withdrew from the study due to an AE.

Comment

In this prospective, open-label, real-world study, we found that when a patient’s response to a biologic is unsatisfactory, adding nonsteroidal tapinarof cream can enhance the treatment effect of the biologic. More than half of patients who were not adequately responding to a biologic reached the TTT goal at week 12, which was maintained in most of the responders for 4 weeks after tapinarof discontinuation. All measures of disease activity and quality of life (measured by the DLQI) improved when tapinarof was introduced, and the combination was well tolerated and without any unexpected safety signals.

Disease activity improvements we observed with the nonsteroidal tapinarof cream were consistent with those reported when topical steroidal therapies were given to patients responding poorly to their current biologic. Our primary end point (proportion of patients with BSA involvement ≤1% after 12 weeks) showed that half (52% [11/21]) of patients whose BSA involvement was 3% or greater with a biologic for 24 weeks or more reached the TTT goal after 12 weeks of tapinarof-biologic treatment. Other studies of halobetasol propionate–tazarotene lotion¹⁶ and calcipotriene/betamethasone dipropionate foam^17,18 added to the current biologic of poor responders found 60% to 68% of patients had reductions in their percentage BSA to 1% or lower at 12 to 16 weeks of treatment. Randomized studies showed etanercept plus topical clobetasol propionate foam²⁰ or adalimumab plus calcipotriene/betamethasone dipropionate foam²¹ similarly enhanced treatment effects vs biologic alone.

A phase 3 PSOARING trial demonstrated benefit from treatment with tapinarof alone, with a remittive effect of approximately 4 months after discontinuation.²⁵ Our data are consistent with these findings, with 40% (8/20) of patients demonstrating a remittive effect 4 weeks after discontinuing tapinarof while receiving a biologic. A similar maintenance effect was reported in another study in 50% (9/18) of patients treated with a biologic plus halobetasol propionate–tazarotene lotion.¹⁶ Additionally, when halcinonide ointment was given to patients receiving tildrakizumab, mean percentage of BSA involvement, PGA scores, PGA×BSA, and DLQI scores improved and were maintained 4 weeks after halcinonide ointment was stopped.¹⁹ Thus, topical therapy can augment and extend a biologic’s effect for up to 4 weeks. While a longer remittive effect is possible in our patients based on the 4-month remittive rate observed in the PSOARING trials,²⁵ further patient observation would be needed to confirm.

In our study, tapinarof cream added to a biologic had a good safety and tolerability profile. Few AEs were recorded, with most being mild in nature, and no serious AEs or discontinuations due to AEs were reported. Only 1 case of mild folliculitis and 1 case of moderate worsening of psoriasis were considered treatment related. Further, no unexpected or new safety signals with the tapinarof-biologic combination were observed compared with tapinarof alone.²⁷Prior studies have found that supplementing a biologic with topical therapy can reduce the probability of patients switching to another biologic.^16,19 We previously found that adding halobetasol propionate–tazarotene lotion¹⁶ or calcipotriene/betamethasone dipropionate foam¹⁷ to a biologic helped reduce the probability of switching biologics from 88% to 90% at baseline to 12% to 24% after 12 weeks of combined therapy. Such combinations also could prevent a less responsive patient from being prescribed a higher biologic dose.¹⁹ These are important research findings, as patients—even when not responding well to their current biologic—are more likely to be tolerating that biologic well, and switching to a new biologic may introduce new safety or tolerability concerns. Thus, by enhancing the effect of a biologic with a topical therapy, one can avoid increasing the dose of the current biologic or switching to a new biologic, either of which may increase safety and/or tolerability risks. Switching biologics also has increased cost implications to the health care system and/or the patient. When comparing the cost of adding halobetasol propionate–tazarotene lotion to a biologic compared with switching to another biologic, the cost was 1.2 to 2.9 times higher to switch, depending on the biologic, compared with a smaller incremental cost increase to add a topical to the current biologic.¹⁶ Similar observations were reported with calcipotriene/betamethasone dipropionate foam plus a biologic.¹⁷ Although we did not evaluate biologic switching here, we anticipate a similar clinical scenario with a tapinarof-biologic combination.

Limitations of our study included the open-label design, lack of a control arm, and the relatively small study population; however, for studies investigating the safety and effectiveness of a treatment in a real-world setting, these limitations are common and are not unexpected. Our results also are consistent with the overall improvement seen in other studies^16-21 examining the effects of adding a topical to a biologic. Future research is warranted to investigate a longer remittive effect and potential health care system and patient cost savings without having to switch biologics due to lack of effectiveness.

Conclusion

This study demonstrated that adjunctive use of nonsteroidal tapinarof cream 1% may enhance a biologic treatment effect in patients with moderate to severe plaque psoriasis, providing an adequate response for many patients who were not responding well to a biologic alone. Clinical outcomes improved with the tapinarof-biologic combination, and a remittive effect was noted 4 weeks after tapinarof discontinuation without any new safety signals. Adding tapinarof cream to a biologic also may prevent the need to switch biologics when patients do not sufficiently respond, preserving the safety and cost associated with a patient’s current biologic.

The estimated prevalence of psoriasis in individuals older than 20 years in the United States has been reported at approximately 3%, or more than 7.5 million people.¹ There currently is no cure for psoriasis, and available therapeutics, including phototherapy,² topical therapies,³ systemic medications,⁴ and biologic agents,⁵ are focused only on controlling symptoms. The National Psoriasis Foundation defines an acceptable treatment response for plaque psoriasis as 3% or lower body surface area (BSA) involvement after 3 months of therapy, with a treat-to-target (TTT) goal of 1% or less BSA involvement.⁶

Cytokines are known to mediate psoriasis pathology, and biologic therapies target the signaling cascade of various cytokines. Biologics approved to treat moderate to severe plaque psoriasis include IgG monoclonal antibodies binding and inhibiting the activity of interleukin (IL)-17 (ixekizumab,⁷ secukinumab⁸), IL-23 (guselkumab,⁹ risankizumab,¹⁰ tildrakizumab¹¹), and IL-12/23 (ustekinumab¹²). Despite targeting these cytokines, biologics may not sufficiently suppress the symptoms of psoriatic disease and their severity in all patients. Adding a topical treatment to biologic therapy can augment clinical response without increasing the incidence of adverse effects^13-15 and may reduce the need to switch biologics due to ineffectiveness. Switching biologics likely would increase cost burden to the health care system and/or patient depending on their insurance plan and possibly introduce new safety and/or tolerability issues.^16,17

In patients who do not adequately respond to biologics, better responses were reported when topical medications including halobetasol propionate–tazarotene lotion¹⁶ or calcipotriene/betamethasone dipropionate foam^17,18 were administered. In randomized or open-label, real-world studies, patients with psoriasis responded well when topical medications were added to a biologic, such as tildrakizumab combined with halcinonide ointment 0.1%,¹⁹ etanercept combined with topical clobetasol propionate foam,²⁰ or adalimumab combined with calcipotriene/betamethasone dipropionate foam.²¹ No additional safety concerns were observed with the topical add-ons in any of these studies.

Tapinarof is an aryl hydrocarbon receptor agonist approved by the US Food and Drug Administration for topical treatment of plaque psoriasis in adults.²² It is a first-in-class small molecule with a novel mechanism of action that downregulates IL-17A and IL-17F and normalizes the skin barrier through expression of filaggrin, loricrin, and involucrin; it also has antioxidant activity.²³ In the phase 3 PSOARING 1 and 2 trials, daily application of tapinarof cream was safe and efficacious in patients with plaque psoriasis,^24,25 with a remittive (maintenance) effect of a median of approximately 4 months after discontinuation.²⁵ In these 2 phase 3 studies, tapinarof significantly (P<0.01 at week 12) relieved itch, which was seen rapidly (P<0.05 at week 2),²⁶ improved quality of life,²⁷ and led to high patient satisfaction.²⁷ When tapinarof cream was combined with deucravacitinib in a patient with severe plaque psoriasis, symptoms rapidly cleared, with a 75% decrease in disease severity after 4 weeks.²⁸

The objective of this prospective, open-label, real-world, single-center study was to assess the effectiveness, safety, and remittive (or maintenance) effect of nonsteroidal tapinarof cream 1% added to ongoing biologic therapy in patients with plaque psoriasis who were not adequately responding to a biologic alone.

Methods

Study Design and Participants—This prospective, open-label, real-world, single-center study assessed the safety and effectiveness of once-daily application of tapinarof cream 1% added to ongoing biologic therapy to help reach the National Psoriasis Foundation’s TTT goal of 1% or less BSA involvement at week 12 in patients with 3% or greater BSA involvement after taking the biologic for 24 weeks or more. The study protocol was approved by an institutional review board, and the study was conducted according to the ethical guidelines of the Declaration of Helsinki and Good Clinical Practice. All participants provided written informed consent before the study began.

Eligible participants were otherwise healthy males and females aged 18 years and older with moderate to severe plaque psoriasis (BSA involvement ≥3%) who had been treated with a biologic for 24 weeks or more. Patients were recruited from the Psoriasis Treatment Center of New Jersey (East Windsor, New Jersey). Exclusion criteria were recent use of oral systemic therapies (within 4 weeks of baseline) or topical therapies (within 2 weeks) to treat psoriasis, recent use of UVB (within 2 weeks) or psoralen plus UVA (within 4 weeks) phototherapy, or use of any investigational drug within 4 weeks of baseline (or within 5 pharmacokinetic/pharmacodynamic half-lives, whichever was longer). Patients who were pregnant or breastfeeding or who had any known hypersensitivity to the excipients of tapinarof cream also were excluded from the study.

Eligible participants received tapinarof cream 1% once daily plus their ongoing biologic for 12 weeks, after which tapinarof was discontinued and the biologic was continued for an additional 4 weeks. A remittive (maintenance) effect was assessed at week 16.

Study Outcomes—Safety and efficacy were evaluated at baseline and weeks 2, 4, 8, 12, and 16. The primary end point was the proportion of patients who reached the TTT goal of 1% or less BSA involvement at week 12. Secondary end points included the proportion of patients with 1% or less BSA involvement at weeks 2, 4, 8, and 16; and PGA scores, composite PGA multiplied by mean percentage of BSA involvement (PGA×BSA), and PASI scores at baseline and weeks 2, 4, 8, 12, and 16. The patient-reported outcomes of Dermatology Life Quality Index (DLQI) and Worst Itch Numeric Rating Scale (WI-NRS) scores also were evaluated at baseline and weeks 2, 4, 8, 12, and 16. In patients who had disease involvement on the scalp or genital region at baseline, Psoriasis Scalp Severity Index (PSSI) and Static Physician’s Global Assessment of Genitalia scores, respectively, were assessed at baseline and weeks 2, 4, 8, 12, and 16. Safety was determined by the incidence, severity, and relatedness of adverse events (AEs) and serious AEs.

Statistical Analysis—Approximately 30 participants were planned for enrollment and recruited consecutively as they were identified during screening against inclusion and exclusion criteria. Changes from baseline in all outcomes were summarized descriptively. Missing data were not imputed. Given the sample size, no formal statistical analyses were conducted. Safety was summarized by descriptively collating AEs and serious AEs, including their frequency, severity, and treatment relatedness.

Results

Thirty participants were enrolled in the study, and 20 fully completed the study. Nine discontinued treatment before week 12 (6 were lost to follow-up, 2 were terminated early by the investigators, and 1 voluntarily withdrew); 1 additional participant was lost to follow-up after week 12. Patients were predominantly male (20/30 [66.7%]) and White (21/30 [70.0%]); the mean age of all participants was 55.4 years, and the mean (SD) duration of psoriasis was 21.4 (15.0) years (Table 1). The mean baseline percentage of BSA involvement and mean baseline PGA, PASI, and DLQI scores are shown in Table 1. Most (19/30 [63.3%]) patients received biologics that inhibited IL-23 activity (guselkumab, risankizumab, tildrakizumab), approximately one-third (9/30 [30.0%]) received biologics that inhibited IL-17 activity (ixekizumab, secukinumab), and 2 (6.7%) received biologics that inhibited IL-12/IL-23 activity (ustekinumab)(Table 1).

For the primary end point, 52.4% (11/21) of patients reached the TTT goal (BSA involvement ≤1% after 12 weeks of treatment with tapinarof cream added to a prescribed biologic). The proportion of patients reaching the TTT goal increased over time with the combined treatment (eFigure 1). Additionally, the mean percentage of BSA involvement (eFigure 2) as well as the mean values for PGA (eFigure 3) and PGA×BSA decreased over time. The mean percentage of BSA involvement was 5.0% at baseline and dropped to 2.0% by week 12. Similar reductions were observed for PGA and PGA×BSA scores at week 12.

After discontinuing tapinarof cream at week 12 and receiving only the biologic for 4 weeks, the proportion of patients maintaining 1% or less BSA involvement fell to 40.0% (8/20) at week 16, which was closer to that observed at week 8 (36% [9/25]) than at week 12 (52.4% [11/21])(eFigure 1).

The mean PASI score was 5.5 at baseline, then decreased over time when tapinarof cream was combined with a biologic (eFigure 4), falling to 3.1 by week 2 and 1.6 by week 12; it was maintained at 1.7 at week 16. Nine (30.0%) participants had psoriasis on the scalp at baseline with a mean PSSI score of 2.6, which decreased to 0.83 by week 2. By week 12, the mean PSSI score remained stable at 0.95 in the 2 (9.5%) participants who still had scalp involvement. The mean PSSI score increased slightly to 1.45 after patients received only the biologic for 4 weeks. At baseline, 3 (10.0%) patients had genital involvement (mean Static Physician’s Global Assessment of Genitalia score, 0.27). Symptoms resolved in 2 (66.7%) of these patients at week 2 and stayed consistent until week 16; the third patient withdrew at week 2.

Both DLQI and WI-NRS scores decreased with use of tapinarof cream added to a biologic up to week 12 (eFigures 5 and 6). Mean DLQI scores were 5.3 at baseline and 3.1 at week 12. At week 16, the mean DLQI score remained stable at 2.8. Mean WI-NRS scores decreased from 4.0 at baseline to 2.7 at week 12 with the therapy combination; at week 16, the mean WI-NRS score fell further to 1.8.

A total of 6 AEs were reported in 5 (16.7%) patients (Table 2). The majority (4/6 [67.0%]) of AEs were considered mild. Two reported cases of COVID-19 were both considered mild and unrelated. Mild folliculitis and moderate worsening of psoriasis in 2 (6.7%) different patients were the only AEs considered related to treatment. No serious AEs were reported, and no patient withdrew from the study due to an AE.

Comment

In this prospective, open-label, real-world study, we found that when a patient’s response to a biologic is unsatisfactory, adding nonsteroidal tapinarof cream can enhance the treatment effect of the biologic. More than half of patients who were not adequately responding to a biologic reached the TTT goal at week 12, which was maintained in most of the responders for 4 weeks after tapinarof discontinuation. All measures of disease activity and quality of life (measured by the DLQI) improved when tapinarof was introduced, and the combination was well tolerated and without any unexpected safety signals.

Disease activity improvements we observed with the nonsteroidal tapinarof cream were consistent with those reported when topical steroidal therapies were given to patients responding poorly to their current biologic. Our primary end point (proportion of patients with BSA involvement ≤1% after 12 weeks) showed that half (52% [11/21]) of patients whose BSA involvement was 3% or greater with a biologic for 24 weeks or more reached the TTT goal after 12 weeks of tapinarof-biologic treatment. Other studies of halobetasol propionate–tazarotene lotion¹⁶ and calcipotriene/betamethasone dipropionate foam^17,18 added to the current biologic of poor responders found 60% to 68% of patients had reductions in their percentage BSA to 1% or lower at 12 to 16 weeks of treatment. Randomized studies showed etanercept plus topical clobetasol propionate foam²⁰ or adalimumab plus calcipotriene/betamethasone dipropionate foam²¹ similarly enhanced treatment effects vs biologic alone.

A phase 3 PSOARING trial demonstrated benefit from treatment with tapinarof alone, with a remittive effect of approximately 4 months after discontinuation.²⁵ Our data are consistent with these findings, with 40% (8/20) of patients demonstrating a remittive effect 4 weeks after discontinuing tapinarof while receiving a biologic. A similar maintenance effect was reported in another study in 50% (9/18) of patients treated with a biologic plus halobetasol propionate–tazarotene lotion.¹⁶ Additionally, when halcinonide ointment was given to patients receiving tildrakizumab, mean percentage of BSA involvement, PGA scores, PGA×BSA, and DLQI scores improved and were maintained 4 weeks after halcinonide ointment was stopped.¹⁹ Thus, topical therapy can augment and extend a biologic’s effect for up to 4 weeks. While a longer remittive effect is possible in our patients based on the 4-month remittive rate observed in the PSOARING trials,²⁵ further patient observation would be needed to confirm.

In our study, tapinarof cream added to a biologic had a good safety and tolerability profile. Few AEs were recorded, with most being mild in nature, and no serious AEs or discontinuations due to AEs were reported. Only 1 case of mild folliculitis and 1 case of moderate worsening of psoriasis were considered treatment related. Further, no unexpected or new safety signals with the tapinarof-biologic combination were observed compared with tapinarof alone.²⁷Prior studies have found that supplementing a biologic with topical therapy can reduce the probability of patients switching to another biologic.^16,19 We previously found that adding halobetasol propionate–tazarotene lotion¹⁶ or calcipotriene/betamethasone dipropionate foam¹⁷ to a biologic helped reduce the probability of switching biologics from 88% to 90% at baseline to 12% to 24% after 12 weeks of combined therapy. Such combinations also could prevent a less responsive patient from being prescribed a higher biologic dose.¹⁹ These are important research findings, as patients—even when not responding well to their current biologic—are more likely to be tolerating that biologic well, and switching to a new biologic may introduce new safety or tolerability concerns. Thus, by enhancing the effect of a biologic with a topical therapy, one can avoid increasing the dose of the current biologic or switching to a new biologic, either of which may increase safety and/or tolerability risks. Switching biologics also has increased cost implications to the health care system and/or the patient. When comparing the cost of adding halobetasol propionate–tazarotene lotion to a biologic compared with switching to another biologic, the cost was 1.2 to 2.9 times higher to switch, depending on the biologic, compared with a smaller incremental cost increase to add a topical to the current biologic.¹⁶ Similar observations were reported with calcipotriene/betamethasone dipropionate foam plus a biologic.¹⁷ Although we did not evaluate biologic switching here, we anticipate a similar clinical scenario with a tapinarof-biologic combination.

Limitations of our study included the open-label design, lack of a control arm, and the relatively small study population; however, for studies investigating the safety and effectiveness of a treatment in a real-world setting, these limitations are common and are not unexpected. Our results also are consistent with the overall improvement seen in other studies^16-21 examining the effects of adding a topical to a biologic. Future research is warranted to investigate a longer remittive effect and potential health care system and patient cost savings without having to switch biologics due to lack of effectiveness.

Conclusion

This study demonstrated that adjunctive use of nonsteroidal tapinarof cream 1% may enhance a biologic treatment effect in patients with moderate to severe plaque psoriasis, providing an adequate response for many patients who were not responding well to a biologic alone. Clinical outcomes improved with the tapinarof-biologic combination, and a remittive effect was noted 4 weeks after tapinarof discontinuation without any new safety signals. Adding tapinarof cream to a biologic also may prevent the need to switch biologics when patients do not sufficiently respond, preserving the safety and cost associated with a patient’s current biologic.