Validation of two predictive models designed to aid clinicians in identifying Stevens-Johnson syndrome/toxic epidermal necrolysis: A single institution retrospective review

Abstract

To the Editor: Stevens-Johnson syndrome and toxic epidermal necrolysis (SJS/TEN) are severe cutaneous drug reactions resulting in epidermal necrolysis of the skin as well as the mucosal membranes. Early recognition can facilitate appropriate triage; however, clinically diagnosing SJS/TEN may be challenging due to variability in disease presentation.1Lerch M. Mainetti C. Terziroli Beretta-Piccoli B. Harr T. Current perspectives on Stevens-Johnson syndrome and toxic epidermal necrolysis.Clin Rev Allergy Immunol. 2018; 54: 147-176Crossref PubMed Scopus (131) Google Scholar In 2019, 2 predictive models were first proposed to aid in distinguishing SJS/TEN from its clinical mimickers.2Weinkle A. Pettit C. Jani A. et al.Distinguishing Stevens-Johnson syndrome/toxic epidermal necrolysis from clinical mimickers during inpatient dermatologic consultation—a retrospective chart review.J Am Acad Dermatol. 2019; 81: 749-757Abstract Full Text Full Text PDF PubMed Scopus (13) Google Scholar We sought to externally validate these models among patients transferred due to the concerns of SJS/TEN. A retrospective chart review of patients transferred to Parkland Memorial Hospital for suspected SJS/TEN from October 1, 2012 to December 31, 2018 was performed. Two groups of patients were identified from an internal dermatology consultation list: those with a confirmed diagnosis of SJS/TEN and those with other conditions. Biopsies were performed for patients with an unclear diagnosis at presentation. If the biopsy results were non-specific, the final diagnosis was based on both clinical and histopathologic data. Initial dermatology consult notes were reviewed for the presence or absence of 5 characteristics (Nikolsky sign, atypical target, fever, lymphopenia, and mucosal involvement), and the numerical predictive probability was calculated using the 2 previously published predictive models (Supplemental material available via Mendeley at https://doi.org/10.17632/22h4ghj8wk.1).2Weinkle A. Pettit C. Jani A. et al.Distinguishing Stevens-Johnson syndrome/toxic epidermal necrolysis from clinical mimickers during inpatient dermatologic consultation—a retrospective chart review.J Am Acad Dermatol. 2019; 81: 749-757Abstract Full Text Full Text PDF PubMed Scopus (13) Google Scholar Receiver operating characteristic curves and area under the curve (AUC) were used to assess the models’ discriminatory capability, with a maximum AUC of 1 meaning that the diagnostic test was perfect in differentiating between the 2 groups.3Hajian-Tilaki K. Receiver operating characteristic (ROC) curve analysis for medical diagnostic test evaluation.Caspian J Intern Med. 2013; 4: 627-635PubMed Google Scholar Statistical analysis was also applied to evaluate additional clinical characteristics in our cohort that could aid in diagnosing SJS/TEN. A total of 193 patients were transferred for the evaluation of possible SJS/TEN. Of these, 68 patients had an ultimate diagnosis of SJS/TEN, and 125 patients had an alternative diagnosis. Using the predictive models, we were able to detect an AUC of 0.9251 (95% confidence interval, 0.8875-0.9628; P < .0001) when lymphopenia was included in the model and an AUC of 0.9451 (95% confidence interval, 0.9144-0.9757; P < .0001) when mucosal involvement was included in the model (Fig 1). Additional clinical characteristics that were compared between the groups are listed in Table I.Table IDemographic and clinical characteristics of patients with Stevens-Johnson syndrome and toxic epidermal necrolysis compared to non–Stevens-Johnson syndrome and toxic epidermal necrolysis diagnosesDemographic and clinical characteristicsSJS/TEN (N = 68)Non-SJS/TEN (N = 125)P value∗Continuous data were analyzed using an unpaired 2-tailed Student t test, and categorical data were analyzed using Fisher exact test in GraphPad Prism version 9.0.0. P values are reported for tests of statistical significance. All comparisons were 2-tailed and the α error was set at 5%.Demographic characteristics (N, %) Gender.4496Male28 (41.2%)60 (48.0%)Female40 (58.8%)65 (52.0%) Age in years, mean (SD)48.47 (21.74)53.28 (21.33).1389Clinical characteristics (N, %) Biopsy performed45 (66.2%)58 (46.4%).0103†Statistically significant result, with P value of < .05. Specialty requesting the consultEmergency department24 (35.3%)77 (61.6%).0005†Statistically significant result, with P value of < .05.Burn surgery39 (57.4%)31 (24.8%)<.0001†Statistically significant result, with P value of < .05.Internal medicine5 (7.4%)17 (13.6%).2400 Mucosal involvement66 (97.1%)60 (48.0%)<.0001†Statistically significant result, with P value of < .05.Ocular mucosa39 (57.4%)9 (7.2%)<.0001†Statistically significant result, with P value of < .05.Oral mucosa61 (89.7%)49 (39.2%)<.0001†Statistically significant result, with P value of < .05.Genital mucosa54 (79.4%)35 (28.0%)<.0001†Statistically significant result, with P value of < .05. Nikolsky sign (+)‡Nikolsky (+) sign was recorded at initial dermatology consultation by either explicit documentation of Nikolsky status, or by evidence of sloughing, desquamation, erosions, or full-thickness necrosis at areas adjacent to lesions and at seemingly unaffected skin.65 (95.6%)51 (40.8%)<.0001†Statistically significant result, with P value of < .05. Atypical target§Atypical target defined as a 2-zone targetoid lesion with poorly defined borders or a dusky macule with a dusky center.254 (79.4%)21 (16.8%)<.0001†Statistically significant result, with P value of < .05. Lymphopenia10 (14.7%)10 (8.0%).2151WBC count,‖Normal range: 3.90-10.70 × 109 cells/L. mean (SD)7.40 (3.47)11.55 (8.53).0002†Statistically significant result, with P value of < .05. Skin pain48 (70.6%)33 (26.4%)<.0001†Statistically significant result, with P value of < .05. Skin itch33 (48.5%)73 (58.4%).2261 Vesicles/bullae55 (80.9%)78 (62.4%).0091†Statistically significant result, with P value of < .05. Fever >100.4 °F¶Fever was documented positive if the patient had an outside value >100.4 °F before transfer or if their first value on arrival at the ED was >100.4 °F.35 (51.5%)18 (14.4%)<.0001†Statistically significant result, with P value of < .05. Body surface area >10%24 (35.3%)18 (14.4%).0016†Statistically significant result, with P value of < .05. Transfer distance in miles, mean (SD)56.74 (78.21)76.62 (236)<.0001†Statistically significant result, with P value of < .05. Length of stay in hospital in days, mean (SD)#Length of stay in hospital in days was counted from the date of admission at Parkland Memorial Hospital and did not include the time spent at outside facilities.15.40 (12.87)10.32 (29.9).1839 Most common alternative diagnosesFixed drug eruption−18 (14.4%)AGEP−10 (8.0%)GBFDE−10 (8.0%)Morbilliform−9 (7.2%)Bullous pemphigoid−9 (7.2%)Erythema multiforme−7 (5.6%)DRESS−7 (5.6%)AGEP, Acute generalized exanthematous pustulosis; DRESS, drug reaction with eosinophilia and systemic symptoms; GBFDE, generalized bullous fixed drug eruption; N, number; SD, standard deviation; SJS/TEN, Stevens-Johnson syndrome/toxic epidermal necrolysis; WBC, white blood cell.∗ Continuous data were analyzed using an unpaired 2-tailed Student t test, and categorical data were analyzed using Fisher exact test in GraphPad Prism version 9.0.0. P values are reported for tests of statistical significance. All comparisons were 2-tailed and the α error was set at 5%.† Statistically significant result, with P value of < .05.‡ Nikolsky (+) sign was recorded at initial dermatology consultation by either explicit documentation of Nikolsky status, or by evidence of sloughing, desquamation, erosions, or full-thickness necrosis at areas adjacent to lesions and at seemingly unaffected skin.§ Atypical target defined as a 2-zone targetoid lesion with poorly defined borders or a dusky macule with a dusky center.2Weinkle A. Pettit C. Jani A. et al.Distinguishing Stevens-Johnson syndrome/toxic epidermal necrolysis from clinical mimickers during inpatient dermatologic consultation—a retrospective chart review.J Am Acad Dermatol. 2019; 81: 749-757Abstract Full Text Full Text PDF PubMed Scopus (13) Google Scholar‖ Normal range: 3.90-10.70 × 109 cells/L.¶ Fever was documented positive if the patient had an outside value >100.4 °F before transfer or if their first value on arrival at the ED was >100.4 °F.# Length of stay in hospital in days was counted from the date of admission at Parkland Memorial Hospital and did not include the time spent at outside facilities. Open table in a new tab AGEP, Acute generalized exanthematous pustulosis; DRESS, drug reaction with eosinophilia and systemic symptoms; GBFDE, generalized bullous fixed drug eruption; N, number; SD, standard deviation; SJS/TEN, Stevens-Johnson syndrome/toxic epidermal necrolysis; WBC, white blood cell. At this time, there are no standardized guidelines for determining the likelihood of a true diagnosis of SJS/TEN. A survey of burn units demonstrated that 74% of patients are admitted based only upon clinical suspicion for SJS/TEN, with 61% having their own protocol in place at the time of admission.4Richard E.B. Hamer D. Musso M.W. Short T. O'Neal Jr., H.R. Variability in management of patients With SJS/TEN: a survey of burn unit directors.J Burn Care Res. 2018; 39: 585-592Crossref PubMed Scopus (11) Google Scholar The misdiagnosis of more benign conditions, such as urticaria, contact dermatitis, or morbilliform eruptions, leads to unnecessary transfers and exorbitant costs.5Georgesen C. Karim S.A. Liu R. Moorhead A. Falo L.D. English J.C. Response: “Distinguishing Stevens-Johnson syndrome/toxic epidermal necrolysis from clinical mimickers during inpatient dermatologic consultation—a retrospective chart review”.J Am Acad Dermatol. 2020; 82: e111-e112Abstract Full Text Full Text PDF PubMed Scopus (5) Google Scholar Our data corroborate with previously reported differences and support the need for a predictive model to aid in distinguishing SJS/TEN.2Weinkle A. Pettit C. Jani A. et al.Distinguishing Stevens-Johnson syndrome/toxic epidermal necrolysis from clinical mimickers during inpatient dermatologic consultation—a retrospective chart review.J Am Acad Dermatol. 2019; 81: 749-757Abstract Full Text Full Text PDF PubMed Scopus (13) Google Scholar We demonstrated that the predictive model that includes mucosal involvement rather than lymphopenia has improved the discrimination for SJS/TEN. Although our study externally validates both models, there are limitations to their use in clinical practice. For example, the models were developed based on physical examination features identified by experienced dermatologists. Non-dermatologists are less skilled in identifying morphologic features such as atypical targets, which are essential for this predictive model. Future studies to evaluate the performance of these models when applied prospectively by non-dermatologists are needed. None disclosed.