Development and Validation of Comorbidity Index to Predict In-hospital Mortality Risk for Patients with Steven-Johnson Syndrome

Abstract Background Stevens-Johnson Syndrome (SJS)/toxic epidermal necrolysis (TEN) is a spectrum of severe drug reactions that has a high in-hospital mortality from 5% to 20%. Early risk prediction for in-hospital mortality in SJS/TEN can be informative for clinical management. This study aimed to develop an in-hospital mortality risk calculator for SJS based on patient comorbidities. Methods Patients who were diagnosed with SJS/TEN were identified in the National Inpatient Sample database between Q4 2015-2020. SJS patients were randomly sampled into two equal experimental and validation groups. The weight for each comorbidity was determined from a multivariable logistic regression to develop into a single SJS Index. Results There were 4519 SJS (mortality rate 5.05%), 873 SJS-TEN (mortality rate 13.06%), and 759 TEN (mortality rate 17.00%) patients identified. The SJS Index had good discrimination power ( c -statistic=0.704, 95% CI=0.656-0.753) and was well-calibrated (Brier score=0.049). SJS Index had similar performance in multivariable regression model with all comorbidities (DeLong p-value=0.621) and validation group (DeLong p-value=0.634). The index had significantly better discriminative power than Elixhauser Comorbidity Index (ECI) (DeLong p-value=0.001). Applying the SJS Index to SJS-TEN patients showed comparable discriminative powers (DeLong test p-value=0.284). In TEN patients, SJS Index remained good calibration (Brier score=0.062) but discriminative power decreased (DeLong p-value=0.005). After adjusting for age, race, and ethnicity, SJS Index had improved performance in all groups. Conclusions The SJS Index can effectively discriminate and predict in-hospital mortality in SJS and was externally validated in NIS. SJS/TEN often has emergent presentation and immediate intervention is required, this risk score can give an initial assessment of the mortality based on medical records alone and thus offer a quick insight for clinicians for management. In addition, this study demonstrated the possibility of the SJS Index being used as one composite metric to develop a more comprehensive risk-calculator score.