Background and&#xa0;Objectives <p>SCORe of Toxic Epidermal Necrolysis (SCORTEN) and ABCD-10 have been developed as scoring systems for predicting mortality associated with Stevens–Johnson syndrome (SJS) or toxic epidermal necrolysis (TEN). These scores were developed based on a small number of patients; hence, their generalizability requires further exploration. The present study used three algorithms, including a machine learning method, to construct a mortality prediction model for SJS/TEN and to identify new candidate predictors of mortality from severe drug eruptions.</p> Methods <p>Data from 5966 patients with SJS or TEN were extracted from the Japanese Adverse Drug Event Report Database. A mortality prediction model was then constructed using stepwise regression, L1 regularized-logistic regression, and random forests based on the patient characteristics (e.g., age, sex, primary disease, adverse events, drug classification, route of administration) and outcomes (death).</p> Results and Discussion <p>The mortality prediction models for SJS/TEN identified sex (men), primary disease (hyperlipidemia, diabetes mellitus, renal dysfunction, and malignant tumors), adverse events (renal dysfunction, liver dysfunction, respiratory dysfunction, bacteremia/sepsis, disseminated intravascular coagulation syndrome, shock, and multiple organ failure), number of concomitant drugs, and route of administration (injection) as common factors associated with mortality.</p> Conclusions <p>Our findings showed that sex, hyperlipidemia as the primary disease, number of concomitant drugs, use of antipyretic analgesics, and route of administration may be considered as predictors of mortality in patients with SJS/TEN. The external validity of these factors needs to be examined in the future.</p>