Supplemental materials for publication “Distinguishing Stevens-Johnson syndrome/toxic epidermal necrolysis from clinical mimickers during inpatient dermatologic consultation – a retrospective chart review.” JAAD. 2019.

Statistical analysis here replicates that performed in the primary manuscript with addition of mucosal involvement (either present or absent) as reported in Supplemental Table 1. Stepwise regression of all significant variables including mucosal involvement yielded an updated model with four predictor variables. A new multivariable model was fit with Nikolsky sign (adjusted OR 49.0, 95% CI 10.6-226.2, p-value <0.001); atypical targets (adjusted OR 26.5, 95% CI 5.5-127.3, p<0.001); fever (adjusted OR 7.6, 95% CI 1.8-33.0, p-value=0.007); mucosal involvement (adjusted OR 21.3, 95% CI 4.1-110.7, p-value <0.001); OR (odds ratio); CI (confidence interval). This new multivariable model for the probability of Stevens-Johnson syndrome/toxic epidermal necrolysis (SJS/TEN) is defined as indicated in Supplemental Figure 1. This updated model yields a 91.5% sensitivity and 94.6% specificity with a corresponding positive and negative predictive values of 87.1% and 96.6%, respectively. The AUC for the final model is 0.98 (95% confidence interval 0.96, >0.99 and p-value<0.001). Supplemental Figure 2 is a nomogram which provides the relative contributions of each predictor, including mucosal involvement, to the probability of SJS/TEN.

本研究的统计学分析复刻了正文手稿中的分析流程，并补充纳入了补充表1中所报告的黏膜受累情况（分为存在或不存在两类）。对包括黏膜受累在内的所有有统计学意义的变量进行逐步回归分析后，得到了包含4个预测变量的更新模型。本研究构建的新多变量模型纳入了以下预测因子：尼氏征（调整后比值比（OR, odds ratio）为49.0，95%置信区间（CI, confidence interval）为10.6~226.2，p值<0.001）、非典型靶皮损（调整后OR为26.5，95% CI为5.5~127.3，p值<0.001）、发热（调整后OR为7.6，95% CI为1.8~33.0，p值=0.007）以及黏膜受累（调整后OR为21.3，95% CI为4.1~110.7，p值<0.001）；其中OR为比值比，CI为置信区间。这款用于预测史蒂文斯-约翰逊综合征/中毒性表皮坏死松解症（SJS/TEN, Stevens-Johnson syndrome/toxic epidermal necrolysis）发病概率的多变量模型，其具体定义详见补充图1。该更新模型的敏感度为91.5%，特异度为94.6%，对应的阳性预测值与阴性预测值分别为87.1%与96.6%。最终模型的受试者工作特征曲线下面积（AUC, area under the receiver operating characteristic curve）为0.98（95%置信区间0.96~>0.99，p值<0.001）。补充图2为列线图，用于展示各预测因子（包括黏膜受累）对SJS/TEN发病概率的相对贡献度。