Supplementary_Material_JAAD-D-25-04018

1.Questionnaire assessing patient expectations of treatment outcomes. 2.Methods. 3.Graphical abstract. 4.Raw data. 5.Code. Table S1. Baseline characteristics of PWS patients. Table S2. Hyperparameter tuning details for machine learning models. Table S3. Multivariate logistic regression quantifying mediation effects. Figure S1. Feature selection and model performance comparison: LASSO, RFE, and random forest analyses. Figure S2. Confusion matrices of five models at the optimal threshold (0.515). Figure S3. Performance evaluation of five models: ROC, confusion matrix, calibration, and decision curves. Figure S4. Feature importance and SHAP analysis of the LightGBM model. Figure S5. SHAP summary and dependence plots of the LightGBM model. Figure S6. Exploratory analysis of associations among clinical and dermoscopic variables. Figure S7. Surrogate decision tree of the LightGBM model1. 评估患者对治疗结局预期的调查问卷 2. 研究方法 3. 图形摘要 4. 原始数据 5. 代码 表S1. 普拉德-威利综合征（Prader-Willi Syndrome, PWS）患者的基线特征 表S2. 机器学习模型的超参数调优细节 表S3. 量化中介效应的多元Logistic回归分析 图S1. 特征选择与模型性能对比：最小绝对收缩和选择算子（Least Absolute Shrinkage and Selection Operator, LASSO）、递归特征消除（Recursive Feature Elimination, RFE）及随机森林分析 图S2. 5个模型在最优阈值（0.515）下的混淆矩阵 图S3. 5个模型的性能评估：ROC曲线、混淆矩阵、校准曲线与决策曲线 图S4. LightGBM模型的特征重要性与SHAP分析 图S5. LightGBM模型的SHAP汇总图与依赖图 图S6. 临床变量与皮肤镜变量间关联的探索性分析 图S7. LightGBM模型的代理决策树