SHAP analysis of feature contributions in machine learning models predicting sMN development in AA. SHAP summary (beeswarm) plots for model 1 (A) and model 2 (C). Each point represents the SHAP value of a feature for an individual observation. Features are ranked by overall importance. The x-axis indicates the SHAP value (ie, the impact of that feature on the model’s output for a given patient), whereas color reflects the feature value (blue = low, red = high). Features to the right of x = 0 were associated with increased predicted risk of sMN, whereas features to the left were associated with reduced risk. Mean absolute SHAP value plots with hierarchical clustering of features for model 1 (B) and model 2 (D). These bar plots quantify the global importance of each feature based on the magnitude of its contribution to model predictions. Features with low mean SHAP values were considered to have minimal impact and were excluded from subsequent clustering. “Remaining 3 features” in panel D refers to “Abnormal Karyotype,” “Del(13q) Karyotype,” and “Other IST Treatment,” which were retained but contributed minimally. ATG, antithymocyte globulin.