Using random forest models to predict drug reaction with eosinophilia and systemic symptoms development from CBC parameters

Summary: Abstract Body: Drug reaction with eosinophilia and systemic symptoms (DRESS) is a severe adverse drug reaction that is often clinically indistinguishable from simple morbilliform drug eruptions at rash onset. We implemented Random Forest models to classify patients as DRESS-positive or DRESS-negative using CBC + differential metrics, monocyte-to-lymphocyte ratio (MLR), neutrophil-to-monocyte ratio (NMR), neutrophil-to-lymphocyte ratio (NLR), and pan-immune inflammation value (PIV: [neutrophil * platelet * monocyte]/ lymphocyte), collected within 10 days of rash onset. Models with varying tree depths (no maximum depth, 3, 5, and 7) were trained on a dataset of 261 patients, with a 75:25 train-test split. The performance of each model was compared to identify the optimal depth for accurate DRESS prediction. The no-depth model achieved the highest accuracy (90.5%) but showed signs of overfitting, with precision and recall scores of 92% for the DRESS-positive group. The 3-depth model underfit, achieving an accuracy of 71.4% and F1 scores of 0.75 and 0.67 for DRESS-positive and -negative groups, respectively. The 5-depth model had better balanced performance, achieving 81% accuracy and F1 scores of 0.82 and 0.80. The 7-depth model demonstrated the highest balance of metrics, with an accuracy of 85.7%, precision of 91%, recall of 83% for DRESS-positive predictions, and F1 scores of 0.87 and 0.84. These results demonstrate the potential of random forest models for early DRESS diagnosis. The 7-depth model, in particular, showed strong diagnostic capabilities, suggesting that routine CBC parameters could support accurate DRESS prediction at rash onset, prior to the development of additional clinical indicators. Benjamin Shwartzman¹, Grace Rabinowitz¹, Austin Piontkowski¹, Celina Dubin¹, Nancy Wei¹, Daniela Mikhaylov¹, Danielle Dubin¹, Emma Guttman-Yassky¹, Benjamin Ungar¹, Nicholas Gulati¹ 1. Kimberly and Eric J. Waldman Department of Dermatology, Icahn School of Medicine at Mount Sinai, New York, NY, United States. Bioinformatics, Computational Biology, and Imaging