Machine Learning Models for Prediction of Severe Pneumocystis carinii Pneumonia after Kidney Transplantation: A Single-Center Retrospective Study

Background: The objective of this study was to formulate and validate a prognostic model for postoperative severe Pneumocystis carinii pneumonia (SPCP) in kidney transplant recipients utilizing machine learning algorithms, compare the performance various models. Methods: Clinical manifestations laboratory test results upon admission were gathered as variables 88 patients who experienced PCP following transplantation. most discriminative identified, subsequently, Support Vector Machine (SVM), Logistic Regression (LR), Random Forest (RF), K-Nearest Neighbor (KNN), Light Gradient Boosting (LGBM), eXtreme (XGB) models constructed. Finally, models’ predictive capabilities assessed through ROC curves, sensitivity, specificity, accuracy, positive value (PPV), negative (NPV), F1-scores. Shapley additive explanations (SHAP) algorithm employed elucidate contributions effective model’s variables. Results: Through lasso regression, five features—hemoglobin (Hb), Procalcitonin (PCT), C-reactive protein (CRP), progressive dyspnea, Albumin (ALB)—were six developed using these after evaluating their correlation multicollinearity. In validation cohort, RF demonstrated highest AUC (0.920 (0.810–1.000), F1-Score (0.8), accuracy (0.885), sensitivity (0.818), PPV (0.667), NPV (0.913) among models, while XGB KNN exhibited specificity (0.909) Notably, CRP exerted significant influence on revealed by SHAP feature importance rankings. Conclusions: algorithms offer viable approach constructing predict development disease recipients, with potential practical applications.