Development and Validation of a Machine Learning-Based Nomogram for Predicting Severe Pneumocystis jirovecii Pneumonia in Kidney Transplant Recipients

BACKGROUND Pneumocystis jirovecii pneumonia (PJP) is a life-threatening opportunistic infection in kidney transplant recipients (KTRs). Early identification of patients liable to progress to severe disease is critical for improving prognosis. This study aimed to construct and validate a machine learning-based nomogram for predicting the risk of severe PJP in KTRs using routine clinical indicators. MATERIAL AND METHODS A retrospective cohort of 169 KTRs diagnosed with PJP was analyzed. Severe PJP was defined as cases requiring intensive care unit (ICU) admission or death. The cohort was randomized into training (n=120) and testing (n=49) sets. Three machine learning algorithms (Boruta, RFE, and LASSO) were utilized for feature selection. A multivariate logistic regression model was established and visualized as a nomogram. Model performance was evaluated via area under the ROC curve (AUC), calibration plots, and decision curve analysis (DCA). Kaplan-Meier analysis was performed to assess risk stratification. RESULTS Four key predictors were identified: procalcitonin (PCT), (1→3)-ß-D-glucan (G_test), C-reactive protein (CRP), and the time from kidney transplantation to PJP onset (Time KT to PJP). Notably, shorter post-transplant time and elevated biomarkers were associated with greater severity. The nomogram demonstrated robust discrimination with AUCs of 0.935 (training) and 0.886 (testing), alongside excellent calibration. DCA confirmed a significant clinical net benefit. Furthermore, Kaplan-Meier analysis revealed that patients stratified as high-risk by the model had significantly lower survival rates compared to the low-risk group (P