Immune dysregulation in Pneumocystis jirovecii infection and the development of a machine learning-based diagnostic model.

BACKGROUND: Pneumocystis jirovecii pneumonia (PJP) is a life-threatening opportunistic infection in immunocompromised patients. A major clinical challenge is the inability of current microbiological diagnostics (e.g., quantitative polymerase chain reaction [qPCR]) to reliably distinguish PJP from Pneumocystis jirovecii colonization(P. jirovecii colonization). OBJECTIVE: This study aimed to develop and validate a diagnostic model for differentiating PJP from P. jirovecii colonization by analyzing 17 core model input predictors that integrate pathogen burden and host immune response profiles. METHODS: We conducted a retrospective analysis of 127 clinically assessed adult patients (78 with PJP, 49 with P. jirovecii colonization). 17 core model input predictors (pathogen burden, systemic inflammatory markers, and peripheral blood cellular immune parameters) were compared between groups. Subsequently, we constructed a random forest machine learning model using these 17 core model input predictors. Model development employed a two-stage grid search strategy for hyperparameter optimization, with strict prevention of overfitting ensured through cross-validation. The final model's performance was evaluated on an independent internal validation set. RESULTS: Compared to the P. jirovecii colonization group, the PJP group exhibited significantly higher pathogen burden (median Ct 31.40 vs. 34.72, P<0.001), heightened inflammatory responses (e.g., elevated IL-10 and IL-6), and a characteristic immunosuppressed state marked by severe lymphopenia, with pronounced reductions in CD4+ T cells and regulatory T cells (Tregs). Based on these findings, a Random Forest algorithm was used to construct a discriminative model incorporating the 17 core model input predictors. In the validation cohort (N = 37), the model demonstrated robust discriminatory performance, with an area under the ROC curve (AUC) of 0.852 (95% CI: 0.709-0.996) and an accuracy of 78.38%. Feature importance analysis identified three top core model input predictors: pathogen load, IL-10 level, and absolute lymphocyte count as the most discriminative predictors. This study confirms that a multi-parameter strategy integrating the 17 core model input predictors can effectively overcome the limitations of current single-marker microbiological tests. The developed machine learning model provides a novel, translationally promising tool for the precise clinical differentiation of PJP from P. jirovecii colonization.