Machine learning triumphs in differentiating fungal and bacterial infections

Patients with multiple myeloma (MM) are immunocompromised and highly susceptible to severe infections. Invasive fungal disease (IFD) carries a particularly high mortality rate, yet its presentation often mimics bacterial infection, leading to misdiagnosis, inappropriate antibiotic use, and delayed antifungal therapy. This case-control study aimed to develop and validate a machine learning-based model to accurately distinguish IFD from bacterial infection in MM patients. In this retrospective case-control study, we analyzed epidemiological, clinical, and laboratory data from 140 MM patients with IFD (cases) and 158 MM patients with bacterial infections (controls). 16 key variables were used to train nine machine learning models (including Random Forest and XGBoost). The dataset was split into a training cohort (70%) and a test cohort (30%) for performance evaluation. The findings demonstrate that the area under the receiver operating characteristic curve (AUC) values for the nine models varied between 0.860 and 0.967. Notably, the logistic regression model exhibited superior performance, achieving an AUC of 0.967, an accuracy of 0.918, a recall of 0.908, and a precision of 0.919. Bootstrap analysis using 500 stratified bootstrap samples revealed that the model’s performance metrics had a standard deviation of less than 0.04 and narrow 95% confidence intervals, with the AUC interval width at 0.021, indicating consistent performance and high accuracy. Additionally, SHapley Additive exPlanation (SHAP) analysis enhanced the model’s interpretability by elucidating the contribution of each predictor. We developed and validated an accurate machine learning-based model that effectively distinguishes IFD from bacterial infections in MM patients. The resulting SHAP analysis provides clinicians with a practical tool for the early identification of high-risk IFD, potentially guiding timely and appropriate treatment decisions.