Data to diagnosis: Machine learning-aided prediction of bloodstream infections

The application of machine learning (ML) algorithms to routine hematological parameters for early prediction of bloodstream infections and to characterize the microbial and antimicrobial resistance (AMR) profile of culture-positive cases is of interest. A retrospective observational study conducted at AIIMS Kalyani between January 2023 and December 2024, in which blood culture results, AST profiles and 16 routine hematological parameters were collected. ML models including Logistic Regression, Decision Tree, Random Forest and SVM were developed using Python. ROC-AUC, accuracy, sensitivity and specificity were computed for model evaluation. <i>Escherichia coli</i>, <i>Klebsiella pneumoniae</i>, <i>Staphylococcus aureus</i> and <i>Pseudomonas aeruginosa</i> were the predominant isolates, with high resistance to cephalosporins and beta-lactam-beta-lactamase inhibitor combinations. The Random Forest model showed the highest predictive power (accuracy: 87%, AUC: 0.70) and key predictors included neutrophil count, CRP, and TLC. The ML models offer promising support for early prediction of BSIs and, when coupled with continuous AMR surveillance, can facilitate rapid diagnosis and guide empirical antibiotic therapy, especially in low-resource settings.