Abstract
Introduction Leptospirosis is a zoonotic infection caused by Leptospira bacteria, which is reemerging in various regions and often poses a diagnostic challenge due to its nonspecific symptoms. While most infections are mild, severe cases occur in 5-10% of patients and are associated with high mortality, especially in areas with poor sanitation and urbanization. This study aims to investigate the association of specific parameters with leptospirosis diagnosis using a machine learning model and geographic mapping tools to identify spatial patterns and high-risk areas for the disease. Methods An observational retrospective study conducted at a tertiary care center analyzed patients clinically suspected of leptospirosis over the course of one year. The study utilized laboratory investigations, geographic mapping, and machine learning models to explore the association between various laboratory parameters and the predictive diagnosis of leptospirosis. Results The study, conducted over one year at All India Institute of Medical Sciences, Kalyani, India, included 325 patients, of whom 43 (13.2%) tested positive for leptospirosis by IgM ELISA. Geographic mapping revealed case clusters around nearby districts of West Bengal, with a few cases from Tripura and Bangladesh. The study found no significant association between individual laboratory parameters and leptospirosis diagnosis. However, machine learning models, particularly k-nearest neighbors (KNN), demonstrated moderate predictive accuracy (accuracy: 74%, area under the curve: 0.6). Conclusion Geographic mapping identified clusters of leptospirosis cases; however, no significant association was found between individual laboratory parameters and the disease diagnosis. Machine learning models, particularly KNN, demonstrated moderate predictive accuracy. The study also highlighted the overlapping clinical features of leptospirosis, dengue, and scrub typhus in West Bengal, although it noted the absence of detailed clinical data as a limitation.