Machine Learning–Based Risk Stratification for Occupational Leptospirosis Surveillance within a One Health Framework

Introduction:Leptospirosis is an important zoonotic disease in Iran and a persistent occupational health concern for high-risk groups in warm and humid regions. Nonspecific early symptoms and diagnostic delays hinder timely prevention. Data-driven risk stratification may support occupational health programs through targeted surveillance and preventive interventions. This study aimed to develop and internally validate a machine learning-based model for occupational leptospirosis risk stratification within a One Health surveillance perspective.

Methods:A retrospective dataset of 796 individual occupational and clinical records was compiled from occupational health surveillance reports and individual-level datasets described in peer-reviewed epidemiological studies conducted in northern, southern, and western provinces of Iran between 2013 and 2025. The outcome was defined as laboratory-confirmed or clinically diagnosed leptospirosis, reflecting routine surveillance practice. Predictors included occupation type, exposure to animal urine or stagnant water, use of personal protective equipment, early clinical symptoms including fever, headache, and myalgia, age, sex, and underlying conditions. Class imbalance was addressed using Synthetic Minority Over-sampling Technique and Random Over-Sampling within training folds. Logistic Regression and XGBoost models were trained using k-fold cross-validation and evaluated using the area under the receiver operating characteristic curve, sensitivity, accuracy, and Brier score.

Results:The XGBoost model with Random Over-Sampling achieved the best performance, with a sensitivity of 82 %, an area under the curve of 0.84, an accuracy of 78 %, and a Brier score of 0.16. The most influential predictors were lack of personal protective equipment use, direct contact with animal urine, and fever within the first five days of symptom onset. Risk stratification classified 14 % of records as high risk and 4 % as very high risk.

Conclusions:This study shows that integrating occupational exposure and early clinical data enables machine learning risk stratification for occupational leptospirosis surveillance within a One Health framework, requiring external validation before implementation.