The invisible stings: public healthcare access as an intrinsic source of bias in reported scorpion stings

Scorpionism is an important and growing public health concern in Brazil, with spatial variation in reported cases that may reflect environmental, socioeconomic, and healthcare-related factors. This study aimed to identify factors associated with reported scorpion stings and map their spatial distribution across São Paulo State. We analyzed 439,046 reported scorpion stings recorded across 645 municipalities from 2007 to 2025. An initial set of 118 climatic, environmental, and socioeconomic predictors was reduced to 25 non-redundant variables using variance inflation factor filtering. Five machine learning approaches were compared using log-transformed incident counts and spatial block cross-validation. A sensitivity analysis compared the selected model with a reduced model excluding public healthcare facilities​. Random Forest provided the best predictive performance ( R ² = 0.762; Pearson’s r = 0.949; RMSE = 151.55; MAE = 45.59). Public healthcare facilities were the most influential predictor (37.42%), followed by October vapor pressure (14.08%), adequate sanitation coverage (6.62%), and population density (3.36%). Excluding healthcare facilities reduced model performance ( R ² = 0.695; r = 0.925) and redistributed importance toward demographic, socioeconomic, and climatic variables. Number of healthcare facilities showed a positive association with predicted reported sting rates, while high predicted notification of scorpion stings was concentrated mainly in the central, eastern, and northeastern regions of São Paulo state. Reported scorpion sting patterns reflect the combined influence of environmental, climatic, socioeconomic, demographic, and healthcare-related factors. Number of healthcare facilities​ was the strongest predictor and substantially influenced model performance and spatial predictions. This highlights the critical need to account for this variable in spatial analyses, as its circular influence overestimates reported scorpion stings in well-served areas while leaving gaps in underserved regions, directly driving the underreporting of stings and heavily biasing ecological predictions.