Environmental Correlates of Opisthorchis viverrini Infection-Free Zones: A Comparative Machine Learning Approach Using Satellite-Derived Indices

Opisthorchis viverrini (OV) infection is still a major problem in Northeast Thailand. In this study, we looked at locations where no OV was detected (0% prevalence) by analyzing 519 georeferenced points alongside 10 environmental indices from satellite data. We found that 457 of these points, or about 88.1%, showed zero prevalence. When we looked at individual factors, the Enhanced Vegetation Index (EVI) tended to be higher and the Standardized Precipitation Index (SPI6) tended to be lower in these zero-prevalence areas. However, once we applied the Bonferroni correction, these differences were not statistically significant. Using Principal Component Analysis, we identified four components that explained 88.04% of the variance. Among the machine learning models we tested, Logistic Regression performed best with a Balanced Accuracy of 0.6053. When we used SHAP analysis based on a Random Forest model, EVI and SPI6 stood out as having the biggest impact. We also found significant spatial clustering of these zero-prevalence locations through spatial autocorrelation. Overall, these results point toward certain environmental factors linked to zero OV prevalence, specifically EVI and SPI6. That said, because the results did not hold up after multiple testing corrections, they should be treated as exploratory rather than definitive. This study offers a way to combine satellite data with spatial analysis and machine learning, but it also shows that we really need larger, more balanced datasets to get clearer answers in the future.