Exploring the relationship between per- and polyfluoroalkyl substances exposure and rheumatoid arthritis risk using interpretable machine learning.

BACKGROUND

Rheumatoid arthritis is a chronic autoimmune disease influenced by environmental exposures, including per- and polyfluoroalkyl substances (PFAS). Although previous studies have suggested links between PFAS and rheumatoid arthritis risk, none have used interpretable machine learning models for prediction. This study aimed to develop such a model to assess risk based on PFAS exposure.

METHODS

We analyzed data from 11,705 participants in the National Health and Nutrition Examination Survey (2003-2018). Twelve machine learning algorithms were evaluated using metrics including area under the curve (AUC), accuracy, sensitivity, specificity, and F1 score. Key predictors were identified using SHapley Additive exPlanations (SHAP). Partial dependence plots and locally weighted scatterplot smoothing (LOWESS) curves were used to examine non-linear associations and exposure thresholds. A web-based risk calculator was developed to enhance clinical and public health applicability.

RESULTS

CatBoost showed the best performance (AUC: 0.82; Accuracy: 74%; F1 score: 0.62) and was selected for further interpretation. SHAP analysis identified perfluorooctane sulfonic acid (PFOS) and 2-(N-Methyl-perfluorooctane sulfonamido) acetic acid (MPAH) as major contributors to risk prediction. PFOS exhibited a U-shaped relationship with increased risk above 15.10 ng/ml, while MPAH showed a risk transition at 0.22 ng/ml. Waterfall plots illustrated the contribution of individual exposures. The interactive web-based calculator allows users to input PFAS levels and receive personalized rheumatoid arthritis risk estimates. It is freely available on Hugging Face Spaces (https://huggingface.co/spaces/Machine199710/RA_ML).

CONCLUSIONS

This study demonstrates the potential of machine learning to predict rheumatoid arthritis risk based on PFAS exposure. The identified non-linear patterns provide insights into environmental contributions to disease risk and may inform future prevention strategies.