Pelvic inflammatory disease prevalence and dietary phosphorus: A cross-sectional analysis of the National Health and Nutrition Examination Survey, 2015-2018.

OBJECTIVE

Emerging evidence suggests dietary components may modulate inflammatory conditions, yet the role of phosphorus in pelvic inflammatory disease (PID) remains unclear. This study investigated the association between dietary phosphorus intake and PID prevalence while developing machine learning (ML) models for PID risk prediction.

METHODS

A total of 2629 non-pregnant women from the National Health and Nutrition Examination Survey 2015-2018 were analyzed. PID was defined using self-reported physician diagnosis. Dietary phosphorus intake was quantified using two 24-h recalls. Multivariable weighted logistic regression, subgroup analyses, and smooth curve fitting were employed to assess linear/nonlinear associations. Six ML algorithms-Random Forest (RF), LightGBM, Decision Tree, XGBoost, SGD Classifier, and Gradient Boosting-were trained using phosphorus intake and demographic features. Model performance was evaluated based on the accuracy, precision, recall, F1-score, and AUC. Feature importance was interpreted using Shapley Additive Explanations (SHAP) values.

RESULTS

Higher dietary phosphorus intake was inversely associated with PID prevalence. In fully adjusted models, each 1 mg/day increase in phosphorus reduced PID odds by 0.1% (adjusted odds ratio [aOR] 0.99, 95% confidence interval [CI]: 0.9988-0.9999, P = 0.017), with the highest quartile (Q4) showing 56.0% lower odds than Q1 (aOR 0.44, 95% CI: 0.221-0.875, P = 0.022). Age significantly modified this association (P = 0.003), with stronger effects observed in older adults. The RF model achieved optimal prediction performance (accuracy = 98.3%, AUC = 0.984), with phosphorus intake ranked among the top three predictive features.

CONCLUSION

These findings provide the first epidemiologic evidence that adequate dietary phosphorus intake may protect against PID, particularly in older women. The integration of ML models with nutritional data demonstrates high predictive accuracy for PID risk, paving the way for personalized dietary interventions.