U-shaped association between serum chloride and hypertension risk with nadir around 103 mmol/L: insights from regression and interpretable machine learning (XGBoost/SHAP) using NHANES 2017-2018.

Background: Hypertension is a major contributing factor for cardiovascular disease. This research attempted to explicate the link between serum levels of chloride and the risk of hypertension occurrence, as well as to identify the threshold level at which the risk undergoes changes across the general population and various demographic segments. Methods: Employing materials from the to 2017-2018 National Health and Nutrition Examination Survey (NHANES; n = 4,591), we employed multivariate regression analysis to gauge the connection between adult serum chloride concentrations and the risk of hypertension occurring. Smooth curve fitting, threshold effects, and saturation effects analyses were carried out to identify the threshold levels of chloride connected with changes in the risk of hypertension. Additionally, to further explore the complex relationship between serum chloride levels and hypertension risk, and to understand the contributions of various features within a high-performance machine learning model, we trained an XGBoost classifier to predict hypertension status and utilized SHAP (SHapley Additive exPlanations) values for interpretation. Results: A substantial connection was acquired between serum chloride levels and the risk of hypertension. After adjusting for variables, the multivariate logistic regression analysis demonstrated a U-shaped connection (OR = 0.94, 95% CI: 0.92-0.97, P < 0.0001). Below 103 mmol/L, the risk of hypertension decreased with increasing chloride levels (OR = 0.906, 95% CI = 0.877-0.936, P < 0.0001), demonstrating a 9.4% decline in the likelihood of hypertension per 1 mmol/L rise in chloride. Conversely, above 103 mmol/L, the risk grew with higher chloride concentrations (OR = 1.119, 95% CI = 1.030-1.216, P = 0.0081), signifying an 11.9% rise in the probability of hypertension by 1 mmol/L increment. Interpretation of an XGBoost machine learning model using SHAP values visually corroborated this U-shaped pattern, further indicating that the lowest contribution of serum chloride to the predicted risk of hypertension occurred around the 103 mmol/L level. Conclusion: In conclusion, using NHANES 2017-2018 data, this study revealed a significant U-shaped association between adult serum chloride levels and hypertension risk, with a nadir at 103 mmol/L. Both low and high chloride levels correlated with increased hypertension risk. This suggests serum chloride could serve as a potential biomarker for hypertension risk stratification, warranting further validation. Given the observational design, future prospective studies are needed to confirm this association and elucidate its underlying mechanisms.