基于机器学习的高血压发病影响因素的巢式病例对照研究

A Nested Case-Control Study on Machine Learning-Based Risk Factors for Hypertension

摘要: 目的：研究高血压发病的影响因素，并为高血压的防治提供科学依据。方法：基于中国健康与养老追踪调查(China Health and Retirement Longitudinal Study, CHARLS)数据库，以2011~2018年中4次随访中新发的1946例高血压患者为新发病例组，按照年龄(±2岁)和性别1:1个体匹配的方法，选取同期未发生高血压患者为对照组，最终纳入研究对象3668例。采用条件Logistic回归模型、限制性立方样条模型及梯度提升机模型(Gradient Boosting Machine Mode, GBM)探讨高血压发病的相关风险因素。结果：多因素条件Logistic回归模型显示甘油三脂(Triglycerides, TG)、C-反应蛋白(C-reactive protein, CRP)、高密度脂蛋白胆固醇(High-Density Lipoprotein Cholesterol, HDL-C)及腰围与高血压存在关联，其OR值分别是1.01 (95% CI: 1.00~1.02)、1.01 (95% CI: 1.00~1.03)、1.0 (95% CI: 1.00~1.01)、1.03 (5% CI: 1.02~1.03)。限制性立方样条模型结果显示，尿酸(Uric Acid, UA)水平与高血压发病之间存在正向非线性剂量–反应关系(UA：P总趋势 = 0.012，P非线 = 0.026)，GBM模型中根据SHAP (SHapley Additive exPlanations)值显示，胱抑素C对高血压的发病具有较大的正向影响。结论：本研究发现腰围、TG、HDL-C、UA、胱抑素C、CRP是高血压发病的相关风险因素，同时发现UA水平与高血压发病之间存在正向非线性剂量–反应关系。

Abstract: Objective: To study the influencing factors on the development of hypertension and provide a scientific basis for the prevention and treatment of hypertension. Methods: Based on the China Health and Retirement Longitudinal Study (CHARLS) database, 1946 patients with new-onset hypertension during the 4 follow-ups between 2011 and 2018 as the case group, and selecting a control group of subjects who did not develop hypertension during the same period using a 1:1 individual matching method based on age (±2 years) and gender, a total of 3668 subjects were included in the study. Conditional logistic regression models, restricted cubic spline models, and gradient boosting machines (GBM) models were employed to explore the influencing factors of hypertension onset. Results: The multifactorial conditional logistic regression model showed that triglyceride (TG), C-reactive protein (CRP), high-density lipoprotein cholesterol (HDL-C), and waist circumference were associated with hypertension, with OR values of 1.01 (95% CI: 1.00~1.02), 1.01 (95% CI: 1.00~1.03), and 1.01 (95% CI: 1.00~1.01), 1.03 (95% CI: 1.02~1.03). Restricted cubic spline modeling results showed a positive nonlinear dose-response relationship between uric acid (UA) levels and the development of hypertension (UA: P total trend = 0.012, P nonlinear = 0.026), and in GBM modeling based on SHAP values showed a large positive effect of cystatin C on the development of hypertension. Conclusion: This study found that waist circumference, TG, HDL-C, UA, cystatin C, and CRP are influencing factors for the onset of hypertension. Additionally, a positive non-linear dose-response relationship was observed between UA levels and the risk of hypertension development.