Machine learning-driven risk assessment of coronary heart disease: Analysis of NHANES data from 1999 to 2018

The high incidence of coronary artery heart disease (CHD) poses a significant burden and challenge to public health systems globally. Effective prevention and early diagnosis of CHD have become key strategies to alleviate this burden. This study aims to explore the application of advanced machine learning techniques to enhance the accuracy of early screening and risk assessment for CHD.

A total of 49 490 study subjects from the National Health and Nutrition Examination Survey (NHANES) database spanning from 1999 to 2018 were included. The dataset was randomly divided into training (70%) and testing (30%) sets. The dependent variable (outcome variable) was whether the subjects were informed of a CHD diagnosis, categorizing them into a CHD group and a non-CHD group. We reviewed the literature on risk factors associated with CHD, ultimately including 68 independent variables. The variable characteristics of the study subjects were analyzed, comparing differences between the CHD and non-CHD groups. Machine learning algorithms, specifically random forest (randomForest_4.7-1.1) and XGBoost (xgboost_1.7.7.1) were utilized for variable selection. A comprehensive analysis of the top 10 variables identified by these 2 algorithms were conducted, selecting those mutually recognized by both. A generalized linear model was used to analyze the relationships between variables and CHD, and classical logistic regression was used to construct the CHD risk prediction model. The model's ability to distinguish between CHD and non-CHD individuals was assessed using the area under the receiver operating characteristic curve (AUC); calibration measurements were conducted with the Hosmer-Lemeshow goodness-of-fit test to evaluate the consistency between predicted values and actual CHD proportions; and decision curve analysis was applied to evaluate the clinical benefits of the model's risk prediction. Finally, a nomogram was constructed to visually present the risk scoring of the final model.

The mean age of the overall population was (49.53±18.31) years, with males comprising 51.8%. Compared to the non-CHD group, the CHD group was older [(69.05± 11.32) years vs (48.67±18.07) years, P<0.001], had a higher proportion of females (67.1% vs 47.4%, P<0.001), and exhibited statistically significant differences in classical cardiovascular risk factors such as body mass index, systolic blood pressure, diastolic blood pressure, and smoking (all P<0.001). Additionally, there were statistically significant differences in non-classical cardiovascular factors, such as energy intake, vitamins E, vitamin K, calcium, phosphorus, magnesium, zinc, copper, sodium, potassium, and selenium (all P<0.05). Six key variables most associated with CHD occurrence were ultimately identified. The CHD risk prediction model constructed was as follows: logit(p)= -7.783+0.074×age+0.003×creatinine-0.003×platelets+0.257×glycated hemoglobin+0.003× uric acid+0.101×coefficient of variation of red cell distribution width. The model demonstrated excellent discriminative ability in predicting CHD, with an accuracy of 0.712 and an AUC of 0.841. Calibration curves indicated good consistency between predicted probabilities and actual values in both the training and testing sets, demonstrating model stability and reliability. Decision curve analysis suggested that the model provided net benefits across a range of threshold probabilities, supporting its potential application in clinical decision-making.

This study successfully identified potential risk factors for CHD using machine learning techniques and developed a concise and practical clinical prediction model. Further prospective clinical cohort studies are needed to validate its potential for clinical application, enabling effective cardiovascular disease prevention and intervention strategies in real-world healthcare settings.