CHD Risk Minimization through Lifestyle Control: Machine Learning Gateway

Abstract

Studies on the influence of a modern lifestyle in abetting Coronary Heart Diseases (CHD) have mostly focused on deterrent health factors, likesmoking, alcohol intake, cheese consumption and average systolic blood pressure, largely disregarding the impact of a healthy lifestyle in mitigating CHD risk. In this study, 30+ years' World Health Organization (WHO) data have been analyzed, using a wide array of advanced Machine Learning techniques, to quantify how regulated reliance on positive health indicators, e.g. fruits/vegetables, cereals can offset CHD risk factors over a period of time. Our research ranks the impact of the negative outliers on CHD and then quantifies the impact of the positive health factors in mitigating the negative risk-factors. Our research outcomes, presented through simple mathematical equations, outline the best CHD prevention strategy using lifestyle control only. We show that a 20% increase in the intake of fruit/vegetable leads to 3-6% decrease in SBP; or, a 10% increase in cereal intake lowers SBP by 3%; a simultaneous increase of 10% in fruit-vegetable can further offset the effects of SBP by 6%. Our analysis establishes gender independence of lifestyle on CHD, refuting long held assumptions and unqualified beliefs. We show that CHD risk can be lowered with incremental changes in lifestyle and diet, e.g. fruit-vegetable intake ameliorating effects of alcohol-smoking-fatty food. Our multivariate data model also estimates functional relationships amongst lifestyle factors that can potentially redefine the diagnostics of Framingham score-based CHD-prediction.Globally, cardiovascular diseases account for nearly 17.9 million deaths with Coronary Heart Disease (CHD) accounting for 80% of these 1 . A myriad of factors have been identified as risk generators, including ethnicity, sex, total cholesterol level, triglycerides, blood pressure, that in turn are affected by life style denominators 2,3 . Together, they determine the risk appraisal function that have been assessed using conventional predictive scoring like body mass index (BMI) and Framingham scores 4,5 together with more advanced population biology or epidemiological estimators. Although there have been numerous advances in the treatment of established CHD, at a population level, assessed through Artificial Intelligence (Machine Learning) based adaptation of established statistical wisdom, remains a major knowledge gap 5 .Ground breaking epidemiological studies have identified key lifestyle and health indicators as risk factors for Coronary Heart Disease (CHD) 6-9 . Lifestyle factors include smoking 10,11 , alcohol consumption 12 , lack of physical activity while key health indicators include obesity, high blood pressure 5 and diabetes 13 . Evidence suggests a diet rich in fruits, vegetables and whole grains can mitigate the onset of CHD 8 . Some of these risk factors are not individually causative but when combined with other risk factors, increase the risk of CHD 10-14 . Further, lifestyle factors can be suitably modified to ameli...

Social Determinants in Machine Learning Cardiovascular Disease Prediction Models: A Systematic Review

“…A total of 3 studies used unsupervised machine learning algorithms such as clustering to group CVD risk levels or principal component analysis to extract features before supervised machine learning classification. 14,32,33 Median sample size was 2,510; more than two thirds of the studies had sample size >1,000, and 13 of those had >10,000. There was a sample size <100 in 5 studies, which mostly used a Bayesian Network method.…”

“…11 Given the foundation of research undergirding the critical importance of SDH as a driver of differential disease risk, it is clear that modeling methods that incorporate such factors, including capturing the interaction and relative influence of such factors in relation to other physiologic CVD risk factors, are needed. 14 Artificial intelligence and machine learning (an application of artificial intelligence for detecting patterns from data) 15 tools are seeing rapid adoption in clinical research, particularly given the proliferation of electronic health records and advanced computing strategies. These approaches have been shown to improve the prediction of CVD risk, incidence, and outcomes 16−18 over traditional risk scores such as those from the American College of Cardiology or the American Heart Association.…”

Machine Learning for Integrating Social Determinants in Cardiovascular Disease Prediction Models: A Systematic Review

“…Three studies used unsupervised machine learning algorithms, such as clustering to group CVD risk levels or principal component analysis (PCA) to extract features prior to supervised machine learning classification. 14,43,44 The most frequently used algorithms are described in Table 1. Of the 35 studies using neural networks, 12 used one hidden layer, 23 used multiple hidden layers, including most commonly three-layer perceptron, convolutional neural network and recurrent neural network.…”

“…13 Given the critical importance of social determinants with respect to disease risk, it is clear that better capturing the interaction and relative influence of such factors in relation to traditional CVD risk factors of hypertension, diabetes and hyperlipidemia provides the most significant opportunity to reduce CVD burden. 5,11,12,14…”

“…13 Given the critical importance of social determinants with respect to disease risk, it is clear that better capturing the interaction and relative influence of such factors in relation to traditional CVD risk factors of hypertension, diabetes and hyperlipidemia provides the most significant opportunity to reduce CVD burden. 5,11,12,14 Meanwhile, artificial intelligence (AI) and machine learning (an application of AI for detecting patterns from data) 15 tools have started to be adopted in clinical research, prompted by recent progress in advanced computing strategies as well as the proliferation of electronic medical record databases. 16 Machine learning methods have demonstrated improvement across multiple metrics for prediction of CVD risk, incidence and outcomes [17][18][19] over traditional risk scores such as those from the American College of Cardiology or American Heart Association.…”