Construction of machine learning-based diet quality index and its association with blood pressure levels and hypertension risk
Objective This paper aimed to propose a data-adaptive diet quality index based on the elastic net variable selection method in machine learning.Additionally,it explored the association of this index with blood pressure levels and the risk of hypertension.Methods The data source was from 1 143 Asian samples in the National Health and Nutrition Examination Survey(NHANES)across four cycles from 2011 to 2018,with 59.06%being male and an average age of(43.12±15.10)years.Other variables included demographic variables,intake levels of 29 food components and diastolic and systolic blood pressure measurements.Control-ling for related confounding factors,the elastic net variable selection method was used to screen important food components.The weighted average of food components was defined as the data-driven diet quality score(ddDQS),using their corresponding regression coefficients as weights.Furthermore,the association between the new index and the risk of hypertension was investigated and compared with four commonly used diet quality indices:the Healthy Eating Index 2015(HEI-2015),the Alternative Healthy Eating Index(AHEI),the Dieta-ry Approaches to Stop Hypertension(DASH)index,and the Mediterranean Dietary Index(MED).Results The prevalence of hypertension was 26.15%,and the average age of individuals with hypertension(52.94±14.77)years was significantly higher than those without hypertension(39.64±13.62)years.The ddDQS in-cluded five important dietary components:refined grains,oils,alcohol,added sugars and starchy potatoes.It was significantly associated with systolic blood pressure(SBP)(β=-2.08,95%CI=-3.24~-0.92,P<0.001)and hypertension(HTN)(β=-0.482,95%CI=-0.72~-0.25,OR=0.62,P<0.001).Intake of starchy potatoes increased the ddDQS,while intake of refined grains,oils,alcohol,and added sugars de-creased the ddDQS.Subgroup analysis results showed that individuals aged 40~60 years and males were more sensitive to the ddDQS,suggesting that these subgroups would benefit more from adhering to healthy dietary patterns.The HEI-2015 was associated with diastolic blood pressure(DBP),SBP,and HTN to some extent,but the P-values were greater than those corresponding to the ddDQS.This study did not find associations be-tween the DASH,AHEI,and MED indices and DBP,SBP,or HTN.Conclusion Compared to commonly used diet quality indices,the ddDQS constructed using machine learning methods demonstrates better predic-tive ability for hypertension.A higher ddDQS is beneficial for reducing blood pressure levels and the risk of hy-pertension.The development of a diet quality index tailored to the Asian population can guide future research on hypertension prevention and control.
hypertensiondiet quality indexelastic net variable selectionmachine learning