Who Is at High Risk of Poor Diet Quality? A Precision Public Health Nutrition Approach using Machine Learning

Area of Science:

• Public Health Nutrition
• Computational Epidemiology
• Health Informatics

Background:

• Suboptimal diet quality significantly impacts public health, influenced by complex socioecological factors.
• Operationalizing a precision public health nutrition framework to address these factors is challenging.
• Identifying specific population subgroups with poor diet quality is crucial for targeted interventions.

Purpose of the Study:

• To evaluate machine learning's utility in operationalizing precision public health nutrition.
• To precisely identify subgroups with the highest prevalence of poor diet quality.
• To determine the most significant predictors of diet quality within these subgroups.

Main Methods:

• Secondary analysis of the 2018-2019 International Food Policy Study in Canada (n=5093).
• Utilized 42 candidate predictors across sociodemographic, policy, literacy, and health domains.
• Employed conditional inference trees (CIT) and conditional random forests (CRF) with the Healthy Eating Index-2015 (HEI-2015) as the diet quality outcome.

Main Results:

• CIT identified 7 subgroups with varying diet quality, with lower diet quality probabilities ranging from 16.9% to 49.9%.
• The subgroup with the highest proportion of poor diet quality comprised individuals with limited cooking technique confidence (≤4).
• Top predictors of diet quality included food label use, cooking confidence, perceived health, food insecurity, and health literacy.

Conclusions:

• Machine learning methods are effective for operationalizing precision public health nutrition.
• This approach enables precise identification of at-risk subgroups and key diet quality predictors.
• Findings support data-driven strategies for improving population diet quality.