Prediction of Diet Quality Based on Day-Level Meal Pattern: A Preliminary Analysis Using Decision Tree Modeling. (14th June 2022)

• Record Type: Journal Article
• Authors: Lin, Annie; Cornely, Adrian; Kalam, Faiza; Sen, Sougata; Makelarski, Jennifer; Colvin, Christopher; Ward, Danielle; Mirsky, Grace
• Is Part Of: Current developments in nutrition. Volume 6(2022)Supplement 1
• Page Start: 417
• Page End: 417
• Publication Date: 2022-06-14
• Subjects: Nutrition -- Periodicals; Nutritional Physiological Phenomena; Nutrition; Periodicals; Periodicals; Fulltext; Internet Resources; Periodicals; 612.3
• DOI: 10.1093/cdn/nzac055.006
• Languages: English
• ISSNs: 2475-2991

Abstract

Abstract: Objectives: Previous studies have investigated if meal timing is associated with energy and macronutrient intake. However, few focus on the combination of food intake and meal timing and their association with diet quality. We use machine learning to examine how day-level meal patterns (food group intake, meal timing) predict diet quality among adults. Methods: We analyzed diet data from interviewer-administered 24-hour recalls from the NHANES 2015–2016 and 2017–2018 cycles (N = 9761). Fifteen food groups were examined: fruit, fruit juice, vegetables, whole grains, meats (red, cured, poultry, seafood), eggs, plant protein, dairy, oils, solid fats, added sugar, and alcohol. Proportion of intake for each food group per participant - relative to that participant's total daily intake of that food group - was included as input parameters for each meal (breakfast, lunch, dinner). Diet quality was computed using the Healthy Eating Index 2015 (HEI-2015); higher scores represented better diet quality. Cutoff threshold for a higher vs. lower quality diet was defined by the 75th percentile of HEI-2015 for the dataset (cutoff = 59.24). Decision tree modeling identified the inputs that contributed to the highest information gain and the optimal classification threshold for each input. Results: On average, participants consumed 0.4 ± 0.8 cup equivalents of fruits and 0.6 ± 1.2 ounce equivalents of whole grains. These two food groups contributed most to the diet quality prediction model, which had a 78% classification accuracy for the dataset. Lower quality diets were associated with: a) < 2% of both total fruit and whole grain intake at breakfast; b) >2% of total fruit but < 2% of total whole grain intake at breakfast; or c) >2% of total whole grain intake at breakfast but < 2% of total fruit intake at breakfast and lunch. Higher quality diets were associated with: a) >2% of both total fruit and whole grain intake at breakfast or b) < 2% of total fruit intake at breakfast but > 2% of both total whole grain intake at breakfast and fruit intake at lunch. Food group intake at dinner was not a top predictor in the preliminary model. Conclusions: Preliminary analyses revealed that the timing of fruit and whole grain consumption are important predictors of diet quality. Future studies should test the preliminary model on additional datasets and should include snacking episodes in the analyses. Funding Sources: None.