Development of computational algorithmics using biochemical data to predict dietary habits: insights from the dietary deal study

Abstract

Objectives: Assessing dietary intake and understanding the underlaying contributions to health is crucial from achieving metabolic wellbeing. Traditional methods to measure food intake such as food questionnaires and dietary recall have limitations in accuracy and reliability. This study aimed to develop a nutritional tool using easily available biochemical data to predict dietary habits. Methods: A total of 138 participants enrolled in the Dietary Deal cross-sectional study were assessed for diet quality using AHEI and MEDAS17 scores, categorized by median adherence (≤p50 or >p50). Adjusted logistic regressions (a-LR) identified biochemical markers associated with higher diet quality ( >p50). Model performance was evaluated using metrics: precision-recall (PR) and area under curves (AUC), sensitivity, specificity, positive (PPV) and negative predictive values (PNV). Results: Individuals in the >p50 category for both scores (AHEI and MEDAS17) consumed more pro-healthy foods and had higher values in diet-nutriscores. Two a-LR models (controlling for age, sex, BMI, physical activity, and SF-36) were developed. Probability classification in MEDAS17 >p50, associations (p-value < 0.1) were observed with glucose (OR=1.06), HDL (OR=1.04), calcium (OR=0.14), retinol (OR=0.01), ascorbate (OR=0.88), D25OH (OR=1.05), and HbA1c % (OR=0.43). Probability classification in AHEI >p50, associations (p-value < 0.1) were observed with platelet (OR=0.99), HDL (OR=0.96), copper (OR=0.98), insulin (OR=0.86), homocysteine (OR=1.33), ascorbate (OR=1.48). Both models showed moderate/high correct classification (AUC: 79% and 85%, sensitivity: 73% and 79%; specificity: 75% and 77%; PPV: 73% and 77%; PNV: 75% and 79% for MEDAS17 and AHEI, respectively). Preliminary computational algorithms were devised for probability classification based on the a-LR as a tool for nutritional practice, incorporating a weighted system to each variable. Conclusions: These findings suggest that simple biochemical data shows potential for predicting dietary habits, a steppingstone for personalized interventions in precision medicine. This study suggests some biomarkers can objectively assess food intake, paving the way for tailored personalized nutrition interventions based on individual needs.