Use of Machine Learning to Identify Determinants of Habitual-Preformed Water Intake.

Area of Science:

• Nutrition Science
• Biostatistics
• Physiology

Background:

• Adequate water intake is crucial for overall health.
• Understanding the determinants of preformed water consumption in adults remains limited.
• Preformed water intake includes water from beverages and food.

Purpose of the Study:

• To apply machine learning (ML) models to identify factors associated with preformed water intake in healthy adults.
• To compare ML approaches with traditional statistical methods for identifying hydration predictors.
• To explore the relationship between preformed water intake and various dietary, physiological, and body composition variables.

Main Methods:

• Secondary analysis of baseline data from 219 participants in the CALERIE™ 2 trial.
• Quantification of habitual preformed water intake using deuterium and oxygen-18 isotope data.
• Development and comparison of various regression models (linear, tree-based, penalized) to identify associated factors.

Main Results:

• A ridge regression model incorporating 25 variables best explained 38% of the variance in preformed water intake.
• Higher preformed water intake correlated with increased intake of fiber, protein, alcohol, and total food, and decreased intake of carbohydrates and sodium.
• ML models highlighted alcohol and potassium intake as significant predictors, which were not identified by traditional linear regression.

Conclusions:

• Data-driven ML models can uncover complex patterns related to preformed water intake that traditional methods may miss.
• These findings can inform strategies for identifying individuals at risk of inadequate hydration.
• ML offers a powerful approach for analyzing intricate datasets in nutritional and health research.