# Phenotype-Driven Variability in Longitudinal Body Composition Changes After a Very Low-Calorie Ketogenic Intervention: A Machine Learning Cluster Approach

## TL;DR

This study uses machine learning to identify how factors like sex and initial weight affect weight loss on a very low-calorie ketogenic diet, helping personalize obesity treatments.

## Contribution

The study introduces a machine learning cluster approach to classify patient phenotypes and predict weight-loss success in a VLCKD program.

## Key findings

- Male gender and higher initial body weight strongly predict greater weight loss on a VLCKD.
- Two distinct patient clusters with unique weight-loss patterns were identified based on age, sex, and follow-up duration.
- Personalized strategies based on these clusters can improve treatment outcomes in obesity management.

## Abstract

Background: Obesity is a major global public health issue with no fully satisfactory solutions. Most nutritional interventions rely on caloric restriction, with varying degrees of success. Very low-calorie ketogenic diets (VLCKD) have demonstrated rapid and sustained weight loss by inducing ketone bodies through lipolysis, reducing appetite, and preserving lean mass while maintaining metabolic health. Methods: A prospective clinical study analyzed sociodemographic, anthropometric, and adherence data from 7775 patients undergoing a multidisciplinary nutritional single-arm intervention based on a commercial weight-loss program. This method, using protein preparations with a specific balanced nutritional profile, aimed to identify key predictors of weight-loss success and classify population phenotypes with shared baseline characteristics and weight-loss patterns to optimize treatment personalization. Results: Statistical and machine learning analyses revealed that male gender (−9.2 kg vs. −5.9 kg) and higher initial body weight (−8.9 kg vs. −4.0 kg) strongly predict greater weight loss on a VLCKD, while age has a lesser impact. Two distinct population clusters emerged, differing in age, sex, follow-up duration, and medical visits, demonstrating unique weight-loss success patterns. These clusters help define individualized strategies for optimizing outcomes. Conclusions: These findings translationally support associations with the efficacy of a multidisciplinary VLCK weight-loss program and highlight predictors of success. Recognizing variables such as sex, age, and initial weight enhances the potential for a precision-based approach in obesity management, enabling more tailored and effective treatments for diverse patient profiles and prescribe weight loss personalized recommendations.