Identifying Young Adults at High Risk for Weight Gain Using Machine Learning.

Area of Science:

• Public Health
• Machine Learning
• Obesity Research

Background:

• Rising rates of weight gain in young adults present a significant public health challenge.
• Early identification of at-risk individuals is crucial for effective intervention.
• Overweight and class 1 obesity are prevalent conditions in this demographic.

Purpose of the Study:

• To develop and evaluate electronic health record-based machine learning models for predicting significant weight gain (≥10% body weight) in young adults.
• To identify key predictors of weight gain within this population.

Main Methods:

• Seven machine learning models were assessed, including regression, random forest, neural network, gradient-boosted decision trees, and support vector machine (SVM).
• Predictors included demographics, obesity-related conditions, laboratory data, vital signs, and neighborhood variables.
• Models were trained and validated on a large cohort, with accuracy measured by the area under the receiver operating characteristic curves (AUC).

Main Results:

• The study included 24,183 young adults, with 14.2% experiencing ≥10% total body weight gain within two years.
• Model performance varied, with the gradient-boosted decision trees achieving the highest AUC (0.675).
• Demographics (age, sex) and baseline body mass index were significant predictors across most models.

Conclusions:

• The developed machine learning models demonstrated modest accuracy in identifying young adults at risk of substantial weight gain.
• Enhancing future predictive models may require the integration of behavioral and genetic data.
• These findings highlight the potential of EHR-based machine learning for public health initiatives in obesity prevention.