Building Machine Learning Models to Correct Self-Reported Anthropometric Measures.

Area of Science:

• Public Health
• Biostatistics
• Machine Learning

Background:

• Population obesity monitoring relies on self-reported data, which is susceptible to recall errors and bias.
• Accurate anthropometric measurements are crucial for understanding obesity prevalence and associated health risks.

Purpose of the Study:

• To develop and validate machine learning (ML) models for correcting self-reported height and weight.
• To estimate obesity prevalence in US adults using corrected anthropometric data.

Main Methods:

• Utilized individual-level data from 50,274 US adults from the National Health and Nutrition Examination Survey (NHANES) 1999-2020.
• Applied nine ML models to predict objectively measured height, weight, and body mass index from self-reported data.
• Assessed model performance using root-mean-square error.

Main Results:

• The best-performing ML models significantly reduced discrepancies between self-reported and objectively measured height (22.08%), weight (2.02%), and body mass index (11.14%).
• Obesity prevalence estimation using ML models showed a 99.52% reduction in discrepancy compared to self-reported data.
• Predicted obesity prevalence (36.05%) was statistically indistinguishable from the objectively measured prevalence (36.03%).

Conclusions:

• Developed ML models can reliably correct self-reported anthropometric data in large population surveys.
• This methodology offers a robust approach to accurately estimate obesity prevalence in US adults.
• The findings support the use of ML-driven data correction for improved public health surveillance.