Alternative regression models to assess increase in childhood BMI

Area of Science:

• Biostatistics
• Pediatric Health
• Epidemiology

Background:

• Childhood body mass index (BMI) data often exhibit skewed distributions, posing challenges for traditional statistical models like linear or logistic regression.
• Existing methods may not adequately capture the complex relationships and variability in BMI data, limiting accurate risk factor assessment.

Purpose of the Study:

• To compare the effectiveness of different regression approaches in modeling childhood BMI.
• To identify the most suitable statistical methods for predicting childhood BMI and understanding associated risk factors.

Main Methods:

• A comparative analysis of generalized linear models (GLMs), quantile regression, and Generalized Additive Models for Location, Scale and Shape (GAMLSS) was performed.
• Data from 4967 children in Bavaria, Germany, were analyzed, considering risk factors such as TV watching, meal frequency, breastfeeding, smoking in pregnancy, maternal obesity, parental social class, and early-life weight gain.

Main Results:

• GAMLSS demonstrated a superior fit for estimating risk factor effects on BMI compared to GLMs, as indicated by the generalized Akaike information criterion.
• Quantile regression provided additional interpretability for specific BMI quantiles related to overweight and obesity.
• TV watching, maternal BMI, and weight gain in early life were significantly associated with body composition, while meal frequency showed an inverse association. Associations for other factors varied by model type.

Conclusions:

• GAMLSS and quantile regression offer more appropriate frameworks than standard GLMs for modeling BMI data and its risk factors.
• These advanced models facilitate the estimation of risk factor-specific BMI percentile curves, enhancing predictive and analytical capabilities.