Using classification and regression trees to model missingness in youth BMI, height and body mass data

Area of Science:

• Public Health
• Biostatistics
• Adolescent Health

Background:

• Missing data in youth body mass index (BMI) self-reports can significantly impact research findings.
• Previous analyses of missing BMI data in adolescents were limited by logistic regression's inability to identify subgroups or variable importance.
• Understanding missing data patterns is crucial for accurate health behavior research in youth.

Purpose of the Study:

• To investigate patterns and identify subgroups associated with missing body mass index (BMI) data in a large cohort of Canadian youth.
• To utilize advanced statistical modeling to uncover factors contributing to missing BMI measurements.

Main Methods:

• Employed sex-stratified classification and regression tree (CART) models on data from 74,501 Canadian youth (2018/19 COMPASS study).
• Examined associations between missing height, body mass, and BMI data and variables related to diet, physical activity, academics, mental health, and substance use.
• 31% of BMI data were identified as missing within the study cohort.

Main Results:

• CART models identified specific subgroups of male and female youth with a high likelihood of missing BMI values.
• Key factors associated with missing BMI included being younger, perceiving oneself as overweight, lower physical activity levels, and poorer mental health.
• Conversely, older youth who did not perceive themselves as overweight were less likely to have missing BMI data.

Conclusions:

• Excluding cases with missing BMI can introduce bias, skewing samples towards healthier youth.
• CART models effectively identify subgroups and variable hierarchies, proving invaluable for analyzing missing data patterns.
• These findings highlight the need for careful consideration of missing data in youth health research to ensure representative results.