At Odds: Concerns Raised by Using Odds Ratios for Continuous or Common Dichotomous Outcomes in Research on Physical...

Area of Science:

• Environmental Epidemiology
• Biostatistics
• Public Health

Background:

• Research on obesity and the built environment frequently employs logistic regression and odds ratios.
• Odds ratios can present challenges in validity, interpretation, and communication for common outcomes such as obesity.

Purpose of the Study:

• To identify key issues with using odds ratios for common outcomes in obesity and built environment research.
• To propose alternative statistical measures and approaches for more accurate findings.

Main Methods:

• Analysis of data from 13,102 New York City residents on walkability and body mass index.
• Identification and illustration of three primary problems associated with odds ratio usage.

Main Results:

• Dichotomizing continuous measures like body mass index (BMI) leads to information loss, reduced statistical power, and amplified measurement error.
• Odds ratios are systematically higher than prevalence ratios, with significant inflation for common outcomes like obesity.
• Odds ratios can lead to erroneous conclusions in interaction analyses, particularly when outcome prevalence varies across subgroups.

Conclusions:

• Utilize continuous outcome data whenever possible to retain relevant information and statistical power.
• Employ prevalence ratios instead of odds ratios for common dichotomous outcomes to ensure accurate interpretation.
• When odds ratios are necessary, authors must report outcome prevalence across different exposure groups to contextualize findings.