A Simulation Study of Categorizing Continuous Exposure Variables Measured with Error in Autism Research: Small...

Area of Science:

• Epidemiology
• Biostatistics
• Genetics

Background:

• Categorizing continuous variables for analysis can introduce bias.
• The impact of cutoff selection on odds ratios (ORs) is often overlooked.
• Exposure measurement error is a common issue in epidemiological studies, particularly in autism spectrum disorder (ASD) research.

Purpose of the Study:

• To investigate the effect of varying cutoffs on odds ratios (ORs) when categorizing a mismeasured exposure.
• To assess how measurement error and exposure-outcome relationships influence the observed ORs.
• To understand the implications for autism spectrum disorder (ASD) research.

Main Methods:

• Simulated cohorts with varying exposure-outcome curves and measurement error variances.
• Logistic regression used to calculate ORs across 61 dichotomized cutoffs and five-category cutoffs.
• Calculation of ORs, sensitivity, and specificity for each scenario and cutoff.

Main Results:

• Distinct OR (versus cutoff) curves were observed, though obscured by increasing measurement error.
• Non-monotonic oscillations in ORs were noted at extreme cutoffs, unrelated to small sample sizes.
• Differential exposure misclassification occurred, with higher sensitivity in cases and specificity in controls.

Conclusions:

• Cutoff selection for continuous variables profoundly affects study results and odds ratios.
• Measurement error can obscure true exposure-disease relationships.
• The OR curve shape may reveal underlying relationships if measurement error is minimal.