Related Experiment Videos
Comparing longitudinal binary outcomes in an observational oral health study
Brent J Shelton1, Gregg H Gilbert, Zhenmei Lu
1Department of Biostatistics, School of Public Health, The University of Alabama at Birmingham, 1665 University Boulevard, RPHB 327-H, Birmingham, AL 35294-0022, U.S.A. bshelton@uab.edu
Statistics in Medicine
|June 13, 2003
Summary
This study introduces sample selection models for longitudinal binary outcomes, crucial for accurate analysis of observational health data. Accounting for selection bias significantly altered conclusions, highlighting its importance in dental care research.
Area of Science:
- Biostatistics
- Health Services Research
- Epidemiology
Background:
- Observational studies are vital for treatment comparisons but face challenges from selection bias.
- Existing sample selection models are limited to cross-sectional data and continuous outcomes.
- Addressing selection bias is critical for reliable findings in health research.
Purpose of the Study:
- To extend sample selection models to longitudinal studies with binary outcomes.
- To apply these models to analyze chewing difficulty in relation to dental care use.
- To compare findings with and without accounting for selection bias.
Main Methods:
- Utilized a two-stage probit model with Generalized Estimating Equations (GEE) for correlated longitudinal binary data.
- Analyzed chewing difficulty outcomes measured bi-monthly over 24 months.
- Incorporated dental care use as a time-varying covariate.
Main Results:
- Accounting for selection bias due to unobserved confounders significantly impacted study conclusions.
- An adverse selection phenomenon was observed, where those needing treatment most were least likely to seek it.
- The proposed model provided different insights compared to a standard GEE model ignoring selection bias.
Conclusions:
- Sample selection models are effective for binary longitudinal observational data.
- The findings underscore the importance of addressing selection bias in health outcomes research.
- This methodology can be applied to various health outcomes studies to improve accuracy.