Related Experiment Videos
A SAS macro for the analysis of multivariate longitudinal binary outcomes.
Brent J Shelton1, Gregg H Gilbert, Bin Liu
1Department of Internal Medicine, UK Markey Cancer Center, Lexington, KY, USA. bshelton@kcp.uky.edu
Computer Methods and Programs in Biomedicine
|September 29, 2004
Summary
This study introduces a SAS macro for analyzing multiple correlated binary outcomes in oral health research. The new method improves accuracy in estimating covariate effects, particularly for problem-oriented visits, compared to separate analyses.
Area of Science:
- Biostatistics
- Oral Health Research
- Longitudinal Data Analysis
Background:
- Multiple binary outcomes are common in health research, necessitating methods that account for outcome correlations.
- Existing software for correlated binary outcomes often requires extensive data pre-processing.
- Accurate statistical adjustment for correlations between outcomes is crucial when comparing covariate effects.
Purpose of the Study:
- To present a SAS macro for estimating covariate effects on multiple, correlated binary outcomes.
- To demonstrate the macro's utility in analyzing longitudinal oral health data.
- To compare results from a multivariate model with traditional univariate analyses.
Main Methods:
- Development and application of a SAS macro for multivariate logistic regression using Generalized Estimating Equations (GEE).
- Analysis of three correlated longitudinal oral health outcomes (problem-oriented visit, dental cleaning, routine check-up) measured over 24 months.
- Comparison of trivariate GEE model estimates against separate univariate GEE models for each outcome.
Main Results:
- The multivariate model identified a significant association between male sex and increased odds of problem-oriented visits (P = 0.0407), which was non-significant in univariate analysis (P = 0.0641).
- Consistent regular attenders showed higher odds of dental cleanings/check-ups versus problem-oriented visits compared to consistent problem-oriented attenders (chi2 = 33.47, P < 0.01).
- Individuals with broken teeth or fillings had greater odds of problem-oriented visits relative to cleanings/check-ups (chi2 = 34.12, P < 0.01 and chi2 = 17.11, P < 0.01).
Conclusions:
- The developed SAS macro effectively estimates covariate effects in multivariate correlated binary outcomes.
- Multivariate analysis provides more accurate and significant findings than separate univariate analyses for oral health data.
- The macro facilitates a deeper understanding of factors influencing distinct but related oral health behaviors.