Related Experiment Videos
[Logistic regression vs other generalized linear models to estimate prevalence rate ratios]
P Traissac1, Y Martin-Prével, F Delpeuch
1Unité de Nutrition IRD, Centre collaborateur de l'OMS, 911, Avenue Agropolis, 34032 Montpellier, France. traissac@mpl.ird.fr
Revue D'Epidemiologie Et De Sante Publique
|February 16, 2000
Summary
Prevalence rate ratios, not odds ratios, should be used to estimate disease risk in cross-sectional studies. Logistic regression can lead to inaccurate results, especially with overdispersion, while log-binomial models offer a better approach.
Area of Science:
- Epidemiology
- Biostatistics
Context:
- Cross-sectional studies frequently use odds ratios and logistic regression to assess disease risk.
- This approach is often inaccurate due to the rare disease assumption and issues with variance estimation.
Purpose:
- To highlight the limitations of odds ratios and logistic regression in estimating prevalence ratios.
- To advocate for the use of log-binomial models for more accurate risk assessment.
Summary:
- Prevalence rate ratios are more appropriate than odds ratios for quantifying disease risk in cross-sectional studies.
- Log-binomial models directly estimate prevalence rate ratios and handle overdispersion better than logistic regression.
- Overdispersion in logistic regression can lead to underestimation of type I error rates.
Impact:
- Promotes more accurate epidemiological research and public health assessments.
- Encourages the adoption of statistically sound methods for disease risk quantification.
- Improves the reliability of findings from cross-sectional studies.