A comparison of two methods for estimating prevalence ratios.

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Prevalence ratios are preferred over odds ratios for non-rare diseases/injuries in cross-sectional studies.
• Existing methods for modeling prevalence ratios face challenges like convergence issues and unreliable standard errors.
• Newer methods require evaluation for accuracy and reliability in multivariable models.

Purpose of the Study:

• Compare the performance of the Robust Poisson method and the log-binomial method for estimating prevalence ratios.
• Evaluate these methods using both simulated and real-world data.
• Identify the most suitable method for different prevalence levels and sample sizes.

Main Methods:

• The Robust Poisson method utilizes a Poisson distribution with a sandwich variance estimator.
• The log-binomial method employs a binomial distribution for maximum likelihood estimation.
• Both methods were assessed through computer simulations and analysis of real data.

Main Results:

• The Robust Poisson method provided less biased prevalence ratio estimates at very high prevalences with moderate sample size.
• The log-binomial method showed slightly less bias at moderate prevalences with moderate sample size.
• The log-binomial method generally offered higher statistical power and smaller standard errors compared to the Robust Poisson method.

Conclusions:

• The log-binomial method is generally preferred due to lower bias in common situations and its ability to ensure estimated prevalences are within the valid range [0, 1].
• While the Robust Poisson method is user-friendly, the log-binomial method's adherence to the correct model and maximum likelihood estimation leads to superior statistical properties.
• The log-binomial method consistently provides better power, smaller standard errors, and valid prevalence estimates.