Comparison of methods for the analysis of longitudinal interval count data

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Survival Analysis

Background:

• Longitudinal studies frequently analyze the recurrence rate of non-fatal events.
• Often, only aggregated event counts over time intervals are available.
• Accurate statistical modeling is crucial for reliable analysis of such data.

Purpose of the Study:

• To compare the performance of Thall's mixed Poisson-gamma regression and Zeger and Liang's quasi-likelihood methods.
• To evaluate their accuracy in estimating standard errors for longitudinal event recurrence data.
• To identify conditions where one method may be superior to the other.

Main Methods:

• Simulation techniques with large sample sizes were employed.
• The mixed Poisson-gamma regression (Thall) and quasi-likelihood (Zeger-Liang) methods were compared.
• Analysis focused on correctly specified mean models and scenarios with time-dependent covariates.

Main Results:

• Both methods yielded similar standard errors under most conditions.
• The Thall method significantly underestimated standard errors when using time-dependent covariates with non-Poisson-gamma data.
• A method to differentiate between the variance structures of the two models was presented.

Conclusions:

• The Zeger-Liang quasi-likelihood method is more robust for longitudinal data with time-dependent covariates than Thall's method.
• The study provides practical guidance for analyzing recurrent event data in clinical trials.
• Findings extend existing knowledge on variance misspecification in longitudinal models.