Robust power and sample size calculations in quasi-likelihood models: methods and practice

Area of Science:

• Biostatistics
• Epidemiology
• Health Services Research

Background:

• Accurate power and sample size calculations are critical for study planning but challenging for quasi-likelihood (QL) models.
• Traditional methods often rely on restrictive distributional assumptions, limiting their use with non-standard response distributions or complex covariate structures.

Purpose of the Study:

• To evaluate the effectiveness of two generalized linear model (GLM)-based effect size measures, 2 Standard Deviations in the Linear Predictor (2SLiP) and Pseudo-Partial [Formula: see text] (P2R2), for power and sample size calculations within the QL framework.
• To assess the utility of these measures for both Wald and score tests across various simulation settings.
• To demonstrate practical application in a real-world health scenario.

Main Methods:

• Extensive simulations were conducted to assess the performance of 2SLiP and P2R2 across diverse outcome types, link functions, and variance structures.
• The measures were evaluated under Wald tests and their applicability to score tests was explored.
• A case study applied these effect size measures to survey data on healthcare workers and personal protective equipment (PPE) adequacy in relation to burnout risk during the COVID-19 pandemic.

Main Results:

• Both 2SLiP and P2R2 demonstrated accuracy in power and sample size calculations for QL models, with estimates within 3% and 2% of targets, respectively.
• The effect size measures are fundamentally moment-based and extend directly to QL models.
• In the case study, estimated effect sizes indicated small but meaningful associations between PPE adequacy and burnout risk, with accurate sample size recommendations.

Conclusions:

• 2SLiP and P2R2 provide robust and reliable alternatives for power and sample size calculations in quasi-likelihood settings.
• These measures require minimal distributional assumptions, enhancing flexibility and applicability for realistic medical and public health study designs.