Sample size considerations in observational health care quality studies

Area of Science:

• Health Services Research
• Biostatistics
• Epidemiology

Background:

• Health care quality studies often compare provider performance using clustered data (e.g., patients within hospitals).
• Observational studies, common in health quality research, may lack randomization, introducing bias.
• The primary inferential target in these studies is the cluster (provider), not the individual patient.

Purpose of the Study:

• To discuss sample size determination methods for observational health quality studies with binary outcomes.
• To highlight the advantages of hierarchical binomial models for cluster-level inference.
• To address sample size calculations for complex designs like unbalanced clusters and stratification.

Main Methods:

• Review of sample size calculations using marginal models.
• Detailed discussion of hierarchical binomial models for clustered binary data.
• Characterization of sample size for unbalanced clusters and stratified designs.

Main Results:

• Hierarchical models are recommended when the research interest is in comparing clusters (healthcare providers).
• The paper provides approaches for sample size determination in complex observational health quality study designs.
• Experiences from a cardiovascular disease quality of care study inform the methodological discussion.

Conclusions:

• Hierarchical models are preferred for analyzing clustered data in health quality research, especially when comparing providers.
• Accurate sample size determination is crucial for the validity of findings in observational health quality studies.
• The presented methods support robust study design for assessing healthcare quality.