TABLE 1.

Summary Comparison of Pooling Methods Considered

Feature	Covariate Sharing	Aggregated Data	Meta-Analysis	Propensity Score-Based Pooling
Privacy issues
Ability to distribute study data to working groups without compromising patient privacy or proprietary data	↓	↓	↑	↑
Ease of complying with HIPAA law and Centers for Medicare and Medicaid Services rules	↓	⇆	↑	↑
Analytic and statistical issues
Ability to cross-tabulate individual covariates and explore data	↑	↑	↓	↓
Ability to match patients across rather than within centers	↑	↓	↓	↑
Ability to use propensity score trimming to find the most representative patient population	⇆	⇆	↓	↑
Ability to evaluate dose-response relationships	↑	⇆	↑	⇆
Ability to detect effect modification by patient-level factors	↑	↓	↓	⇆
Ability to detect effect modification among center populations	↑	⇆	↑	↑
Ability to evaluate performance of confounder adjustment techniques	↑	⇆	⇆	↑
Operational issues
Ease of transferring and compiling centers’ datasets into a single analytic database	⇆	↑	N/A	⇆
Ability to have limited expertise in statistical analysis within each center	↑	↑	↓	⇆
Flexibility in modifying outcome models to include or exclude certain covariates	↑	⇆	↓	↓
Flexibility to perform subgroup analysis on subgroups that were not specified a priori	↑	↓	↓	↓
Flexibility to match cohorts on factors that were not specified a priori	↑	↓	↓	⇆
Flexibility to add or modify exclusion criteria that were not specified a priori	↑	↓	↓	↓
Ease of reuse of data for parallel research questions or other outcomes	↑	↑	↓	⇆
Computing time required	↑	↑	⇆	⇆
Speed of study execution	⇆	⇆	⇆	⇆
Investigators’ overall ability to understand and sense transparency in the analyses and results	↑	↑	⇆	⇆

↑ indicates method is well-suited for noted issue;

⇆, method is moderately well-suited for noted issue;

↓, method is poorly-suited for noted issue.