Table 2.
Simulation Metrics for Plasmode Analysis Resultsa, Small Covariate Set, 20% Index-Exposure-Prevalence Interaction Scenarios
| Scenario b | Bias | Variance | Square Root of MSE | AMB |
|---|---|---|---|---|
| Defaultc | ||||
| Crude | −0.103 | |||
| CEM | 0.327 | 0.226 | 0.577 | 0.967 |
| PSM | 0.067 | 0.040 | 0.210 | 0.886 |
| MDM | 0.131 | 0.041 | 0.242 | 0.792 |
| FS | 0.070 | 0.027 | 0.178 | 0.946 |
| Exaggeratedc | ||||
| Crude | −0.091 | |||
| CEM | 0.341 | 0.220 | 0.580 | 0.967 |
| PSM | 0.079 | 0.040 | 0.214 | 0.886 |
| MDM | 0.143 | 0.038 | 0.242 | 0.792 |
| FS | 0.080 | 0.023 | 0.172 | 0.946 |
Abbreviations: AMB, average proportional decrease in Mahalanobis balance; CEM, coarsened exact matching; FS, fine stratification by propensity score; MDM, Mahalanobis distance matching; MSE, mean squared error; PSM, propensity score matching.
a Using a data set based on data from the Pharmaceutical Assistance Contract for the Elderly (nonsteroidal antiinflammatory drug cohort), United States, 1999–2002.
b The product term represented the interaction between age and score on the Charlson Comorbidity Index.
c The “default” scenario maintained the original product term and the “exaggerated” scenario was based on a product term that was 200% greater than the default product term.