Table 2.
Performance of different strategies when dealing with additional missing covariate data: Non-informative censoring (NIC), complete case analysis (CC), multiple imputation before ascertainment (MI), multiple imputation after ascertainment (MI (asc.)), IPW estimates after ascertainment with logistic regression based weights (IPW (asc.,lw)), and Bayesian model averaging weights (IPW (asc.,maw)). Missing covariate data was either multiple imputed (MI) or treated as a separate category (MC) when estimating the weights. Levels of ascertainment relate to the percentage of missing outcome information that was ascertained.
| Asc. | IPW (asc.) (lw,MI) | IPW (asc.) (lw,MC) | IPW (asc.) (maw,MI) | IPW (asc.) (maw,MC) | NIC | CC | MI | MI (asc.) | |
|---|---|---|---|---|---|---|---|---|---|
| 20% | SL | 0.8580 | 0.9230 | 0.8150 | 0.8590 | 6.4900 | 4.4660 | 1.9240 | 0.5720 |
| ML | 0.1100 | 0.1155 | 0.1067 | 0.1052 | 0.0821 | 0.2069 | 0.0612 | 0.0545 | |
| MSEw | 0.7000 | 0.9200 | 0.6500 | 0.8900 | |||||
|
| |||||||||
| 50% | SL | 0.2370 | 0.2460 | 0.2350 | 0.2390 | 6.5720 | 4.5540 | 1.9960 | 0.2060 |
| ML | 0.0649 | 0.0655 | 0.0641 | 0.0641 | 0.0844 | 0.1994 | 0.0742 | 0.0570 | |
| MSEw | 0.1600 | 0.2300 | 0.1500 | 0.2100 | |||||