Table 2.
Handling of missing data in primary analysis among 73 trials who reported missing outcome data
| N (%) 1 | |
|---|---|
| Proportion of patients with missing outcome1 | |
| No missing data | 4 (5) |
| 0 – 1% | 2 (3) |
| 1 – 5% | 11 (14) |
| 5 – 10% | 24 (31) |
| > 10% | 36 (47) |
| Reported number of patients with missing outcomes by randomized treatment arm | 71 (97) |
| Reported reasons why missing | 66 (90) |
| Mentioned attempts to avoid missing data before and during trial | 26 (36) |
| Methods | |
| Complete case | 33 (45) |
| Simple imputation | |
| Linear interpolation | 3 (5) |
| Worst-case | 8 (11) |
| LOCF | 9 (12) |
| Multiple imputation | 6 (8) |
| Model based | |
| GEE (un-weighted) | 3 (4) |
| Mixed model/hierarchical/multilevel | 11 (15) |
1The denominator for the proportion of patients with missing outcome is 77; the other denominators are 73 (the number of studies with any missing data).