Table 2.
Imputation model summary
| Name | Model(s) | MI1? | Note |
|---|---|---|---|
| RI2 | – | No | Imputed gated as 03 |
| Ignore | Row-mean4 | No | Typical approach |
| Ignore (weighted)5 | Row-mean4 | No | Linear weights |
| Ignore20 | Row-mean4 | No | 20% missingness cut |
| RF | RF | No | 100 trees |
| kNN | kNN | No | – |
| MICE Default | PMM/logreg6 | Yes | – |
| MICE CART | CART | Yes | 1 tree |
| MICE CART+Aux | CART | Yes | 100 auxiliary variables |
| MICE RF | RF | Yes | 10 trees |
1 MI, multiple imputations
2 RI, rule-based imputation
3 Gated variables were PFQ, RXD and VIQ blocks (Supplemental Table SV)
4 Mean value of the available deficit data for each individual
5 Results in Supplemental
6 PMM for continuous (lab) variables, logreg for ordinal/binary (self-reported) variables