Table 3.
Results from feature selection by least absolute shrinkage and selection operator (LASSO) for 3 days (area under the curve of the receiver operating characteristics are reported with the SE). The best performing model refers to the model with a lambda value associated with minimum cross-validation error. The adjusted model refers to a LASSO model with the largest value of lambda such that the error remains within 1 SE of the minimum.
| Criteria, outcome, and imputation method | Day 1 | Day 2 | Day 3 | ||
| AUROCa for best performing model | |||||
| 30-day mortality | |||||
| HDb | 0.7858 (0.0033) | 0.7685 (0.0041) | 0.7302 (0.0043) | ||
| PMMc | 0.7876 (0.0039) | 0.7708 (0.0046) | 0.7391 (0.0053) | ||
| In-hospital mortality | |||||
| HD | 0.7983 (0.0040) | 0.7804 (0.0046) | 0.7476 (0.0042) | ||
| PMM | 0.8007 (0.0047) | 0.7838 (0.0049) | 0.7582 (0.0054) | ||
| Indicators among selected predictors by the best performing model, n (%) | |||||
| 30-day mortality | |||||
| HD | 23 (43) | 24 (48) | 19 (707) | ||
| PMM | 26 (45) | 26 (47) | 17 (68) | ||
| In-hospital mortality | |||||
| HD | 28 (46) | 29 (48) | 21 (60) | ||
| PMM | 29 (47) | 27 (49) | 24 (62) | ||
| AUROC for adjusted model | |||||
| 30-day mortality | |||||
| HD | 0.7826 (0.0034) | 0.7646 (0.0043) | 0.7262 (0.0041) | ||
| PMM | 0.7840 (0.0038) | 0.7667 (0.0045) | 0.7339 (0.0044) | ||
| In-hospital mortality | |||||
| HD | 0.7944 (0.0043) | 0.7762 (0.0047) | 0.7439 (0.0041) | ||
| PMM | 0.7961 (0.0049) | 0.7793 (0.0050) | 0.7536 (0.0045) | ||
| Indicators among selected predictors by the adjusted model, n (%) | |||||
| 30-day mortality | |||||
| HD | 20 (45) | 16 (48) | 22 (67) | ||
| PMM | 19 (45) | 16 (52) | 31 (62) | ||
| In-hospital mortality | |||||
| HD | 20 (47) | 13 (42) | 16 (64) | ||
| PMM | 18 (50) | 11 (41) | 16 (62) | ||
aAUROC: area under the curve of the receiver operating characteristic.
bHD: hot deck.
cPMM: predictive mean matching.