| | Referent model | Gait speed model | Model-based PFP |
|---|
Model performance (95% CrI) |
R2 | 0.22 (0.17 to 0.27) | 0.27 (0.22 to 0.32) | 0.37 (0.32 to 0.41) |
Approximate LOO cross-validation |
LOO-R2 | 0.22 (0.17 to 0.27) | 0.26 (0.21 to 0.31) | 0.35 (0.29 to 0.40) |
LOO-ELPDdiff (SE)b | −90.9 (14.9) | −50.7 (12.6) | 0.0 |
Classification performance (95% CI) |
Cohen kappac | 0.36 (0.30 to 0.42) | 0.40 (0.35 to 0.45) | 0.47 (0.42 to 0.52) |
AUROCd | 0.67 (0.64 to 0.69) | 0.74 (0.71 to 0.77) | 0.77 (0.74 to 0.80) |
Overall NRId | - | 0.05 (−0.02 to 0.13) | 0.11 (0.05 to 0.18) |
Event NRI | - | 0.20 (0.14 to 0.25) | 0.23 (0.18 to 0.27) |
Non-event NRI | - | −0.14 (−0.09 to −0.19) | −0.11 (−0.07 to −0.16) |
- Abbreviations: PFP Physical Frailty Phenotype, CrI credibility interval, CI confidence interval, SE standard error, LOO-CV leave-one-subject-out cross-validation, LOO-R2 leave-one-subject-out R-squared statistic, ELPDdiff pairwise difference in leave-one-out expected log posterior density, AUROC area under the receiver operating characteristic curve, NRI net reclassification index
- aModel performance of the model-based physical frailty phenotype (PFP) (a model with non-dichotomized PFP criterion predictors) was compared to that of the referent model (a model with only the PFP count score) and the gait speed model (a model with gait speed and standard covariates)
- bPairwise difference in leave-one-out (LOO) expected log posterior density (denoted using ELPDdiff) between models and its standard error (SE). As ELPDdiff was estimated with respect to the best-performing model, an absolute ELPDdiff of greater than twice its SE was taken as evidence that the best-performing model (with a ELPdiff of 0) had better out-of-sample predictive performance than the alternative model
- cCohen’s quadratic-weighted kappa coefficients computed based on frailty index-defined robust (≤0.10), pre-frail (>0.10 to 0.21), and frail (>0.21) categories
- dAUROC (area under the receiver operating characteristic curve) and net reclassification index (NRI) computed based on frailty index-defined robust (≤0.10) and pre-frail/frail (>0.10) categories
