Table 2.
Construction of a frailty score in the training cohort
| HR (95% CI) | P | Score | |
|---|---|---|---|
| ADL | |||
| Independent | 1 | 1 | |
| Dependent | 2.07 (1.59-2.71) | <.001 | 2 |
| CCI | |||
| Score 0-1 | 1 | 1 | |
| Score 2 | 1.53 (1.14-2.04) | .004 | 1.5 |
| Score ≥3 | 1.92 (1.45-2.55) | <.001 | 2 |
| GNRI | |||
| Absent/low | 1 | 1 | |
| Moderate | 2.01 (1.49-2.70) | <.001 | 2 |
| Severe | 2.31 (1.61-3.30) | <.001 | 2.5 |
| Age, y | |||
| <85 | 1 | 1 | |
| ≥85 | 2.25 (1.70-2.98) | <.001 | 2 |
| IPI score | |||
| Low (1) | 1 | — | |
| Low-intermediate (2) | 1.13 (0.72-1.80) | .589 | — |
| High-intermediate (3) | 1.71 (1.13-2.60) | .012 | — |
| High (4-5) | 2.75 (1.83-4.12) | <.001 | — |
Multivariate Cox regression analysis with HR for death from all causes (OS), adjusted for IPI score. The model was developed in the training cohort (n = 522 patients). Missing values were imputed using MICE. Follow-up was limited to 2 years for GNRI to obtain proportional hazard, otherwise follow-up was limited to 5 years. Weights for the geriatric assessment variables were obtained by rounding the HR to the closest 0.5 score. The frailty score was then created by multiplication of the rounded HRs, producing a frailty score ranging from 1 to 20.