Primary analysis: multivariate penalized Fine-Gray model analysis of CHARM variables
| Variable . | Coefficient (log HR scale)∗ . | P value . | Coefficient (adjusted for non-CHARM variables)† . | Median . | Quartile with elevated NRM risk . | Subdist HR‡ . | 95% CI lower . | 95% CI upper . |
|---|---|---|---|---|---|---|---|---|
| HCT-CI | 0.1 296 686 | <.0001 | 0.1 357 042 | 3 | 4 | 1.138 | 1.069 | 1.212 |
| Log(CRP)∗ | 0.1 118 738 | .0493 | 0.1 141 597 | 2 | 6 | 1.131 | 1.000 | 1.278 |
| Albumin | −0.5 989 142 | .0008 | −0.5 920 071 | 4 | 3.7 | 1.197 | 1.078 | 1.329 |
| Weight loss2 | 0.001 110 624 | .0089 | 0.001 190 834 | 3.2% | 8.1% | 1.063 | 1.016 | 1.114 |
| PROkps2 | −6.45035E-05 | .1115 | −6.74881E-05 | 90 | 80 | 1.116 | 0.975 | 1.277 |
| Age at HCT | 0.03 748 089 | .0317 | 0.04 346 934 | 67.5 | 71.2 | 1.149 | 1.012 | 1.304 |
| MoCA | −0.04 649 406 | .1106 | −0.04 011 149 | 26 | 24 | 1.097 | 0.979 | 1.230 |
| Variable . | Coefficient (log HR scale)∗ . | P value . | Coefficient (adjusted for non-CHARM variables)† . | Median . | Quartile with elevated NRM risk . | Subdist HR‡ . | 95% CI lower . | 95% CI upper . |
|---|---|---|---|---|---|---|---|---|
| HCT-CI | 0.1 296 686 | <.0001 | 0.1 357 042 | 3 | 4 | 1.138 | 1.069 | 1.212 |
| Log(CRP)∗ | 0.1 118 738 | .0493 | 0.1 141 597 | 2 | 6 | 1.131 | 1.000 | 1.278 |
| Albumin | −0.5 989 142 | .0008 | −0.5 920 071 | 4 | 3.7 | 1.197 | 1.078 | 1.329 |
| Weight loss2 | 0.001 110 624 | .0089 | 0.001 190 834 | 3.2% | 8.1% | 1.063 | 1.016 | 1.114 |
| PROkps2 | −6.45035E-05 | .1115 | −6.74881E-05 | 90 | 80 | 1.116 | 0.975 | 1.277 |
| Age at HCT | 0.03 748 089 | .0317 | 0.04 346 934 | 67.5 | 71.2 | 1.149 | 1.012 | 1.304 |
| MoCA | −0.04 649 406 | .1106 | −0.04 011 149 | 26 | 24 | 1.097 | 0.979 | 1.230 |
Final CHARM score calculation = 0.1 296 686 × HCT-CI + 0.1 118 738 × log (CRP) – 0.5 989 142 × albumin + 0.001 110 624 × weight loss2 – 0.0 000 645 035 × KPS2 + 0.03 748 089 × age at HCT – 0.04 649 406 × MoCA. An online CHARM calculator is available at: https://cibmtr.org/CIBMTR/OffNav/DevSandbox/CHARM-Risk-NRM-Calculator.
log(CRP), log of C-reactive protein; PROkps2, indicates patient-reported KPS squared; weight loss2, percentage of weight loss comparing weight 1 year before allogeneic HCT to weight at the time of study evaluation squared.
Natural log; positive coefficients indicate higher risk with higher variable values; negative coefficients indicate higher risk with lower variable values.
The model was adjusted for characteristics of conditioning intensity, donor/recipient cytomegalovirus serology matching, or donor type/HLA matching forcing covariates in the model to see whether the effect of the primary-CHARM health variables differed after adjustment for these factors.
Subdistribution HR from the Fine-Gray model refers to the relative change in the instantaneous rate of the occurrence of the event (NRM) in those subjects who are event free or who have experienced a competing event (relapse). Subdistribution HR and associated CIs are for the comparison of the quartile with elevated risk vs the median value for each CHARM variable, to account for the highly varying scales of each measurement. Quartiles and medians are expressed on the original scale rather than the transformed (log or squared) scales. Quartile with elevated risk is defined as the 75th percentile for measures where increasing values are associated with higher NRM risk (HCT-CI, log(CRP), weight loss, and age) and the 25th percentile for measures where decreasing values are associated with higher NRM risk (albumin, PROkps, and MoCA).