Estimated treatment effect (HR) and false positive rate from 3 models if toxicity of chemotherapy causes more early deaths
| Relapse ∼ Weibull (λ = 0.05, γ = 1.5); Nonrelapse death ∼ Weibull (λ = 0.15, γ = 1.5); . | ||||
|---|---|---|---|---|
| Post-relapse death ∼ Weibull (λ = 0.3, γ = 1.5); β(X1) = 0.5; β(X2) = 0.75; . | ||||
| Time from CR1 to allo-HSCT ∼ U (0, 5) . | ||||
| . | HRRFS . | FPRRFS . | HROS . | FPROS . |
| Naïve Cox | 0.309 | 1 | 0.313 | 1 |
| Time-dependent Cox | 1.007 | 0.049 | 0.962 | 0.053 |
| Super Landmark | 1.007 | 0.061 | 0.956 | 0.067 |
| Relapse ∼ Weibull (λ = 0.05, γ = 1.5); Nonrelapse death ∼ Weibull (λ = 0.15, γ = 1.5); . | ||||
|---|---|---|---|---|
| Post-relapse death ∼ Weibull (λ = 0.3, γ = 1.5); β(X1) = 0.5; β(X2) = 0.75; . | ||||
| Time from CR1 to allo-HSCT ∼ U (0, 5) . | ||||
| . | HRRFS . | FPRRFS . | HROS . | FPROS . |
| Naïve Cox | 0.309 | 1 | 0.313 | 1 |
| Time-dependent Cox | 1.007 | 0.049 | 0.962 | 0.053 |
| Super Landmark | 1.007 | 0.061 | 0.956 | 0.067 |
Simulation replicates = 1000 and sample size = 600; HR, hazard ratio, is averaged over the results from 1000 replicates; FPR, false-positive rate, is the proportion of significant effect estimates of allo-HSCT in 1000 replicates, where the true treatment effect does not exist; RFS: relapse-free survival; OS: overall survival.
Allo-HSCT is fixed in the naïve Cox model and is time-dependent in the multivariate time-dependent Cox model and in the super landmark model; relapse is included as a covariate (fixed or time-dependent, as appropriate) in all 3 models for OS; the 2 prognostic factors (X1 and X2) are also included.