Table 2.
Bias, variance, and coverage of logistic model parameters within the multistate cure model
| Baseline hazard | 2 4, 1 4 Assumption |
Logistic model | # Failed (out of 500) | Run time (min/sim) | ||
|---|---|---|---|---|---|---|
| Intercept |
|
|
||||
| Scenario 1: no covariate missingness or unequal follow-up | ||||||
| Weibull |
|
0 (0.33) 93 | 0 (0.42) 95 | 0 (0.44) 94 | 0 | 2.02 |
| Weibull |
|
0 (0.34) 94 | 0 (0.51) 94 | 0 (0.51) 94 | 0 | 2.12 |
| Weibull |
|
0 (0.59) 93 | 0 (0.41) 95 | 0 (0.45) 95 | 0 | 2.07 |
| Weibull | None | 1 (0.71) 93 | 0 (0.49) 96 | 0 (0.51) 97 | 46 | 2.16 |
| Cox |
|
0 (0.33) 93 | 0 (0.42) 97 | 0 (0.44) 95 | 0 | 7.68 |
| Cox |
|
0 (0.34) 93 | 0 (0.51) 95 | 0 (0.52) 95 | 0 | 8.04 |
| Cox |
|
1 (0.89) 84 | 0 (0.43) 94 | 0 (0.48) 95 | 0 | 7.98 |
| Cox | None | 11 (0.37) 50 | 2 (2.64) 60 | 2 (2.83) 64 | 1 | 8.46 |
| Scenario 2: covariate missingness | ||||||
| Weibull |
|
0 (0.32) 96 | 0 (0.44) 97 | 0 (0.60) 95 | 0 | 5.65 |
| Weibull |
|
0 (0.33) 96 | 0 (0.54) 97 | 0 (0.70) 95 | 1 | 5.71 |
| Weibull |
|
0 (0.63) 90 | 0 (0.47) 97 | 0 (0.60) 95 | 0 | 5.73 |
| Weibull | None | 2 (0.72) 89 | 0 (0.59) 95 | 0 (0.73) 93 | 141 | 5.57 |
| Cox |
|
0 (0.32) 96 | 0 (0.46) 96 | 0 (0.60) 95 | 0 | 27.5 |
| Cox |
|
0 (0.33) 94 | 1 (0.58) 96 |
2 (0.71) 93 |
0 | 27.3 |
| Cox |
|
1 (1.59) 81 | 1 (0.50) 96 |
3 (0.60) 94 |
0 | 27.4 |
| Cox | None | 6 (0.63) 80 |
2 (2.15) 61 |
2 (2.28) 67 |
82 | 26.7 |
| Scenario 3: unequal follow-up | ||||||
| Weibull |
|
0 (0.35) 96 | 0 (0.51) 95 | 0 (0.48) 96 | 0 | 8.49 |
| Weibull |
|
0 (0.35) 96 | 0 (0.64) 95 | 0 (0.63) 94 | 0 | 8.61 |
| Weibull |
|
0 (0.63) 92 | 0 (0.54) 94 | 0 (0.51) 96 | 0 | 8.56 |
| Weibull | None | 3 (0.69) 92 | 0 (0.66) 95 | 0 (0.64) 93 | 102 | 8.58 |
| Cox |
|
2 (0.35) 94 |
1 (0.53) 93 | 1 (0.48) 95 | 0 | 18.3 |
| Cox |
|
4 (0.43) 87 |
2 (0.77) 93 |
2 (0.81) 90 |
0 | 18.5 |
| Cox |
|
5 (1.95) 74 |
2 (0.60) 94 |
2 (0.57) 94 |
0 | 18.5 |
| Cox | None | 4 (0.70) 83 |
1 (2.65) 53 |
2 (2.44) 60 |
19 | 17.8 |
Results across 500 simulations are presented using the following notation: bias (empirical variance), coverage of 95% confidence interval, each multiplied by 100. The number of simulations (out of 500) with numerical issues and the median run time per simulation are also shown.
































