Table 1 Combinations of intercurrent event strategies that were illustrated in this case study and their corresponding analysis methods
Figure | Intercurrent event strategies | Analysis | ||||
|---|---|---|---|---|---|---|
Strategy | While no IE | Composite | Hypothetical | Treatment policy | QoL data included in the analysisa | Estimation of mean QoL at each cycle |
Figure 3 | ||||||
1a,b | Death | PD, TD | All outcomes until each patient’s respective death in the first 40 cyclesb, including outcomes after TD or PD | 1a. GEE with independence correlation structure, where cycle number was included as a categorical variable [26] 1b. An LMM with random intercept and slope was estimated, where the cycle number was included as a categorical variable. Individual outcome predictions from this LMM were computed at each cycle and averaged only over those patients alive at that cycle. | ||
2 | Death | PD, TD | All outcomes until each patient’s respective death in the first 40 cycles, including outcomes after TD or PD. After death, the outcome was set to 0 until cycle 40 | Estimated means using a GEE with independence correlation structure, where cycle number was included as a categorical variable (Applying an LMM would also be possible here) | ||
3a,b | Death | PD, TD | All outcomes until each patient’s respective death in the first 40 cycles, including outcomes after TD or PD | 3a. Marginal means from an estimated LMM with random intercept, where the cycle number was included as a categorical variable 3b. same as 3a, but with additional random slope | ||
Figure 4 | ||||||
1 | Death, TD | PD | All outcomes before each patient’s respective TD or death in the first 40 cycles, including outcomes after PD | GEE with independence correlation structure, where cycle number was included as a categorical variable. (Averaging individual LMM predictions over those still on treatment at each cycle would also be possible) | ||
2a | Death, TD | PD | All outcomes before each patient’s respective TD or death in the first 40 cycles, including outcomes after PD | Marginal means from an estimated LMM with random intercept, where the cycle number was included as a categorical variable | ||
2b | Death | TD | PD | All outcomes before each patient’s respective TD or death in the first 40 cycles, including outcomes after PD | Estimating an LMM and averaging individual predictions at each cycle over those still alive at that cycle | |
3 | Death | PD, TD | Same as estimates 1a,b in Fig. 3 | |||
Figure 5 | The analyses for Fig. 5 below are analogous to those for Fig. 4 but with “TD” replaced by “PD” and vice versa | |||||
1 | Death, PD | TD | All outcomes before each patient’s respective PD or death in the first 40 cycles, including outcomes after TD | GEE with independence correlation structure, where cycle number was included as a categorical variable. (Averaging individual LMM predictions over those still without PD and alive at each cycle would also be possible) | ||
2a | Death, PD | TD | All outcomes before each patient’s respective PD or death in the first 40 cycles, including outcomes after TD | Marginal means from an estimated LMM with random intercept, where the cycle number was included as a categorical variable | ||
2b | Death | PD | TD | All outcomes before each patient’s respective PD or death in the first 40 cycles, including outcomes after TD | Estimating an LMM and averaging individual predictions at each cycle over those still alive at that cycle | |
3 | Death | TD, PD | Same as estimates 1a,b in Fig. 3 | |||