Table 1.
Setup of the simulation study base case and scenarios
| Characteristic | Base-case value | Rationale | Scenario analysis value(s) |
|---|---|---|---|
| Study design settings | |||
| Number of patients in study | 300 | 300 patients is approximately what has been seen in immune-oncology studies to date (though this does vary) |
150 (scenario 1) 500 (scenario 2) |
| Cohort age, years | 65 | The approximate age of patients enrolled in to contemporary immunotherapy studies |
55 (scenario 3) 75 (scenario 4) |
| Male: female ratio | 1:1 | Although the gender ratio in studies is driven by the prevalence of conditions. In the simulation study, however, this only affects background mortality so is not varied | |
| Utility measurement interval | 120 days | Utilities are usually measured at increasing intervals over time, for simplicity a uniform pattern has been imposed |
90 days (scenario 5) 180 days (scenario 6) |
| Administrative censoring for utility values | 48 months for all patients | Utilities are generally only collected until the end of the study period. A ‘typical’ data collection period has been used, which is varied in sensitivity analysis to include other observation periods seen in trials |
18 months for all patients (scenario 7) 60 months for all patients (scenario 8) Until progression or maximum 60 months (scenario 9) Until 30 days after progression or maximum 60 months (scenario 10) |
| Missing data | 0% | Missing data can be an issue in clinical studies. In the base case this is assumed to be zero, with different mechanisms for missingness explored in sensitivity analysis |
10% of observations MCAR (scenario 11) 10% of patients lost to follow up at a random timepoint (all subsequent data censored; permanent MCAR) (scenario 12) Increasing likelihood of censored values as utility decreases (MNAR) (scenario 13) Censoring probability linked to time to death (scenario 14) |
| Survival simulation | |||
| Ratio of patients exhibiting poor/intermediate/background survival | 13:7:4 | In immune-oncology studies a number of patients have experienced durable survival, this proportion however varies between studies |
13:7:0 (scenario 15)—no long-term survivors 13:7:2 (scenario 16)—a lower rate of long-term survivors 13:7:6 (scenario 17)—a higher rate of long-term survivors |
| Time to progression for patients with poor outcomes (months) | Gamma (shape = 3, scale = 1) | Immuno-oncology studies exhibit a changing hazard over time with a short period on enrollment before many progression and survival events are observed, which decrease in frequency over time, with few being observed beyond 18 months [3] | |
| Time to progression for patients with intermediate outcomes (months) | Weibull (shape = 1.3, scale = 8) | ||
| Post-progression survival | Weibull (shape = 1.5, scale = 14) | ||
| Percentage of deaths pre-progression | 20% | ||
| Pseudo-progression | 0% | A known issue with immuno-oncology is that the immune response can lead to swelling, which may be (incorrectly) categorized as disease progression. Whilst new measures have been developed to account for this, the impact is explored in sensitivity analysis where a portion of patients are miscategorized for regressions as having PFS as per the intermediate group | 10% of long-term survivors incorrectly assumed to have progressed in line with the patterns seen for other groups (scenario 18) |
| Link between pre- and post-progression survival | Independent distributions | The assumption is made that response to treatment, and post-progression survival are uncorrelated i.e. patient characteristics are not both predictive and prognostic | A scenario analysis (scenario 19) is presented where simulated post-progression survival is multiplied by 1.25 for long-term survivors, and 0.75 for short-term survivors. This implicitly assumes responders to treatment are healthier patients |
| Utility simulation | |||
| Patient utility distribution before progression or being close to death | Beta (α = 80, β = 20) i.e. mean 0.80, quartiles 0.77, 0.80, 0.83 | In line with the literature on utilities which show reasonably high levels pre-progression, falling on disease progression [18] | |
| Progressed utility (in progression scenarios) distribution | Beta (α = 60, β = 20) i.e. mean 0.75, quartiles 0.72, 0.75, 0.78 | ||
| Time at which utility fell before death (in time-to-death scenarios) | Uniform (minimum, 90 days; maximum, 270 days) | Various observations have been reported in the literature, and thus a range is used which varies by scenario | |
| Utility fall before death (distribution) | Normal (mean, 0.5; SD, 0.2) | The absolute fall seen in studies have differed, but all have been substantial | |
MCAR missing completely at random, MNAR missing not at random, N number of patients, OS overall survival, PFS progression-free survival