| To Register SMDM Homepage | |
|
|
|
||||
Purpose: To develop methods to undertake expected value of sample information (EVSI) calculations for 2 treatment survival trials using cost-effectiveness modelling.
Method: Cox proportional hazard methods are common to evaluate the relative risk of survival, but the approach is non-parametric, and we sometimes need to extrapolate baseline survival for cost per QALY calculations. Within an illustrative (but representative) model, we use a Weibull survival curve for prior baseline survival (bivariate lognormal to characterise uncertain Weibull parameters) and a lognormal distribution for prior relative risk. Monte Carlo simulation of notional matched pairs of individual level trial participants is undertaken given this prior uncertainty (sample size = n). Data are censored for a pre-specified duration of follow-up d. For each simulated trial dataset, Bayesian updating of the Weibull and lognormal relative risk probability distributions enables probabilistic sensitivity analysis. EVSI are undertaken using Laplace approximation. Five alternative follow-up durations and sample sizes are examined.
Results: The illustrative model results shows that for a sample size of n=100, and a duration of follow-up of 1 year, EVSI=£3.2m. For n=100 and 2 years follow-up, EVSI=£3.9m. For n=1,000 and 4 years follow-up, EVSI=£4.9m. Other combinations show the trade-offs between sample size and follow-up duration.
Conclusions: The methods described provide an approach to undertaking EVSI calculation to design survival trials. Such methods could be used alongside traditional sample-size calculations to compare results in practice. Further work is needed on the issue of possible correlation between baseline survival and relative risk.
See more of Poster Session - Public Health; Methodological Advances
See more of The 26th Annual Meeting of the Society for Medical Decision Making (October 17-20, 2004)