Purpose: In the assessment of cardiovascular disease prevention strategies variation between individuals in their susceptibility to disease (called “frailty”) is rarely considered, even though it may have a substantial effect on the cost-effectiveness of such strategies. While this effect cannot be estimated using standard survival models it is unclear whether or not it can be estimated accurately with frailty models that account for heterogeneity at population-level.

Methods: In a simulation study we assessed a cholesterol lowering therapy that was cost-effective for individuals with a 10-year cardiovascular risk exceeding 20%. Framingham risk equations were used and frailty was introduced using Gamma, Uniform and Non-parametric distributions, with high and low variation (100 simulations of six datasets). Three frailty models, Weibull-Gamma, Weibull-Uniform and Weibull-Non-parametric, were estimated on each simulated dataset. With these models we determined the expected number of suboptimal treatment decisions and the expected loss of net monetary benefit (NMB) caused by the heterogeneity. Results were compared with the actual number of suboptimal treatment decisions and NMB loss, derived using a hypothetical perfect model.

Results: Using the perfect model we found that heterogeneity with high (low) variation caused 14% (8%) of all decisions based on the frailty models to be suboptimal, regardless of frailty distributions. Consequently, the actual NMB ranged from 34-60% of the NMB without heterogeneity. Using the frailty models to estimate the number of incorrect allocation to (non-)treatments and the NMB loss results were, on average, similar, but the varied per simulation. The estimated number of suboptimal allocations was 1.03 (95% CI: 0.79-1.30) times the actual number for the Weibull-Gamma model (Weibull-Uniform, 0.93, 95% CI: 0.63-1.37, Weibull-Non-parametric, 1.07, 95% CI: 0.55-1.87) and did not depend substantially on the frailty distributions used. The NMB loss caused by heterogeneity was best estimated with the Weibull-Gamma model and was 1.05 (95% CI: 0.84-1.23) times the actual NMB loss.

Conclusions: The extent to which the cost-effectiveness of disease prevention strategies is lowered by the presence of heterogeneity can be assessed fairly accurately using frailty models. The Weibull-Gamma model provides the best performance for different distributions of heterogeneity in the target population. It could be used in practice to determine for which, increased, risk threshold the expected cost-effectiveness with heterogeneity becomes equal to the expected cost-effectiveness without heterogeneity.