Purpose: Although cost-effectiveness analyses (CEA) of two competing therapies are often based on a lifetime horizon, shorter analytic time horizons are increasingly common because of a growing focus on evidence from clinical trials, non-societal third party payer perspectives, a desire to minimize regulatory hurdles, and practical considerations. Short analytic time horizons may lead to different results. Our aim was to explore the effects of varying the time horizon of analysis on incremental cost-effectiveness ratios (ICERs).

Methods: We compared two hypothetical therapies, "new" and "usual", over a range of analytic time horizons. In each case, we assumed that "new" had a lower long-term mortality rate. We developed three models to encompass common clinical situations: Model 1: "new" having higher cost but no initial procedural risk or cost; Model 2: "new" having higher or lower cost, a substantial procedural cost but no procedural mortality; Model 3: "new" having either higher or lower cost and both an initial procedural mortality and cost. We modeled cohort survival with Gompertz functions and then calculated cohort costs for each therapy. For each model, differences in cumulative cohort costs and life expectancies were compared to obtain ICERs for different analytic time horizons and discount rates.

Results: Even for the same underlying assumptions about survival and cost, ICERs change considerably for different analytic time horizons. In Model 1, as the incremental benefit and cost of “new” therapy progressively increased over time, decreasing ICERs with increasing time horizons were often seen. In Model 2, the procedural cost of the “new” therapy produced ICERs that were quite high initially but fell sharply and monotonically with increasing analytic time horizons. In Model 3, when both procedural risks and costs for the “new” therapy were high, an incremental benefit was realized only with analytic time horizons long enough to outweigh the procedural mortality. The “new” therapy was dominated for short analytic time horizons; ICER sometimes increased with a long analytic time horizon. Sensitivity analyses showed increasing discount rates attenuated these effects.

Conclusions: ICERs can change substantially with differences in the analytic time horizon. Therefore, even when the underlying clinical evidence has limited duration, CEAs must be explored over a spectrum of time horizons by projecting survival and cost beyond the evidence-based trial data.