OBJECTIVES: To systematically evaluate the impact of time horizon choice on the incremental cost-effectiveness ratio (ICER) under varying assumptions regarding treatment effectiveness, costs and discount rate. METHODS: We developed a Markov model, comparing two hypothetical strategies and predicting marginal quality adjusted life years (QALYs), costs and ICERs as a function of time horizon (ranging from 5 to 50 years). We assume that clinical trial data is available for 5 years of follow up. The model was analyzed for the following scenarios: extrapolation of survival benefit (optimistic, intermediate, conservative), prediction of treatment costs (one time costs only, constant continued incremental cost with/without cost decrease after 10 years), and discount rate (undiscounted, same discount rate for QALYs and costs, differential discount rate for QALYs and costs). RESULTS: The effect on ICER is greatest when comparing a 5 year to a 10 year or longer time horizons. For most scenarios the ICER does not change much when extending the time horizon beyond 15 years (less than 20% relative change in ICER).The choice of time horizon conditional on the extrapolation method of survival benefit (optimistic and intermediate scenarios) for one time cost scenarios had the greatest impact on reducing the ICER (reduction of >60% if using 10 year instead of 5 year time horizon). In scenarios with continued costs, the ICER was sensitive to a cost decrease after 10 years but less sensitive to the discount rate. CONCLUSION: Current guidelines suggest adopting a lifetime time horizon when a mortality benefit is present. Our study measures the bias associated with adopting the time horizon of a clinical trial, a common practice. Most susceptible to bias are scenarios with one time costs as the ICER is most sensitive to the chosen extrapolation method for survival benefits. The choice of time horizon and its impact on ICER warrant careful consideration.