OBJECTIVES: To apply Bayesian approaches to evaluate the impact of time-delay on the cost-effectiveness of primary angioplasty versus hospital-administered thrombolysis for patients with acute myocardial infarction. METHODS: A probabilistic model was developed to evaluate the life-term cost-effectiveness of primary angioplasty. A health service perspective was adopted with outcomes estimated using Quality-Adjusted Life Years (QALYs). Evidence on short-term event rates was obtained by updating a recent meta-analysis of randomised trials comparing angioplasty with thrombolysis (Lancet 2003; 361:13-20). Data from 22 trials were combined using Bayesian hierarchical modelling and meta-regression. This approach enabled the simultaneous estimation of posterior distributions and correlation structure for: 1) major cardiovascular events including death, re-infarction and stroke, and 2) different time endpoints (4-6 weeks and 6 months). The impact of PCI time-delay to treatment was analysed using mean time delay compared to thrombolysis as a covariate of the random effects model. Long-term costs and QALYs were estimated using a Markov model populated from UK registry data. RESULTS: Based on the average time-delay reported in the trials (54 minutes), primary angioplasty resulted in a mean gain of 0.29 QALYs and an additional cost of £2,680 compared to thrombolysis. The associated incremental cost-effectiveness ratio (ICER) was £9,241 per QALY. At a threshold of £20,000 per QALY there was a 90% probability that primary angioplasty was cost-effective. Adjusting the time-delay resulted in considerable variation in the cost-effectiveness estimates. For a shorter delay of 30-minutes the ICER was £6,850 per QALY; increasing the time-delay to 90-minutes resulted in a marked increase in the ICER to £64,750 per QALY (98% and 36% probability cost-effective at £20k, respectively). CONCLUSIONS: This study demonstrates the policy importance of time-delay when considering the cost-effectiveness of primary angioplasty, and the advantages of using Bayesian approaches to model multiple endpoints, treatment effects and baseline event rates.