OBJECTIVES: Existing studies exploring the validity of using mapping algorithms to predict utilities in external datasets have found mixed results. We apply a series of published EQ-5D mapping algorithms to individual patient level data from a trial which in fact collected this outcome, with a view to assess the impact of using predicted versus actual EQ-5D values on the results of a cost-effectiveness model. METHODS: The RITA-3 trial compared early interventional strategy for patients with non-ST-elevation acute coronary syndrome (NSTE-ACS) against a conservative strategy. EQ-5D data were collected at baseline, 4 months, 12 months and yearly thereafter. A range of other clinical and quality of life outcomes were also collected including the Seattle Angina Questionnaire (SAQ), SF-36 and the Canadian Cardiovascular Society (CCS) Functional Classification of Angina. Using mapping algorithms found in the published literature, we predicted EQ-5D scores using these outcome measures. We then explored what effect the predicted utilities had upon cost-effectiveness estimates in a model compared to those with the actual EQ-5D data. The comparisons were made across five patient subgroups. RESULTS: The cost-effectiveness estimates varied according to the original outcome measure used to map and the patient subgroup. The EQ-5D values predicted using CCS scores produced cost-effectiveness estimates much higher than those with the actual EQ-5D data in the trial, whilst the estimates using EQ-5D values predicted using SAQ scores were lower. The values predicted using SF-36 scores gave cost-effectiveness estimates very similar to those with the EQ-5D data, irrespective of the patient subgroup. CONCLUSIONS: Analysts should take caution when mapping EQ-5D values from algorithms that have not been externally validated, especially where these algorithms have used clinical outcomes or disease-specific measures of health-related quality of life. Our results suggest that mapping from generic outcome measures might be reasonable.