OBJECTIVES: Information about acute myeloid leukaemia (AML), including the costs of treatment and survival-estimates, are usually derived from clinical trial data. However, it is not known whether this information is generalizable to non-trial patients. This study was carried out to evaluate the differences in medical costs and survival between trial and non-trial patients with AML; and hence assess the external validity of trial data to the general patient population. METHODS: The Haematological Malignancy Research Network (HMRN, www.hmrn.org) is an established population-based patient cohort that registers around 2000 newly diagnosed patients each year. All adults (≥18) newly diagnosed with AML between September 2004 and August 2007 and treated with induction intent were included.  Patients were followed until August 2012, and the comparative outcomes were medical costs and survival. Standard statistical analyses were used to measure unadjusted difference in outcomes, and propensity score analyses were applied to measure differences by adjusting for baseline imbalance in pre-treatment characteristics between trial and non-trial patients. RESULTS: Overall, 173 patients treated with induction intent were included, of which 106 were trial and 67 non-trial. Trial participation was associated with younger age, fewer comorbidities, better prognosis, and being treated at teaching hospitals. Before controlling for patients’ characteristics, trial patients had better survival and incurred higher costs (p<.0001 for both). After controlling for patients’ characteristics by carrying out propensity score analyses, these differences remained significant in both survival (median survival 28.7 vs. 8 months; p<.0001) and medical costs (mean costs £84,497 vs. £49,624; p<.0001). CONCLUSIONS: For AML patients treated with induction intent, significant differences were observed in treatment costs and survival according to trial status, both before and after controlling for patients’ pre-treatment characteristics. Data generated solely from clinical trials may therefore not be generalizable to non-trial patients and should be treated with some caution when used to facilitate decision-making.