OBJECTIVES: To compare the economic results of a non-randomized study using propensity score (PS) methodology to adjust for confounding versus a restriction approach based on clinical opinion. METHODS: We used data from a published non-randomized study, which enrolled 195 patients (58.2%) to receive open surgical repair (OSR) and 140 patients (41.8%) to receive endovascular aneurysm repair (EVAR) for the treatment of abdominal aortic aneurysm. OSR patients were classified as being at low risk (LR) or high risk (HR) for post-surgical complications based on clinical opinion and scoring algorithms while all EVAR patients were classified as HR. The database included baseline characteristics, patient level 1-year cost and survival data. One-to-one PS matching was used within the full population to select a more balanced patient sub-population. Incremental cost-effectiveness ratios (ICERs) were assessed within the HR sub-population and the PS-matched sub-population. RESULTS: The HR sub-population was composed of all 140 EVAR patients (100.0%) and 50 OSR patients (25.6%). EVAR was identified as the dominant treatment option within the HR sub-population. The PS-matched sub-population was composed of 77 HR EVAR patients (55.0%) and 77 OSR patients (39.5%), of which 48 (62.3%) were OSR-LR patients and 29 (37.7%) were OSR-HR patients. Unlike results obtained within the HR sub-population, the ICER of EVAR was estimated at $93,608 per life-year gained within the PS-matched sub-population. Differences in the results may be explained by confounding; although balance was improved within the PS-matched sub-population, unbalance remained on several patient characteristics. CONCLUSIONS: Results of this study highlight the fact that PS matching may not always fully adjust for confounding. Clinical opinion may be influenced by unmeasured confounders which may not be adjusted for by PS matching. Balance within patient subsets following PS matching must be evaluated when conducting economic evaluation based on non-randomized studies, especially in studies with small sample sizes.