OBJECTIVES: Bipolar disorder (BD) is a chronic disease associated with several medical and psychiatric comorbidities that can have serious economic impact. We aimed to identify, describe and critically assess health economic studies in BD-I in the US during the last ten years and provide recommendations for future researchers. METHODS: We searched MEDLINE, EMBASE, NHS EED and ISPOR’s database to identify articles/posters with economic evaluations (cost-effectiveness-CEA, cost-utility-CUA, cost-minimization-CMA, cost-benefit-CBA and budget impact analysis-BIA) performed on BD-I patients. We collected information on: population, intervention/comparator, setting, modeling, data quality, clinical/economic outcomes and uncertainty analysis. For quality assessment, two checklists (Drummond/Philips) recommended by the Cochrane Collaboration were applied. RESULTS: Six studies were included (4 articles, 1 poster, 1 budget impact model-BIM) covering various medications alone or combined (lamotrigine, lithium, olanzapine, quetiapine, aripiprazole, lurasidone and risperidone) in several disease settings (overall maintenance, maintenance after stabilization or with recent manic episodes, acute depression or hospitalization following acute mania). Regarding study types (type of modeling) we found 4 CEA/CUA (Markov model), 1 CEA (decision tree) and 1 BIA (discrete event simulation). Measures of benefit included: number of acute episodes, euthymic days, QALY, remission rate and utilities. Four CEAs used ICER as economic outcome and one used Net Benefit Analysis (NBA). Quality assessment showed medium-to-high risk of bias, and very weak clinical bases supporting analyses. CONCLUSIONS: Further economic studies on BD-I should take into consideration the main comparators used in real-life. When no head-to-head comparison exists, researchers should perform systematic review and possibly network meta-analysis. Since BD-I is a chronic disease, a lifetime framework should be considered for time horizon. Parameters from previous publications should be used carefully, since many of them may be invalid. Economic analyses could be developed further to include other treatment options and Markov model should cover subsequent treatment lines.