OBJECTIVES: To investigate the consequences of medication non-adherence on relapse, recurrence, remission and recovery, and to estimate costs associated with non-adherence over 21-months in the treatment of bipolar disorder following a manic or mixed episode. METHODS: EMBLEM is a prospective, observational study on patient outcomes with a manic/mixed episode. Data was collected at baseline, during the first 12 weeks of treatment (acute phase), at 6 month post-baseline and at 6-month intervals up to 24 months (maintenance phase). Medication adherence was assessed at each visit by investigators as: i) not prescribed medication; ii) almost always adheres; iii) adheres half of the time; and iv) never adheres to medication. Adherence was defined as always answering ‘always adheres’ and non-adherence was defined when one or more response of ‘adheres half of the time’ or ‘never adheres’. In this post-hoc analysis, Cox proportional hazards models with non-adherence, adjusted for patient characteristics, investigated its impact on clinical outcomes. Multivariate analyses were performed to estimate the cost of resource use associated with non-adherence during the maintenance phase, using a log-link function. UK unit costs were applied to resource use. RESULTS: Of the 1341 patients analysed, 23.6% were non-adherent over 21-months. Non-adherence was significantly associated with a higher risk of relapse (HR:2.40, 95%CI:2.04-2.83) and recurrence (HR:1.72, 95%CI:1.27-2.33) as well as lower remission rates (Hazard ratio (HR):0.71, 95%CI:0.59-0.86) and recovery (HR:0.65, 95%CI:0.51-0.81). In addition, costs incurred by non-adherent patients during this period were significantly higher than those of adherent patients (£10231 vs. £7379). This disparity resulted mainly from differences in in-patient costs (£4796 vs. £2150). CONCLUSIONS: Non-adherence in bipolar patients after a manic/mixed episode is associated with higher than twice the risk of relapse of adherent patients which has economic implications for healthcare providers. A study limitation is that adherence was assessed by investigators using a single-item measure.