This study extends findings of a previous short-term retrospective cohort study that reported reduced risk of hospitalization and lower psychiatric treatment costs of patients with bipolar disorder treated with mood stabilizer and adjunctive aripiprazole compared to adjunctive ziprasidone, olanzapine and quetiapine during a 90-day follow-up period [13
]. In this 1-year follow-up study, risk of hospitalization was lower in patients treated with aripiprazole with or without mood stabilizer compared to ziprasidone, olanzapine and quetiapine. Duration of therapy on atypical antipsychotic therapy was comparable across all atypical antipsychotics in this study, although the duration was brief relative to the follow-up period, lasting less than 3 months in 75% of cases. However, treatment guidelines recommend regimen simplification after patients are stabilized [7
]. Therefore, in our sample, it is possible that the short duration of atypical antipsychotic therapy reflects stabilization of patients that allowed discontinuation of the atypical antipsychotic. Gianfrancesco et al. found somewhat longer treatment durations in a study of commercially insured patients treated with antipsychotics, with treatment durations of 7-10 months [19
]. However, they allowed a gap of up to 120 days before ending a treatment episode, whereas our threshold of 15 days was much more conservative. To allow for meaningful comparative analysis of the cost data, intent-to-treat analysis was conducted for the cost analysis and patients were followed up for 1 year after their initial atypical antipsychotics treatment.
Antipsychotic doses observed also tended to be lower than label-recommended doses and demonstrated little titration over the course of treatment. These observations are consistent with other reports on atypical antipsychotic dosing in bipolar disorder [20
]. Although we are not able to determine the reasons for these dosing patterns, it is possible that, due to concerns regarding tolerability or safety, physicians were reluctant to start patients on higher doses.
Along with the lower risk of psychiatric hospitalizations associated with aripiprazole compared to three of the four comparators, patients who initiated aripiprazole had lower psychiatric inpatient costs. These results suggest that treatment with aripiprazole tends to provide a valuable cost-offset in saving from decreased hospitalization risk and associated inpatient costs. In particular, the lower risk of hospitalization combined with lower total costs compared to quetiapine represent two attractive outcomes for formulary decision-makers responsible for the entire costs of care [22
Observational studies can provide important insights into the outcomes of clinical practice in real-world settings, where dosing, titration and concomitant medications are not constrained by trial protocols. Such studies evaluate the effectiveness of treatments as they are actually used rather than when optimally dosed. We included the full range of observed dosing in our analysis based on the assumption that, in selecting medications, physicians also use what they believe is the most appropriate dosing and titration for that medication.
This study has several limitations. As a non-randomized retrospective study of observational data, it is possible that despite the use of propensity score matching and multivariate modeling, unobserved treatment selection bias may confound the results. Propensity score matching, however, is a widely accepted method for minimizing the effects of treatment selection bias in observational data [15
]. Other approaches such as instrumental variables and Heckman's sample selection bias method may also be used in such settings [23
], although the potential for residual confounding remains with all such methods. The consistency of our results in propensity score-matched and unmatched samples suggests that these findings are robust. However, the dataset we analyzed consists of patients from a single commercial health plan; results may not be applicable to chronic populations that are more likely to be covered by public sector insurance. Replication in other observational datasets is necessary to validate the robustness of these results.
Additionally, by restricting the analysis to an inception cohort, we were only able to study the effects of the initial choice of medication following an antipsychotic-free period and are thus limited to conclusions on initial antipsychotic selection rather than the effectiveness of a given medication under all circumstances. Based on our results, aripiprazole appears to be the most effective initial choice among atypical antipsychotics for the acute treatment of bipolar disorder, and these effects appear to persist in the post-acute phase. Finally, the study only followed patients until their first psychiatric hospitalization and did not address outcomes following adding, switching, or discontinuing antipsychotics, which may be common in this population. The analysis of such complex treatment patterns within claims data may be subject to high levels of unobservable confounding and difficult to interpret with respect to the contribution of individual medications across complex regimens. Specifically, it may be challenging to account for residual effects of prior medications following a switch, which is why we chose inception cohort design. Moreover, the reasons for adding versus switching antipsychotics would require detailed clinical information not available in this dataset to adjust for treatment selection bias. Therefore, our results are limited to outcomes only while the patient is on their initial antipsychotic medication for that episode of treatment.