Bipolar disorder (BD) is associated with frequent hospitalizations, a high risk of suicide, and multiple comorbidities. BD pharmacotherapy is often administered by trial and error due to a lack of personalized treatment algorithms accounting for patient age, medical history, and concomitant drugs in use. In addition, BD medications are often poorly tolerated due to adverse effects, which leads to low adherence to treatment. Drug-related long-term comorbidities are still poorly understood, because adverse outcomes vary in their rates and severity, particularly in unique subpopulations. The main question expressed by patients is “What drug is best for me?” Systematic reviews and comparative effectiveness studies addressing this question are still lacking. Randomized clinical trials have provided evidence of limited external validity because they have often excluded youth, patients aged ≥65 years, and patients with mental/medical comorbidities; have had a short follow-up with relatively small sample sizes; and rarely have compared multidrug regimens. We set out to address these gaps in evidence using observational studies, selecting the outcomes most important to patients. Our patient partners identified several primary outcomes of interest for our research: the risks of psychiatric hospitalization (PH), self-harm, metabolic abnormalities, and kidney disorders (KDs). We sought to verify the importance of these outcomes, as well as identify other areas of concern for a larger group of patients, guiding study design and dissemination with iterative feedback.
Our specific aims were to (1) identify patient-centered outcomes of longitudinal BD treatment, (2) compare patient-centered outcomes for commonly prescribed BD medications, and (3) assess heterogeneity of treatment effects in youth (ie, aged ≤18 years) and in older patient (ie, aged ≥65 years) subpopulations.
To identify patient-centered outcomes, we conducted 3 focus groups with patients with BD and their relatives. Participants shared their feedback on the challenges faced while managing BD via questionnaires and prioritized their research questions via an affinity-mapping exercise. A causal systems model of BD challenges was built based on the data obtained, using a Theory of Constraints approach. A series of retrospective observational studies were then conducted using an administrative claims database to compare BD drug regimens for the risk of PH, self-harm, KDs, and diabetes mellitus (DM). Intermediate results from our comparative effectiveness studies were shared with participants of 6 subsequent focus groups, and patient feedback was obtained via questionnaires. The data obtained were used to interpret the findings, enhance the study methodology, and plan the information dissemination. During the last 3 focus groups, the study results were presented to youth and patients aged ≥65 years with BD, who shared their feedback on age-specific challenges of BD management. Our retrospective observational studies used the IBM MarketScan administrative claims database (2003-2016) containing information on 1.3 million commercially insured US patients with BD. Dozens of psychotropic drugs and drug combinations were compared for the risk of PH, self-harm, KDs, and DM among patients with BD. Lithium and a no-drug regimen were chosen as comparators. Survival analyses, employing Cox regression, competing risk regression, and time-fixed and time-varying covariates were used. Machine learning was applied to identify uncoded self-harm events in the database.
From a systems perspective, 2 core conflicts underlie most of the patient-reported challenges of BD management: taking medications vs not, and health care quality vs quantity. The patients' top research priorities were alternatives and adjuncts to pharmacotherapy and choice of pharmacotherapy. Drug-free options are of particular concern for youth, and drug physical safety is of concern for those aged ≥65 years. Compared with lithium, antidepressants and first-generation antipsychotics (FGAs) were associated with a higher risk of PH, whereas aripiprazole, valproate, and bupropion had a lower risk of PH. Multidrug pharmacotherapy was associated with an increased risk of KDs and DM. Lithium and monoamine oxidase inhibitors (MAOIs) were associated with a significantly higher risk of KDs than was no-drug treatment. The risk of DM was higher with antipsychotics than with no drug, but lithium, mood-stabilizing anticonvulsants (MSAs), and bupropion had lower DM risk. Compared with lithium, a lower risk of self-harm was associated with the use of lamotrigine, valproate, oxcarbazepine, olanzapine, risperidone, aripiprazole, bupropion, serotonin-norepinephrine reuptake inhibitors (SNRIs), and selective serotonin reuptake inhibitors (SSRIs). A higher risk of self-harm was associated with FGAs and some multidrug combinations. We also found that only 1 in 19 of imputed self-harm events were coded in the database and that most patients with BD discontinue monotherapy within 2 months after their treatment starts.
Relatively beneficial profiles for several BD outcomes were observed for valproate, aripiprazole, and bupropion. Our data support existing concerns about lithium nephrotoxicity and highlight the potential risks of MAOIs, MSAs, and antipsychotics for KDs. Antidepressants and MSAs are associated with lower self-harm risk in BD. The inconsistent drug adherence and poor self-harm coding identified in our study highlight areas where patient care and observational studies may be improved.
The assignment of patients to treatment groups was nonrandomized. Certain unmeasured indications could distort drug risk estimates (eg, family and medical history, ethnicity, and lifestyle). No correction was made for medication dosage, route of administration, or release mechanism (eg, long-acting injectables vs oral).
