Personalized Pharmacotherapy for Bipolar Disorder: How to Tailor Findings From Randomized Trials to Individual Patient-Level Outcomes

RCTs yield information about treatment outcomes for diagnostic or subdiagnostic groups as a unit, but the practitioner almost always must extrapolate such findings to the unique clinical circumstances of a given patient. A key concept in this regard involves the notion of enrichment of patient groups enrolled in clinical trials, meaning that study participation may be restricted to individuals only with certain characteristics—such as those who responded acutely to a particular medication before entering a relapse prevention trial—or study participation may be confined to a specific age range, or to patients with bipolar disorder without rapid cycling or active suicidal features, active substance misuse, or psychiatric comorbid conditions. Ironically, most patients treated for bipolar disorder in routine practice are ultimately deemed ineligible to participate in RCTs. Screen failure rates for intervention trials for bipolar mania range from 55% to 96% (11)—most often because of active substance misuse or psychiatric comorbid conditions, suicidality, poor adherence, or failure to meet full diagnostic criteria; this suggests that most individuals identified as having bipolar disorder in the community differ in fundamental phenomenological ways from their RCT-participant counterparts. Practitioners and investigators alike then wonder why “real-world” patient outcomes often differ markedly from those that might otherwise be expected from the evidence-based literature.

One way in which clinicians and readers of the literature can better distinguish similarities and differences between RCT participants and their own clinical populations—and thereby anticipate how much extrapolation may be needed from the study group to the individual—is to understand and recognize clinical and biological characteristics that increase the likelihood of a favorable treatment response for a given patient. In conducting clinical trials, investigators have begun to recognize the need to distinguish individual from group treatment responses and are increasingly using refined data-analytic approaches that pool patient data to allow for participant-level analyses (12). Such studies usefully reveal factors about an overall study population that may increase or decrease the likelihood of individual participants to respond to an intervention. For example, one such analysis of olanzapine for the treatment of bipolar mania found especially robust separation between drug and placebo among patients with high baseline symptom severity (13). Hopefully, the concept of “personalized medicine” in psychopharmacology will validate possible biomarkers (e.g., gene expression products and proteomics, pharmacogenetics, and inflammatory cytokines) as well as clinical characteristics associated with drug responsivity for a given patient. At present, clinical features can be identified and clustered to devise risk profiles that may favor drug response or that parsimoniously target heterogeneous phenotypes (e.g., one drug targeting multiple effects). Clinicians who follow the empirical literature can formulate their own hypotheses about tailoring a treatment regimen that best matches the unique profile of a given bipolar patient on the basis of their own experience relative to findings from RCTs.

The search for patient features that might influence the chances of a particular result for one patient relative to another involves the statistical concept of moderators and mediators of outcome. As described by the biostatistician Helena Kraemer (14, p. 672), moderators are baseline characteristics that identify “on whom or under what conditions treatment choice differentially affects outcomes” (i.e., who should get the treatment), whereas mediators are factors that suggest “how or why the treatment condition might be preferred to the control condition in the population sampled” (i.e., factors that could either facilitate or jeopardize the best possible treatment outcomes after the treatment has begun). Examples of well-recognized moderators and mediators of pharmacotherapy outcomes in RCTs for treatment of bipolar disorder are summarized in Table 1 and Table 2 (9, 15–31).

Our field came to appreciate the disparity between treatment outcomes for homogeneously “ideal” patient profiles from the greater heterogeneity (and more variable prognosis) of those in real-world settings through observational studies with lithium in the 1980s. From decades of experience since lithium came into use for bipolar disorder, one can now construct a “profile” of the idealized patient best suited to respond to lithium. Rather than consider lithium to be the gold-standard mood stabilizer, as often described (as if it possessed broad universal properties for most forms of bipolar disorder), one can probably think of it more accurately as a well-studied but narrow spectrum agent. Analogous to penicillin, lithium works best in a definable, narrow subgroup of bipolar patients (<10% by some reports (32)) with a distinctly uncomplicated clinical profile, defined by the following characteristics: pure euphoric more than mixed features (33); positive family history of lithium-responsive bipolar disorder (31); introduction within the first few lifetime episodes (34); absence of rapid cycling (19); better prophylaxis against recurrent mania than depression (35); mania polarity “proneness” (16); episode patterns in which manias precede depressions (36); higher age at illness onset (36); absence of somatic comorbid condition (32); absence of substance use comorbidity (37); hyperthymic temperament (38); absence of psychosis or cognitive disorganization (39); and absence of anxiety (40).

Additionally, lithium may hold a relative advantage over other mood-stabilizing agents in settings where prevention of suicidal behavior is a prominent concern (41), as a countermeasure to leukopenia (e.g., during clozapine therapy), and in patients without psoriasis or other chronic dermatological comorbid conditions.

Contrary to early expectations based on studies of divalproex and carbamazepine in the late 1990s, there does not appear to be a “class effect” among anticonvulsant drugs with respect to putative “mood-stabilizing” properties. Indeed, as summarized in Table 3, the mood-specific effects of most newer generation anticonvulsants, as established in RCTs, are narrowly confined to divalproex and carbamazepine (predominantly antimanic>antidepressant) and to lamotrigine (predominantly antidepressant>antimanic and more potent for relapse prevention than the treatment of acute episodes). Check marks in the table indicate robust, well-established efficacy. No positive, large-scale, placebo-controlled trials in any phase of bipolar disorder exist with any of the other anticonvulsants listed in the table. However, anticonvulsant drugs show a diverse range of psychotropic properties; some exert beneficial effects on manic or depressive symptoms, whereas others may cause worsening of affective symptoms (e.g., new-onset depression has been observed in up to 10% of patients with epilepsy taking topiramate (42) or adverse cognitive effects from, notably, topiramate or zonisamide). Other psychotropic effects in addition to those described in Table 3 include the following: established antinocioceptive properties, particularly neuropathic pain (e.g., gabapentin, pregabalin) or trigeminal neuralgia (e.g., carbamazepine); weight loss and appetite suppression (e.g., zonisamide [43]) and reversal of psychotropic-induced weight gain (e.g., zonisamide [44, 45]); treatment of sleep apnea via carbonic anhydrase inhibition (e.g., zonisamide [46]); treatment of insomnia and promoting sleep continuity (e.g., gabapentin prolongs stage 3 sleep); treatment for restless legs syndrome (gabapentin enacarbil); and a possible added benefit of adjunctive topiramate for bipolar disorder with comorbid obsessive-compulsive disorder (OCD [47]).

One could readily envision crafting a tailored pharmacotherapy regimen for a given patient with bipolar disorder that incorporates certain anticonvulsants meant to target specific illness characteristics or comorbid conditions such as those described in Table 3 (48–73)—without conflating such “ancillary” effects with the core targeting of affective symptoms. Illustrative examples are provided at the end of this article.

Second-generation antipsychotics (SGAs) have gained increasing attention over the past 2 decades as having psychotropic effects independent of their antipsychotic properties, including various degrees of antimanic; antidepressant; anxiolytic; and, possibly, procognitive effects in patients with bipolar disorder. All SGAs that have been studied in patients with acute mania in combination with lithium or divalproex have shown greater efficacy than those seen with mood stabilizer monotherapy. Additionally, secondary analyses of RCTs for treatment of acute mania with virtually all SGAs show greater efficacy than those with placebo, regardless of the presence or absence of rapid cycling. RCTs have demonstrated several additional psychotropic effects of certain SGAs, including the following: a beneficial effect for global cognitive functioning (74); possible weight loss effects with adjunctive ziprasidone (75); adjunctive use of aripiprazole for comorbid bipolar disorder and OCD (76); the superiority of aripiprazole over placebo in treating depression, anxiety, and aggression in people with borderline personality disorder (77); and the use of D2 and D3 partial agonists (notably, aripiprazole) to counteract antipsychotic-associated hyperprolactinemia.

The full summary is provided in Table 4 (78–99).

RCTs have not, as yet, demonstrated the value of SGAs in treating comorbid alcohol abuse and dependence (100) or substance use disorder symptoms, either in general or in people with bipolar disorder (101). Impulsive aggression also has not been demonstrated as an independent dimension of psychopathology that is responsive to SGAs in bipolar disorder, apart from the use of some short-acting injectable SGAs to treat acute agitation in emergency departments or inpatient settings; nor have RCTs shown value for SGAs in the treatment of anorexia nervosa (102).

It should be noted that some of the negative or failed trial findings identified in Table 4 may be artifacts of particular study designs. For example, in the negative findings of two RCTs, aripiprazole monotherapy (78), in the treatment of acute bipolar depression, ultimately failed to separate from placebo at the end of each study, not because of a weak drug effect but because of an unexpected improvement among placebo recipients near the conclusion of each trial. The relapse prevention trials of aripiprazole oral (79) or long-acting injectable (80) formulations, or of risperidone long-acting injectable (86) formulations, each began after index episodes of manic or mixed episodes, thereby enriching the samples for a higher propensity toward mania than toward depression relapse. In other words, the study designs used in these trials may have failed to reveal the fullest potential of these agents for preventing bipolar depressive episodes that might have provided more evidence using different study designs (with enrollment following index-depressive rather than solely manic or mixed episodes).

Clinicians sometimes combine medications without necessarily knowing whether there exists an evidence base to support the value of some combinations other others. Table 5 summarizes known pairings of psychotropic drugs in which combination therapies were superior to monotherapies plus placebo, across various illness states in bipolar disorder (103–108).

Still a topic of intense debate on which even experts cannot always agree (109), the issue of whether and when to use antidepressants in bipolar disorder plagues psychopharmacologists for three main reasons.

The first reason is questionable efficacy. No large-scale RCT has ever shown that the use of an antimanic drug plus any antidepressant is superior to using an antimanic drug alone for treating acute bipolar depression. As a class, antidepressants have a disturbingly high “number needed to treat” (NNT) of 29 for bipolar depression (110), casting doubt on the reliability with which they might deliver a desired effect.

The second reason is risk for induction of mania or cycle acceleration. Often, this phenomenon, listed in manufacturers’ package inserts for all antidepressants, is presumed by many clinicians to be the greatest hazard posed by antidepressants in the treatment of bipolar disorder. Over the course of its editions, the Diagnostic and Statistical Manual for Mental Disorders (DSM) has varied in its conceptualization of mania or hypomania that arises in the aftermath of antidepressant initiation, as well as the likelihood with which such an outcome represents the unmasking of a bipolar diathesis versus simply an adverse drug effect. In the DSM-5 (111), treatment-emergent mania or hypomania is considered to reflect the unmasking of a bipolar diathesis if the symptoms of mania or hypomania persist at a fully syndromal level after the suspected catalyst has been stopped and its physiological effects are no longer present (e.g., after five elimination half-lives). Although this definition is sensible for differentiating antidepressants as a cause (producing an adverse effect) versus a catalyst, it fails to take into account the reality that clinicians often will intervene with an antimanic pharmacotherapy (rather than wait and observe for the dissipation of mania as simply a drug side effect), and still others may continue an antidepressant despite the emergence of mania symptoms on the basis of the belief that antidepressants are necessary to address concurrent depressive symptoms. In clinical studies, judgments about alleged antidepressant-associated mania are further hampered by inaccuracies in retrospective recall, subjective impressions about whether diagnostic criteria are actually met for a manic or hypomanic syndrome (as contrasted with more nebulous phenomena such as “activation”), and distinguishing the natural course of illness from a true causal association. (On the latter point, some experts consider the possible association of emerging mania-hypomania symptoms with antidepressant exposure-dose increases to be “definite” if within 8 weeks and “likely” or “possible” if within 12 weeks; beyond that time frame, differentiation from the natural course of illness becomes virtually impossible (112).

Across studies, the risk for antidepressant-induced mania in bipolar disorder appears to be much lower than many clinicians assume: about 10%–15%, with a “number needed to harm” (NNH) of about 200 (110), making the anticipated lack of efficacy a greater hazard than the induction of mania for most depressed patients with bipolar disorder who take a traditional antidepressant.

The third reason involves additional limitations about forming hard and fast conclusions about antidepressant safety and efficacy in treating bipolar disorder. These include the following: Most contemporary antidepressants have not been studied in the treatment of bipolar depression, with no existing placebo-controlled studies of serotonin and norepinephrine reuptake inhibitors (SNRIs), vilazodone, vortioxetine, mirtazapine, or escitalopram; no controlled trials exist that examine the safety and efficacy of antidepressants in treating bipolar disorder for purposes other than treating depression (e.g., OCD, specific anxiety disorders, or bulimia); it is largely unknown whether treatment-emergent manic or hypomanic symptoms arise as a dose-dependent or all-or-none phenomenon; and, finally, the value of coadministering an antidepressant with an antimanic mood stabilizer (advocated in many practice guidelines as a safety measure to minimize the risk for induction of mania-hypomania) is largely an unproven hypothesis that has, nevertheless, become widely accepted as if dogmatic.

The existing database of RCTs points to an identifiable clinical profile of individuals with bipolar disorder for whom antidepressant use may be more or less favorable and safe, as summarized in Table 6 (113–121). A further consideration, true of psychopharmacology in general, involves an individual patient’s personal history of response to a particular drug. In the world of evidence-based medicine, establishing an unequivocal and sustained response to a specific medication for a given individual is incontrovertible evidence of its effect on that given individual. Someone’s known own favorable response to an antidepressant, even if to an agent that has no established database in the RCT literature, constitutes evidence for that one patient. However, that information cannot generalize or transfer to other individuals on the basis of the experience of one patient alone.

Practitioners inevitably must strike a reasonable balance between drug benefits and tolerability, but there are instances in which concerns about safety and tolerability may override efficacy or breadth of spectrum when choosing from among pharmacotherapy options. Adverse effects such as weight gain and metabolic dysregulation, sexual dysfunction, sedation, or cognitive dulling can often contribute to treatment nonadherence or otherwise make for situations in which, to paraphrase Sir Francis Bacon, the cure becomes worse than the disease. Well-tolerated but ineffectual treatments make no sense beyond their placebo effect (which, admittedly, contributes in no small part to many treatment outcomes). However, assuming the added value of established pharmacodynamic efficacy (particularly for high-severity mood episodes), clinicians must not disregard high-potency, potentially life-saving interventions (e.g., lithium or clozapine) that may be uniquely beneficial simply in response to concerns about anticipated side effects. By the same token, extensive and “irrational” polypharmacy regimens sometimes arise (alongside the additive accrual of side effects) that may be nonpurposeful and ineffective, in which case streamlining and the pruning of a regimen may be necessary to achieve the best overall patient fit (122).

Other factors that may drive pharmacotherapy decision making may include medical comorbid conditions (e.g., lithium discontinuation in a patient who develops stage 3B chronic kidney disease), a significantly elevated QTc interval on an electrocardiogram for a patient requiring antipsychotic medication, or teratogenicity concerns in a woman taking divalproex who wants to become pregnant (123). Pharmacogenetic test results sometimes influence medication decisions regarding tolerability (e.g., poor metabolizers of cytochrome P450) or possible lack of efficacy (e.g., ultrarapid metabolizers), although the Food and Drug Administration (FDA) issued a consumer warning in November 2018 on the limitations of pharmacogenetic testing (https://www.fda.gov/MedicalDevices/Safety/AlertsandNotices/ucm624725.htm), cautioning practitioners against changing medications solely on the basis of genetic testing results.

While existing treatments for bipolar disorder can be effective across all phases of illness, many patients still have incomplete remissions or take regimens that are ill suited to the idiosyncrasies of their unique presentations. Future studies must address outcomes related not only to mood but also cognition, impulsivity, psychosis, circadian dysregulations, and anxiety and other comorbid conditions in the hope that both treatment adherence and overall illness burden might improve from targeting all relevant symptom domains. Alongside the search for new drug mechanisms and novel compounds, intervention studies (and practice guidelines) should routinely account for known moderators and mediators of treatment outcome, which, in turn, can help practitioners translate RCT findings in ways that best fit the distinct clinical profile of individual patients.

Bipolar disorder is a phenomenologically heterogeneous condition with multiple symptom targets that extend beyond mood alone. Randomized trials that establish drug efficacy for an overall diagnosis and illness state—such as bipolar mania or depression—can provide a wealth of further information to help practitioners recognize characteristics that may increase the probability of a desired outcome. Clinicians might usefully think of an individual patient’s candidacy for a pharmacotherapy, which is based not solely on the randomized trial evidence for the overall disease entity but, more specifically, on matching recognized treatment moderators and the unique clinical profile of an individual patient beyond the diagnosis alone.