03. Can We Predict Who Responds to Mood Stabilizers?

Text version

Predicting Treatment Outcomes in Bipolar Disorder

We all know that for our bipolar disorder patients, finding the right medication can feel like navigating a maze. It often involves months or even years of trial and error with treatments that simply don’t work for them. But what if we could bring more precision to that process, helping our patients find stability sooner?

That’s exactly what a recent review in Biological Psychiatry by Scott and colleagues has explored. It offers some truly insightful directions for our clinical practice. So today, I want to discuss this paper and explore the challenge that weighs heavily on all of us who treat patients with mental health conditions—predicting what treatments will lead to success in bipolar disorder.

Limited Efficacy: Current Treatments Help Only One-Third

We know that effective long-term treatment can stabilize bipolar symptoms and help patients regain their lives. But here’s the rub I want to highlight. We have nine first-line and seven second-line treatments according to the Canadian CANMAT Bipolar Guidelines, yet none are useful in more than a third of all cases.

This means we’re left to a process of trial and error. Imagine this: a patient tries a medication and it could take months to evaluate its benefit because of the illness’ variable course, even with spontaneous remissions. More than two-thirds of our patients might be on ineffective treatments for prolonged periods, leading to extended periods of poorly stabilized illness and unnecessary exposure to side effects.

This review highlights the urgent need for a reliable method to identify potential non-responders. The paper focuses mainly on lithium, which has been studied most extensively, and to a lesser extent on anticonvulsants like valproate and lamotrigine and the antipsychotics. When we talk about response in long-term treatment, it’s not just about acute symptom changes—it’s often about the overall reduction in illness morbidity over time, measured by things like recurrence frequency, hospitalizations, or global functioning scales.

Clinical Predictors Of Lithium Response

Let’s dive into some clinical pearls from this review. First off, who is most likely to respond well to lithium? The most promising results show that certain clinical features are strongly associated with a good lithium response:

• Episodic course of illness
• Absence of rapid cycling (≥4 episodes/year)
• Later age of onset
• Absence of psychotic symptoms
• Hypomanic or manic onset episode
• Low rates of comorbidities (e.g., substance use, OCD, ADHD)
• Family history of bipolar disorder, and notably a family history of good lithium response, suggest a familial, potentially genetic component to lithium responsiveness.

Lamotrigine and Antipsychotics Have Different Profiles

For lamotrigine, response tends to be better in patients with:

• A predominantly depressive polarity (2/3 or more of their episode time spent depressive)
• More lifetime episodes

For atypical antipsychotics, there’s some evidence for better response in patients:

• Without a history of rapid cycling
• With more severe manic symptoms in the short term

Genetic And Biological Markers Show Promise

Now, beyond clinical observations, this paper delves into biological markers. Imagine if we can measure something in a lab and know before treatment whether a patient will respond. Genetics is a rapidly evolving area.

While early genetic studies had mixed results, genome-wide association studies (GWAS) have found significant associations. A crucial finding is the association of lithium non-response with higher polygenic risk scores for schizophrenia and major depression. While these genetic findings alone explain only a small fraction of treatment outcome variability, their predictive accuracy improves substantially when combined with clinical variables. This points to the power of integrating different types of data.

Cellular Level Insights Reveal Medication-Specific Responses

At the cellular level, there’s some truly exciting evolving literature. Studies using induced pluripotent stem cell (iPSC)-derived neurons from patients with bipolar disorder have identified hyperexcitability. What’s remarkable is that:

• This hyperexcitability was selectively reversed by lithium only in cells from patients who responded well to the medication in real life
• In cells from lithium non-responders, their hyperexcitability was rescued by anticonvulsants like lamotrigine and valproate, but not lithium

This suggests a biological distinction between responders and non-responders, offering a potential future pathway for precision medicine—matching the patient’s biology to the right medication.

Limitations and Challenges

Heterogeneity

The authors are very transparent about the current limitations and the challenges ahead. One of the biggest hurdles is the significant variability within bipolar disorder itself and across different patient populations.

Machine learning models, despite their promise, often show poor predictive performance when applied to data from different clinical sites than where they were trained. This suggests that clinical predictors might vary based on geographic, cultural, or environmental factors, making it tough to create universally generalizable models. The cause of the site’s specific variability is largely unknown.

Correlates vs predictors

Much of the existing data come from uncontrolled observations, not randomized controlled trials. This means that the features associated with treatment response should often be viewed as correlates rather than definitive predictors.

Are we truly predicting a specific drug’s effect or just identifying patients with a generally more benign illness course that might respond well to any mood stabilizer? While the iPSC studies lean towards specificity for lithium, this question remains vital for our clinical practice.

Need Models for Individual Medications

Next is medication-specific models. We desperately need models that can distinguish between predictors for a specific agent like lithium and predictors for an overall good prognosis. Without this, we risk overprescribing lithium to patients who might do equally well on other treatments with maybe fewer side effects.

We also risk underusing lithium in those who would be excellent responders.

Predicting Side Effects Remains Crucial

A final issue is predicting harms. The paper rightly points out that efficacy isn’t the only determinant for precision decision making. We also need to predict the risk of various treatment options.

For example, while lithium is highly effective, concerns about kidney impairment—which can affect a third of patients to some degree treated for over 10 years with lithium, though severe renal impairment is similar to when they’re treated with other options—or concerns about tremors might lead a patient to prefer an anticonvulsant or atypical antipsychotic. Conversely, a patient might choose lithium to avoid some of the weight gain associated with valproate or atypical antipsychotics.

We lack comprehensive real-world data on side effects to develop robust models for predicting individual patient risks.

The Path Forward: Precision Medicine in Bipolar Disorder

To sum up, the journey toward precision medicine in bipolar disorder is promising. We are moving beyond trial and error with insights from clinical features, genetics, and even cellular biology. Significant challenges remain, particularly in addressing illness heterogeneity and building generalizable, medication-specific models that also incorporate risks.

This review underscores that we are on the path to making more informed, tailored treatment decisions for our patients. It highlights some features around what a patient’s illness looks like and their family history that we can use in clinical practice today to help guide our treatment choices and potentially improve their likelihood of good long-term response.

Abstract

Prediction of Treatment Outcome in Bipolar Disorder: When Can We Expect Clinical Relevance?

Long-term pharmacological treatment is the cornerstone of the management of bipolar disorder (BD). Clinicians typically select mood-stabilizing medications from among several options through trial and error. This process could be optimized by using robust predictors of treatment response. We review clinical features and biological markers studied in relation to outcome of long-term treatment of BD. To date, the literature focuses mostly on lithium and to a lesser extent on the anticonvulsants valproate and lamotrigine. The most promising results show association of lithium response with certain clinical features (episodic clinical course and absence of rapid cycling, low rates of comorbid conditions, family history of BD and lithium response) as well as low polygenic risk for schizophrenia and major depression. The clinical application of these findings remains limited, however, due to heterogeneity of the illness as well as unanswered questions about specificity of the effects of different medications.

Reference

Scott, Katie et al. (2025). Prediction of Treatment Outcome in Bipolar Disorder: When Can We Expect Clinical Relevance?. Biological Psychiatry, Volume 98, Issue 4, 285 – 292.