A target trial emulation just published in Lancet Psychiatry reported something that should reopen a clinical argument many of us thought was settled: in patients after a first episode of psychotic mania, mood stabilizer-based strategies, whether as monotherapy or in combination, carried a meaningfully lower relapse risk than second-generation antipsychotic monotherapy during continuation and maintenance treatment. The BD-CAUSAL Collaboration pulled this signal from real-world data across North America, Chile, and Spain. That’s a geographically wide net, and the finding is consistent enough across sites to deserve serious attention.

My reaction as a clinician-PI is not surprise. My reaction is: why did it take observational data at this scale to surface what many of us see every week in practice?

The Antipsychotic-First Assumption Deserves Scrutiny

The clinical logic that drove antipsychotic-first prescribing in early psychotic mania was never unreasonable. Acute psychosis demands rapid symptom control, SGAs deliver it reliably, and continuation of the acute-phase agent feels like the path of least resistance. The APA’s practice guideline for bipolar disorder acknowledges that manic or mixed episodes with psychotic features warrant antipsychotic coverage, which in practice often translates into sustained SGA use well into the maintenance phase. The problem is that “what works in the acute phase” and “what protects against relapse over months and years” are different clinical questions, and we have been conflating them.

The neuropharmacological basis for mood stabilizer superiority in maintenance is not obscure. Lithium’s action on GSK-3 beta signaling and its downstream neuroprotective effects on hippocampal volume are distinct mechanisms from dopamine D2 blockade. Divalproex, FDA-approved for acute mania since 1995, modulates GABA and histone deacetylase activity in ways that speak more directly to the cycling biology of bipolar disorder. An SGA holding dopamine tone steady does not do the same thing. The BD-CAUSAL signal is plausible on mechanistic grounds, which is partly why a meta-analysis of maintenance RCTs in bipolar disorder already showed that no single antipsychotic or mood stabilizer monotherapy demonstrated a significantly reduced risk for both manic and depressive relapse simultaneously. The current trial emulation is consistent with that prior evidence, it just makes the pattern harder to dismiss.

Target trial emulation is an increasingly rigorous tool for exactly this kind of question. A 2024 call-to-arms in Lancet Psychiatry from CAUSALab at Harvard made the methodological case for this approach in psychiatric research, arguing that properly specified emulations can reduce confounding in ways that observational psychiatry studies rarely achieve. The BD-CAUSAL collaboration is doing what that framework recommends. But emulation is not the same as a randomized trial, and the authors are honest about that, their own conclusion calls for randomized trials to confirm these findings.

What First-Episode Trials Are Still Getting Wrong

Here is the design problem the BD-CAUSAL data exposes: virtually every continuation and maintenance trial in early psychotic bipolar disorder has been built around the drug in front of it, not around the sequencing question. Trials test whether drug X prevents relapse better than placebo. Almost none are powered to answer whether initiating with a mood stabilizer backbone in the first episode changes the trajectory of illness over two or three years compared to initiating with an SGA and adding a stabilizer only after breakthrough.

Having run SUD and mood disorder trials and treated over 30,000 patients through guided ketamine medicine sessions at Mindbloom, I can tell you that the heterogeneity within “first-episode psychotic mania” is enormous. A 22-year-old with a clear precipitant, family history of lithium response, and no comorbid substance use looks nothing, biologically or clinically, like a 35-year-old with a first recognized episode that is probably not actually their first. The R-LiNK consortium’s work on predictive biomarkers of lithium response, following 169 patients over 24 months, points toward exactly this phenotype-stratification problem: response to mood stabilizers is not uniformly distributed, and we are enrolling as if it were.

Future continuation and maintenance trials need to embed phenotype stratification from enrollment, not as a post-hoc subgroup. That means baseline variables for family history of mood stabilizer response, comorbid substance use disorder status (which affects both medication adherence and relapse risk independently), and a genuine first-episode confirmation window that rules out prior subclinical episodes. Spanish-speaking and Latin American patients were included in the BD-CAUSAL dataset, which is clinically meaningful, but bilingual consent and culturally adapted adherence monitoring are not standard trial infrastructure, and without them, the populations that appear in observational data disappear in RCTs.

I am watching for whether any Phase 3 sponsor moves toward a head-to-head sequencing design in early psychotic bipolar disorder in the next 12 to 18 months, and whether any protocol takes the BD-CAUSAL phenotype heterogeneity seriously enough to pre-specify subgroup analyses around mood stabilizer response predictors at baseline. If they do not, we will have another generation of trials that answer the wrong question cleanly.

References

  1. Lancet Psychiatry, “Comparative effectiveness of mood stabiliser monotherapy versus second-generation antipsychotic monotherapy versus combination therapy for continuation and maintenance treatment after a first episode of psychotic mania: BD-CAUSAL Collaboration”
  2. American Psychiatric Association, “Practice Guideline for the Treatment of Patients With Bipolar Disorder, Second Edition”
  3. Expert Review of Neurotherapeutics, “Divalproex sodium in the treatment of adults with bipolar disorder” (FDA approval 1995 reference)
  4. International Journal of Neuropsychopharmacology, Meta-analysis of maintenance RCTs in bipolar disorder: relapse risk by treatment strategy
  5. Lancet Psychiatry, July 2024, Szmulewicz AG (CAUSALab, Harvard T.H. Chan School of Public Health): “Target trial emulation in psychiatric research”
  6. R-LiNK Consortium, “Predictive biomarkers of lithium response: preliminary results from the European R-LiNK initiative” (169 patients, 24-month follow-up)
+ posts

Dr. Leonardo Vando, MD is Principal Investigator of Psychiatry at Vando Medical Services.