The FDA Measures All Schizophrenia Symptoms at the Same Clock. A Six-Trial IPD Analysis Just Proved That’s Wrong.

Picture the scene: a principal investigator at a Phase 3 schizophrenia site reviews the Week 6 PANSS assessment for a patient who started a novel antipsychotic four weeks ago. Hostility scores have dropped substantially. Excitement is controlled. The patient’s family reports the household feels safer. But the negative symptom subscale is barely moving, and the disorganization cluster looks no better than baseline. Under the protocol’s responder criteria, the patient is a non-responder. The site considers whether to escalate the dose or flag the patient for early discontinuation. On paper, the trial’s stopping rules demand a decision. In the room, the clinician’s instinct is to wait.

That instinct now has 4,661 individual participant data points behind it.

A Lancet Psychiatry individual participant data (IPD) meta-analysis drawing on six antipsychotic drug trials has mapped the temporal course of symptom domain improvement in schizophrenia with a precision that aggregate-level analyses simply cannot achieve. The finding is both clinically obvious to experienced practitioners and operationally explosive for trial design: hostility and excitement respond first, positive symptoms follow, and negative and cognitive and disorganization symptoms lag significantly behind. These are not parallel tracks reaching the same destination at the same time. They are different trains on different schedules, and the clinical trial infrastructure measuring them has been treating them as one.

A Framework Built on the Wrong Clock

The regulatory standard for schizophrenia endpoint selection has long privileged the total PANSS score or BPRS total score as the primary endpoint, with domain scores playing a supporting role at best. The EMA’s guideline on schizophrenia trial standards, tracked by RAPS, confirms that short-term efficacy assessments should center total scale instruments like PANSS or BPRS. The FDA’s benefit-risk framework, finalized in 2023, similarly evaluates a drug’s overall clinical profile without mandating domain-level temporal disaggregation as a protocol design requirement. Both agencies are measuring the right thing. They are measuring it at the wrong resolution.

The deeper problem is that treating total PANSS change as a uniform outcome across a trial’s assessment windows conflates mechanistically distinct processes into a single number. When a sponsor reports “statistically significant improvement in PANSS total score at Week 8,” that number may be driven almost entirely by hostility and excitement subscale reductions in the first two weeks, with negative and cognitive symptoms contributing little or nothing. The drug looks like it works across the board. The label ends up claiming broad efficacy. And then clinicians spend years puzzled by patients whose aggression resolved but whose social withdrawal and cognitive fog persisted.

This is not a new suspicion. A 2006 paper in the Journal of Psychiatry and Neuroscience by Agid, Seeman, and Kapur challenged the idea of “delayed onset” of antipsychotic action, demonstrating that antipsychotic effects begin within the first day of treatment, with early change predictive of later outcomes. What the field lacked was the granular, domain-specific temporal map that IPD methodology can produce. Pooling six trials at the individual participant level provides that map for the first time at this scale.

A 2006 randomized trial comparing risperidone and haloperidol in 522 first-episode schizophrenia patients found that 77 percent of the 400 subjects who responded showed that response within the first four weeks. But “response” in that study used a 20 percent improvement in total PANSS, not domain-level thresholds. That’s the measurement gap the Lancet Psychiatry IPD analysis now closes: by disaggregating response by domain, researchers can see not just whether patients respond, but which part of their illness responds first, and how far behind the rest follows.

What Six Trials Reveal That One Cannot

IPD meta-analysis methodology earns its authority through what it preserves that aggregate data destroys: individual-level timing, within-patient variance, and the ability to model trajectories rather than point-in-time comparisons. A February 2024 Lancet Psychiatry network meta-analysis combining RCT and real-world data demonstrated the methodological power of combining evidence streams across antipsychotic trials, but even that study operated at the trial-level for most temporal comparisons. The Lancet Psychiatry IPD analysis goes further by working directly with participant-level data from six trials, enabling the kind of fine-grained trajectory modeling that reveals the gap between domain response curves.

The clinical implications arrive in a specific sequence. Hostility and excitement reach substantial response earliest. Positive symptoms follow. Negative symptoms, cognitive symptoms, and disorganization symptoms trail both. The authors are explicit that treatment changes should not be made prematurely when negative and cognitive and disorganization symptoms persist, because improvement in these domains lags behind by design, not by failure.

That distinction carries enormous operational weight for anyone running a schizophrenia trial right now.

Consider what standard trial protocols do at Weeks 4 through 6. Under INTEGRATE international guidelines, clinicians are advised to give the current antipsychotic at least four weeks at a therapeutic dose before declaring inadequate response. Dose increases for atypical antipsychotics are suggested only after 14 to 21 days of inadequate response. Those are dose-management guidelines for clinical practice, not trial protocol rules, but they bleed into sponsor thinking about assessment windows and futility stopping rules. If a trial’s futility analysis occurs at Week 6 and uses a composite PANSS responder threshold that weights all domains equally, and if negative and cognitive symptoms are expected to lag by the very mechanism the Lancet IPD analysis documents, then the futility analysis is structurally biased against drugs that work on the slower-responding domains. Drugs targeting negative symptoms are being evaluated at a clock they can’t win.

The Protocol Rewrite Nobody Has Scheduled Yet

Sponsors designing CNS Phase 2 and Phase 3 protocols in 2026 face a practical choice. They can continue treating the total PANSS assessment window as a single temporal frame, accepting that any drug with a selective effect on negative or cognitive domains will underperform at interim analyses relative to drugs that hit hostility and excitement first. Or they can design domain-stratified assessment schedules that reflect the temporal architecture the IPD data now describes.

The operational translation is specific enough to put in a protocol amendment. The primary efficacy timepoint for hostility and excitement domains can be moved earlier, to Weeks 2 through 4, where the signal is strongest. Negative symptom primary endpoints should be assessed at a later timepoint, Week 12 or beyond, where domain-specific improvement curves have had time to diverge. Interim futility analyses should apply domain-appropriate responder thresholds, not uniform PANSS total score cutoffs, to avoid stopping trials that are succeeding slowly in the domains that matter most to long-term functional outcomes.

Pre-submission meetings with FDA’s Division of Psychiatry become essential here. Any sponsor planning to structure primary and key secondary endpoints around domain-disaggregated timepoints needs to surface that design rationale before the IND goes active, not in a Type B meeting after enrollment has started. The FDA’s benefit-risk framework provides room for domain-specific evidence to inform labeling, but the agency will need to see prospective justification for non-standard assessment windows in the protocol, not a post-hoc explanation in the clinical study report.

There is a harder question underneath the operational one. The EMA’s requirement for total scale primacy and the FDA’s composite benefit-risk framing both emerged in an era when the temporal disaggregation the Lancet IPD analysis demonstrates was not yet available at population scale. Six trials pooled at the participant level is not a small signal. It is a dataset that regulators on both sides of the Atlantic will need to engage with formally, either by updating guidance on assessment windows for schizophrenia trials or by issuing scientific advice that explicitly addresses how domain-lagged endpoints should be handled in trial design review.

Back in that assessment room, the PI who waited instead of escalating was, the data now suggests, making the right call. Her patient’s negative symptoms had not failed to respond. They had not yet had time to respond. The trial’s protocol didn’t know the difference, but after this IPD analysis, future protocols have no excuse not to.

References

1. Lancet Psychiatry — “Temporal course of symptom domain improvement in schizophrenia: individual participant data analysis of six antipsychotic drug trials”
2. Lancet Psychiatry — “Efficacy and effectiveness of antipsychotics in schizophrenia: network meta-analyses combining evidence from randomised controlled trials and real-world data” (January 2024)
3. RAPS — “FDA Finalizes Guidance on Benefit-Risk Assessment for New Drug and Biological Products” (2023)
4. American Journal of Psychiatry — “Time Course for Antipsychotic Treatment Response in First-Episode Schizophrenia” (April 2006)
5. DrOracle — INTEGRATE guidelines on minimum antipsychotic treatment duration before dose adjustment
6. Journal of Psychiatry and Neuroscience — Agid, Seeman, Kapur: “The ‘delayed onset’ of antipsychotic action — An idea whose time has come and gone” (March 2006)
7. RAPS — EMA guideline establishing PANSS/BPRS total score standards for schizophrenia trial development