"I used to ask people: how many times have you been asked to make a trial go faster? Everybody holds up their hands. How many have been asked to slow down? Nobody holds up their hands. That huge imbalance tells you something."
Dr. Steve Brannan joined Karuna Therapeutics in 2017 as employee number two. By the time BMS acquired the company and Cobenfy was approved, he had witnessed one of the rarest outcomes in CNS drug development: a clinical program that never stumbled. No blown pivotals. No unexplained failures. No sudden reversal of fortune. Just a clean path from the science to the market.
His explanation for how that happened is methodical and specific. It starts with understanding what actually kills clinical trials — not the science, not the mechanism, but the operational decisions that introduce avoidable variance. Going too fast. Using sites that aren't ready. Not watching the data in real time. Letting the pressure to accelerate enrollment override judgment about patient quality.
This conversation covers the full methodology: how to select and manage research sites, how to detect problems early through blinded data analytics, why the "soft touch" is not softness but strategy, and what digital biomarkers and ecological momentary assessment will change about psychiatric trials over the next decade. It also touches on the emerging challenge of neuroplastigens — drugs like psychedelics that work once and then create effects for months — and why they pose unique design problems that the field has not yet solved.
The Right Pace Principle: Speed Is Not a Virtue in Clinical Research
Dr. Brannan's central observation about clinical trial culture is that the institutional bias toward speed is overwhelming and essentially universal. In his experience, sponsors at every level — large pharma, mid-size biotech, senior leadership — apply near-constant pressure to go faster. He has never once been asked to slow a trial down.
That asymmetry, he argues, reflects a misunderstanding of what speed actually does to data quality in psychiatry. In trials with objective, hard endpoints — HbA1c, tumor volume on imaging — going fast is often fine. The measure is stable, hard to fake, and not particularly sensitive to how quickly you enroll. Psychiatric trials are different. The endpoints are soft — how depressed do you feel, how bad are the voices — and they fluctuate. Bringing patients in too quickly means more B and C sites. More heterogeneous patient populations. More variance in the ratings. And variance is the enemy of signal detection.
The example he returns to is a program at Takeda with Lundbeck, where an antidepressant went zero for eight in a first wave of studies and four for five in a second wave. Same compound. The difference: the second wave was designed around what had gone wrong. Less speed. Fewer sites selected more carefully. Stricter monitoring. The signal was there all along — it just needed a trial design that could find it.
The A-Site Problem: Why Using Too Many Sites Destroys Quality
One of the most underappreciated operational problems in clinical trial execution, in Dr. Brannan's view, is what happens when sponsors try to hit enrollment targets by adding sites. There is a finite pool of high-quality research sites for any given indication. Once you've engaged the A sites, adding more doesn't increase quality — it introduces B sites, then C sites, then D sites. Each level down adds more variance to the data.
His solution is the same as for pace: accept that going slower with fewer, better sites produces cleaner data than going faster with a larger, more variable site network. The instinct to add sites to compensate for lagging enrollment is understandable from a timeline perspective and almost always wrong from a scientific one.
Protocol feedback is part of the same philosophy. Before finalizing a protocol, Dr. Brannan routinely sends it to the CRO and a handful of sites and asks for input. Not because he believes any single site knows more than he does about trial design, but because no single person sees everything. Every time he's done this, multiple practical issues surface — scheduling conflicts between visits, operationally difficult combinations of assessments, logistical gaps. Sites that are consulted before a trial starts feel invested in the outcome. They know they're being listened to. That relationship pays dividends throughout execution.
Soft Touch Site Management: The Relationship as the Asset
Trust-based site management is, for Dr. Brannan, one of the least replicable competitive advantages in clinical development — and one of the most undervalued. When the BMS team came with him to the American Society of Psychiatry meeting shortly after the Karuna acquisition, and the site group literally brought out candles and sang a song for Steve, the BMS people were genuinely puzzled. "They really like you," they observed. His response: that's important. There are three other trials competing for the same time and space. Sites that like you give you a fair shot. Sites that don't give you whatever's left.
Building that trust requires a specific discipline: being reachable, being collaborative, being willing to hear things you don't want to hear, and communicating about problems in ways that don't make site staff feel attacked. His example is the site that enrolled three consecutive patients over 55 in a trial where the median age was 42-44. Rather than confronting them directly, he called the PI and asked with humor how long the geriatric convention was in town. The next three patients were all under 55. The point was made without creating an adversarial dynamic.
The same logic applies to hiring. When he evaluates people for clinical teams, he's looking for whether they're in the work for reasons beyond themselves — a genuine desire to help patients, not just to advance careers. That motivation shapes how people communicate with sites under pressure, and under pressure is when site relationships get tested.
Blinded Data Analytics: Catching Problems Before They Become Crises
What Dr. Brannan calls "looking for funny business" — real-time monitoring of blinded trial data for anomalies — is one of the most practical elements of his methodology. The concept is simple: statistical patterns that are highly unlikely to occur by chance should trigger investigation.
In one Karuna study, a site enrolled three consecutive patients over age 55. One such patient is possible. Two attracts attention. Three in a row is not random. Investigation revealed the site was steering older patients to Karuna while routing the younger ones into a competitor trial that had a 55-year age cutoff. The practice was discovered and corrected without antagonizing the site.
In a earlier program at Takeda, the first wave of studies contained 23 duplicate patients — the same person enrolled in two different studies simultaneously. The problem was discovered through data analytics and became the foundation for systematic fraud detection procedures. Dr. Brannan's analogy is apt: fraud detection is like TSA at airports. It's a burden that didn't used to exist and now does, because the alternative is worse. Having rigorous systems, however inconvenient, deters the people looking for an easier target.
Rating monitoring adds another layer. In psychiatric trials with scale-based outcomes like the PANSS, he expects to see a normal distribution of scores across sites. Sites where patients cluster suspiciously at threshold values, or where rater scores deviate consistently from central raters, are flagged for review. The process is constant and intensive — not a periodic audit but an ongoing watch.
The Case for Digital Biomarkers in Precision Psychiatry
One of the deepest limitations in psychiatric drug development, in Dr. Brannan's view, is diagnostic imprecision. Schizophrenia and psychotic bipolar disorder can be nearly indistinguishable in the first week of presentation. Depression is probably three or four distinct biological conditions with the same surface phenotype. Without biomarkers that distinguish them, trials include patients who may respond very differently to the same compound — and variance increases accordingly.
Digital biomarkers represent the most promising near-term solution. Speech analysis is the example he highlights. It has been known for over a century that speech patterns in schizophrenia are distinctive — altered tone, unusual pauses, different prosody. What is new is the ability to record, process, and analyze speech at scale with AI, extracting objective features from what was previously only perceived impressionistically. The early data shows meaningful differences between patient populations and between treated and untreated states.
His expectation: digital biomarkers become standard in psychiatric trials within 5-10 years, enriching patient selection and providing more sensitive outcome measures. The barrier is regulatory conservatism — both the FDA and large sponsors move slowly to adopt new approaches. But the need is clear enough that adoption is, in his view, inevitable. An enrichment strategy built on vocal or facial biomarkers would likely yield a labeling caveat describing the study population, but that caveat would also become a precision prescription guide in clinical practice.
Ecological Momentary Assessment: Getting Data Between Visits
Traditional psychiatric trial measurement happens at clinic visits, which may be weekly at best. Patients are asked to recall how they've been feeling over the past week. The limitations are obvious — most people can't reliably recall their breakfast three days ago, much less the severity and frequency of psychiatric symptoms.
Ecological momentary assessment (EMA) addresses this by sending brief prompts via smartphone throughout the day: where are you, who are you with, what are you doing, are the voices bothering you? The result is a dense data stream that captures symptom fluctuation in real time rather than retrospectively. Dr. Brannan's team piloted EMA in a Karuna safety study and found the data quality striking — more statistical power, better characterization of the symptom trajectory, and a more honest picture of daily life than weekly clinic visits can provide.
The challenge is regulatory adoption. FDA guidance on EMA as a trial endpoint lags behind the science. But the technology is advancing quickly, and Dr. Brannan's position is that the field should be using it now, in exploratory analyses and safety studies, to build the evidentiary base that will eventually move it into primary endpoints.
AI in Clinical Development: Opportunities and What to Watch
Dr. Brannan's view on AI in clinical development is cautiously optimistic and highly specific. The strongest near-term applications are in pattern detection — exactly what blinded data analytics already does, but faster and more sensitive. AI can identify anomalies in enrollment patterns, rater behavior, and patient-reported data that human monitors would miss or catch too late.
The area where he urges caution is in autonomous decision-making about patient selection or trial endpoints. Psychiatric populations are heterogeneous in ways that algorithmic models may not capture well, and the consequences of miscalibration — including placebo enrichment or inadvertent exclusion of treatment responders — can invalidate a trial. The right model, in his view, is AI as augmentation: surfacing patterns for experienced humans to evaluate, not replacing clinical judgment.
Neuroplastigens: A Design Problem the Field Hasn't Solved
Neuroplastigens — drugs that trigger neuroplastic changes lasting far longer than the drug's residence time in the body — pose a set of clinical development challenges that Dr. Brannan finds genuinely difficult. Psychedelics are the clearest example: a single dose may produce effects that persist for four or five months. How do you do a dose-ranging study when the dose you gave three months ago is still producing its effects? How do you design a placebo-controlled trial when the psychedelic experience is itself a blinding problem? How do you establish a dose-response relationship when the response is so delayed and so durable?
These are not hypothetical concerns. They are active design challenges facing every sponsor in the psychedelic space. The conventional clinical trial toolkit — titrate to effect, randomize, measure at a fixed endpoint — was built for drugs that work quickly and clear quickly. Neuroplastigens don't fit that model. The field is in early days of figuring out what a rigorous trial design looks like for them.
Clinical Operations: The Unsung Foundation
Dr. Brannan's closing argument is about where credit gets assigned in drug development. Scientists get the credit. CEOs get the credit. The clinical operations team — the people who set up the site networks, manage the monitoring, watch the data streams, fly to conferences to maintain relationships — rarely do. In his view, that is backwards.
At Karuna, Sharon Sach joined as a clinical operations consultant from the start, bringing years of GSK experience and a network of trusted site relationships. Her contribution to the program's clean run was as significant as any scientific decision. The sites enrolled the right patients because they trusted her and knew she would be watching. The protocols worked in practice because she had lived through enough protocol failures to know what to check.
Clinical operations excellence is not a support function. It is, for Dr. Brannan, the difference between a well-designed trial that fails operationally and a moderately designed trial that succeeds because every execution element was right. In a field where failures are common and expensive, that distinction is everything.
What You'll Learn
- Why the relentless pressure to accelerate enrollment is the leading cause of avoidable trial failure in psychiatry
- How to select the right research sites and know when adding more sites hurts more than it helps
- What "soft touch" site management looks like in practice — and why relationships with PIs are a real competitive advantage
- How real-time blinded data analytics can catch fraud, rater drift, and site behavior problems before they contaminate the trial
- Why digital biomarkers like speech analysis may finally give psychiatric trials the precision endpoints the field needs
- What ecological momentary assessment can do for symptom capture that weekly clinic visits cannot
- How AI fits into clinical development — and where caution is warranted
- Why neuroplastigens like psychedelics break the standard dose-response trial design model
- How the Karuna program achieved FDA approval of the first new-mechanism schizophrenia drug in decades without a single major stumble
Episode Highlights
- [00:00] Intro: Leading Innovation in Psychiatric Drug Development
- [04:01] The Story Behind Karuna's Breakthrough Schizophrenia Drug
- [11:23] Navigating Big Pharma vs Biotech: Lessons from Both Worlds
- [17:12] The Art of Managing Trial Variance in Psychiatry
- [26:29] Building Trust Through "Soft Touch" Site Management
- [37:56] The Promise of Digital Biomarkers in Psychiatric Trials
- [46:52] AI's Role in Clinical Development: Opportunities and Limitations
- [49:47] Neuroplasticity and the Future of Psychiatric Medicine
- [56:37] Clinical Operations: The Unsung Heroes of Trial Success
