"If all 15 patients are already on a trial, where does your new drug go?" Lessons from the frontline of Phase I/II design

The fireside chat "Data that holds up under pressure: how to generate Phase I to II efficacy data that survives scale-up, scrutiny and value inflection" brought together four people who've each watched a trial succeed, fail, or both, sometimes on the same programme. Moderated by Judith Koliwer (Senior Industry Advisor, Körber Pharma), the panel featured Nina Kotsopoulou (CTO, AAVantgarde), Azadeh Kia (VP Gene Therapy Research, Beacon Therapeutics), and Kieran Breen (research advisor, Parkinson's Europe, and vice chair of the EMA's Committee for Advanced Therapies).

Koliwer opened the session not with a question about data, but with a personal one: what actually keeps each of them doing this work. Her own answer set the emotional register for the whole conversation. A clinical trial she ran for an autism spectrum condition failed after Phase I, not because the data was bad, but because it lacked the stability to attract Phase II investment. What keeps her going, she said, is that a US company later picked up the trial and built on her data. The programme she couldn't finish is still moving science forward under someone else's name. That's not a small thing to admit on stage, and it framed everything that came after: this panel wasn't interested in a tidy checklist. It wanted to talk about where things actually go wrong, and why.

Picking the wrong patients is still the single biggest failure mode

Breen went first on trial design, and his answer was unambiguous: the most common mistake he sees is getting the patient group wrong, and it usually comes from wanting to hedge rather than commit.

Using Parkinson's as his example, he described the challenge of heterogeneity: early, mid and late-stage patients respond completely differently, and a drug addressing symptoms needs a different cohort to one addressing disease progression. His most striking illustration was a real Parkinson's ATMP trial that used a mid-stage patient group rather than early-stage, hoping a broad net would catch some responders. It failed. His diagnosis was blunt: teams choose a wide group in the hope that some patients will respond, rather than a narrow group of the patients who are genuinely most likely to.

His mental model for getting this right is a cone. Start narrow, with the patients most likely to show a response, to prove the therapy genuinely works. Only once that's established do you broaden the population in Phase III. He drew a direct parallel to preclinical work: you don't validate a therapy in the wrong mouse model and hope for the best, and the same discipline applies to your first human cohort.

Alongside patient selection, he stressed getting outcome measures right from day one, both clinical and patient-reported, and taking that trial design to regulators for scientific advice before committing to it. Getting a national regulator or the EMA to sign off on your primary endpoint before you start is, in his words, simply better than finding out afterwards that you chose the wrong one.

Cell and gene therapy doesn't get a second attempt

Kia was direct about what makes advanced therapies different from conventional drug development: cost. Where traditional medicines can iterate through multiple trial cycles, cell and gene therapy programmes often can't afford to go back and get it right a second time. Get the patient group wrong here, and you haven't just lost time, you may have lost the only shot that population was going to get.

Kotsopoulou added the structural reason this matters so much: in rare disease, a short Phase I/II is frequently followed directly by a pivotal trial, with no traditional Phase III in between. That means the early trial has to generate genuine efficacy signal, not just safety data, because the investment case for moving to pivotal depends on it. Her framing was pointed: think about the end from the very beginning, and that applies to product design, not only trial design.

The decisions that get locked in before you ever dose a patient

Kia's account of what actually determines Phase I/II success was the most technically detailed part of the session, and her core point was that the outcome is largely decided long before the first patient is dosed. Transgene cassette design determines potency and expression, which in turn determines dose, which in turn affects the safety profile you'll see in patients. Capsid selection gets locked in once you commit to a GMP batch, and reversing that decision later is genuinely difficult.

Her advice on endpoints was equally direct: choose what's meaningful for understanding mechanism of action and efficacy, not what's simply easiest to measure. And on potency assays specifically, while regulators don't require one analytically at Phase I, the groundwork needs to start early regardless, particularly for anything targeting an intracellular protein, where establishing a reliable assay is genuinely hard.

Her closing point on this topic doubled as a challenge to the room: researchers need to shed an academic mindset that treats data as an end in itself. The goal isn't a publication, it's a product that reaches a patient, and that requires research to be in close, continuous dialogue with CMC, regulatory and commercial from the outset, not handed a data package once the science is "done."

Kotsopoulou backed this with a cautionary example from beta thalassaemia programmes that had to codon-optimise mid-trial because expression wasn't sufficient at the outset, an expensive fix that can derail a programme's timeline and, in some cases, its viability. Her practical recommendation: even though a potency assay isn't formally required until pivotal trials, having one in place early pays off later, both for managing comparability and for maintaining continuity in your data as the product evolves.

When there's no control arm to compare against

Breen raised a challenge specific to genuinely rare disease: patient numbers are sometimes so small that a placebo-controlled trial isn't feasible at all. Increasingly, the answer is retrospective patient data, real-world evidence and patient registries standing in as the control arm. The catch, he stressed, is that this only works if the prospective trial is designed from the outset to collect data in a form that genuinely matches what was collected retrospectively. Get that mismatch wrong, and the comparison falls apart later, when it's too late to fix.

He extended this into a broader point about sequencing: manufacturing, preclinical work and clinical trial design need to be planned as one continuous pipeline from day one, not treated as sequential handoffs. His most sobering illustration was a rare disease with a total patient population of around 15, all of whom were already enrolled in an existing trial. A new drug developer arriving after that point has, in practical terms, nowhere left to go, regardless of how strong the science is. Kia's response captured the commercial stakes precisely: that's a commercialisation failure, not a scientific one, and Luxturna's own patient-population miscalculation stands as a real precedent for how a clinically successful therapy can still fall short commercially.

What the panel wants to see change

Asked what needs to improve, the answers converged fast on two words: collaboration and communication. Kotsopoulou wants to see research, clinical, manufacturing and commercial functions genuinely working as one machine from the earliest stage, alongside early regulatory engagement to confirm the development path is actually viable before too much is committed. Kia agreed, noting that most of the failures she's seen trace back to poor communication between functions rather than poor science, and called specifically for more work on managing immune response and cytokine storm risk in gene therapy, an issue she linked directly to the very high vector doses (in the order of 10^14 vg/kg) now being used in some programmes, and the deaths that have resulted.

Breen offered a genuine example of collaboration done well: an ongoing joint Swedish-UK cell therapy trial for early-stage Parkinson's, rebuilt from the ground up after a similar trial failed roughly a decade and a half earlier. This time, the team properly characterised their cells, selected the right patient group, took advice from both the MHRA and EMA, and worked directly with patient groups in both countries to define outcome measures that actually mattered to patients. He called it as close to a gold standard as the field currently has, precisely because it was built on the lessons of a documented failure rather than a fresh guess.

Koliwer's own closing point was about access rather than design: making it genuinely easier for patients and their families to find relevant trials in the first place, rather than relying on patient organisations and informal networks to do that work by default.

The one lesson each panellist wanted the room to leave with

Kotsopoulou: characterise and understand your product as early and as thoroughly as possible, because the more you know, the smoother your comparability and regulatory path will be. Kia: excellent science is necessary but not sufficient, think about payers and reimbursement from the start, not after approval. Breen: the quality of your data is what everything else depends on. And Koliwer's final addition tied the whole session together: structure and collect your data so the next person, whoever inherits your programme, can actually build on it.

Why this session matters beyond the room

What made this fireside chat land wasn't a framework, it was four people willing to describe exactly where their own trials, or trials they'd advised on, had gone wrong, and precisely what they'd do differently. That kind of candour is rare, and it's exactly the sort of session that's worth being in the room for.

That same depth and honesty is what Advanced Therapies Europe brings back to the stage every year.

Want the rest of the story? Register your interest to attend Advanced Therapies Europe 2027 and be first to hear when sessions and speakers go live.