R2H
insights

The Profile Fits. Will the Patient Respond?

By Ravza Nazli Muyesseroglu, MD, MSc Scientific Lead & Co-Founder, R2H·November 11, 2026

AI can quickly find the patients who meet a trial’s criteria. Whether the drug will actually work for them is a different question, and answering it still takes biology and clinical experience.

The Profile Fits. Will the Patient Respond?

When a clinical trial fails, the first reaction is usually to blame the molecule. Maybe the target was wrong, the dose wasn’t right, or the mechanism didn’t work as expected.

But that may not be the whole story. Over the past few months, I’ve written about how some trials may fail not because the drug doesn’t work, but because it wasn’t tested in the right patients. A drug that works for a specific group can appear ineffective when its results are averaged across a much broader population.

Patient selection is becoming a bigger part of the conversation in drug development. A recent industry article argued that the next competitive advantage may come not from designing better molecules, but from identifying the patients most likely to benefit from them. I agree, but I think there is another question worth asking: even if AI can identify patients who match a profile, how do we know that profile captures the biology that determines who responds?

The Same Diagnosis, Different Biology

A recent piece in Springer Nature’s research communities made a related point. Its argument was that trials keep failing not for lack of data, speed, or AI, but because the features used to select patients often don’t reflect what is actually driving their disease.

The example it uses is HER2-positive breast cancer, which is defined by high levels of a protein involved in cell growth. HER2 is one of the real successes of this approach: testing for it changed how the disease is treated. But even within that group, the piece points out, tumours can differ in ways that affect how they respond. The same diagnosis doesn’t always mean the same underlying biology. If patients can differ even within a well-defined group like this, the problem may be greater in trials that rely on broader disease labels.

Finding a Match Isn’t Predicting a Response

None of this means AI has no role. Finding patients means working through vast amounts of information, from medical records and genetic data to scans and lab results. AI can do this at a scale and speed that human teams would struggle to match, picking out the patients who fit a given profile.

But matching a patient to a profile is not the same as knowing they will respond. That depends on whether the profile reflects the biology driving the disease and the way the drug works. Answering it requires understanding the drug’s mechanism, the evidence behind the biomarker, and clinical experience of how the disease behaves in patients, which a dataset doesn’t fully capture.

The difficulty is clearest when a biomarker-guided trial fails, because the reason is rarely obvious. The drug may not have reached its target as expected. It may have reached it, and the mechanism mattered less than hoped. Or the biomarker may have failed to pick out the patients most likely to respond. These can all produce the same disappointing result, but they point to very different decisions about what to do next. Working out which one happened is careful work for clinicians, biologists and data scientists together, and no matching algorithm does it for them.

Patient selection can also be oversold. The PROFILE trial in Crohn’s disease is a useful reminder. A blood-based biomarker that had looked promising in earlier research was tested to see whether it could guide treatment decisions. It did not help predict which patients would benefit from which strategy, even though early intensive treatment proved far more effective than the conventional step-up approach. The lesson isn’t that biomarkers are useless. It’s that a promising signal still has to prove its value in a trial.

AI has made the early stages of drug discovery faster, and it is making it easier to find patients who fit a profile. This is also how we approach the work at R2H: AI helps bring together and evaluate the scientific evidence, while clinical and biological expertise helps us understand what that evidence means and where its limits are.

Finding patients who match a profile is useful. Knowing whether that profile can actually predict a response is what matters.

Accelerate Your Translational Research

Experience how Clarisyn transforms complex scientific problems into structured, evidence-based reports.

Get Started with Clarisyn