The Gap Between the Podium and the Patient

June 23, 2026

Commercial ExcellenceDiseases Finding eligible patients Finding patients earlier inAdvance inCare inFlow Insights inTrigue Medical AffairsOncology Patient Journey Services Value & Access HEOR

By Léon Van Wouwe, Clinical Innovation Director, Volv Global

ASCO 2026 produced landmark oncology data. The question no one asked from the podium: can clinical infrastructure get patients across the gap to these drugs?

Every June, the oncology world convenes around the data. This year was exceptional.

At ASCO 2026, evidence was presented that sacituzumab tirumotecan delivered a 65% reduction in the risk of progression or death in first-line PD-L1-positive non-small cell lung cancer.^[1] Izalontamab brengitecan was shown to demonstrate a 71% reduction in PFS risk in later-line triple-negative breast cancer, with an overall survival benefit that translated to 3.4 additional months of life compared with physician’s-choice chemotherapy.^[2] Daraxonrasib was shown to cut the risk of death by 60% in second-line metastatic pancreatic cancer, a disease where meaningful survival improvement has eluded the field for decades.^[3]

These are all wonderful successes, moments that change what medicine is capable of.

And yet. After the applause fades and the abstracts are filed, there is a question I have never once heard asked from a plenary podium, and it’s one that will determine how many patients actually benefit from these achievements. The question is: does the clinical infrastructure exist to get the right patients to these drugs, at the right time, before the treatment window closes?

The honest answer across most of the patient populations represented by these data … is no.

A three-layered problem

The problem has three layers, and we have been focussed on solving only the first.

The first layer is diagnostic interpretation, where AI is applied to radiology scans, pathology slides, and biomarker readouts to improve the accuracy of interpretation of an ordered test. This is where the pharmaceutical and medical-device industry has concentrated the majority of its AI investment. The AI-in-medical-imaging market alone was approximately $1.3 billion in 2024 and is growing at 27% annually.^[4] Through mid-2024, 76% of all FDA-authorised AI medical devices were in radiology.^[5] Roche acquired PathAI for approximately $1 billion.^[6] GE HealthCare acquired Icometrix.^[7] Sanofi committed $180 million to AI-enhanced cancer biomarker identification.^[8] This work is valuable, and in the era of precision oncology it is often structurally necessary. Companion diagnostics are not optional.

But in this layer, the patient is already on their diagnostic journey. The algorithm is being applied to a test that has already been ordered, for a patient who has already been referred. The more difficult problem is upstream: getting the right patients to that test, and from that test to the right therapy, at the right moment in their disease trajectory. This problem is receiving a fraction of the attention and almost none of the capital.

The second layer is clinical trial recruitment, in particular, AI applied to electronic health records and claims data to find patients matching eligibility criteria for a specific study. This is a commercially mature area, already served well by companies such as Tempus, Flatiron, Deep 6 AI, and a dozen others. But it is episodic by definition. I.e., the algorithm finds candidates for a time-bounded enrolment activity, but does not follow the patient longitudinally after the trial closes.

The third layer, the one that connects the ASCO podium to the bedside of the patient and their community, is what I would call systemic care intelligence: the continuous, real-world application of data science to the question of which patients, right now, are not in the right place in their treatment pathway, and what it would take to change that.

This is the layer that barely exists at scale. And it is the layer that will determine whether the ASCO 2026 announcements save the lives they are capable of saving.

Three faces of the same problem

The mismatch shows up differently in different disease settings, but the underlying logic is identical.

As an example, in later-line triple-negative breast cancer (mTNBC), iza-bren’s Phase 3 control arm tells the story more clearly than any commentary could: median progression-free survival on physician’s-choice chemotherapy was 3.1 months.^[2] mTNBC patients progress rapidly, and community oncologists make the next treatment decision at the bedside, at the moment of deterioration, without advance planning. By the time a claims signal of taxane failure reaches anyone with the capacity to act on it, the treatment choice has often already been made, or defaulted. The window is 90 days, often less, which means that patient identification infrastructure that operates retrospectively, or that is activated only at approval, arrives too late.

Meanwhile, in first-line PD-L1-positive non-small cell lung cancer (NSCLC), the problem presents itself differently, albeit equivalent in consequence. Pembrolizumab monotherapy for TPS ≥50% was established as standard of care by KEYNOTE-024 in October 2016^[9], nearly a decade ago. That is a decade of established prescribing behaviour and clinical habit that community oncologists have no structural reason to revisit. The oncologist who should be initiating sacituzumab tirumotecan plus pembrolizumab when the combination label lands will, in many practices, continue initiating pembrolizumab alone because the treatment pathway at their institution has not been recalibrated. This makes the care gap behavioural, not informational.

Now, if we extend the logic further upstream, to the patients who have not yet been diagnosed at all … In alpha-1 antitrypsin deficiency (AATD), an estimated 100,000 Americans have the disease. Fewer than 10,000 have been diagnosed.^[10] More than fifteen international clinical guidelines have recommended testing for AATD in COPD and unexplained liver disease for over two decades. But compliance has remained around 15%, even in academic centres with EMR-embedded reminders.^[11,12] The reason is a behavioural phenomenon that more guidelines cannot fix, known as screening fatigue. Clinicians who have tested dozens of COPD patients without ever finding an AATD case will have, rationally, stopped testing. No algorithm improvement to a radiology scan changes this, nor does any trial recruitment tool address it. It is a design problem of the health system, which means it requires a health system design solution.

The same pattern appears across MASH, IBD, rare neurological disease, Pompe disease, ATTR cardiomyopathy. We have large populations of patients who are in the system, generating data, interacting with clinicians, but nonetheless remaining invisible to the diagnostic and therapeutic pathways that could help them.

What the solution actually requires

Machine learning applied to electronic health records and claims data can identify these patients, ahead of the clinical signal, at population scale, with meaningful accuracy. The academic evidence there is no longer in question: multiple peer-reviewed studies across diseases such as common variable immunodeficiency, acute hepatic porphyria, aromatic L-amino acid decarboxylase deficiency, psoriatic arthritis, and genetic aortopathy demonstrate feasibility.^[13–17] The science works; what we have to work on is to make it work in routine care.

This is a design problem of health systems, and it requires answers to questions the model alone cannot provide, i.e., who receives the signal, who owns the next clinical action, how are false positives managed, how is the data governed, and – critically – how the infrastructure earns and maintains the trust of the clinician whose behaviour it is trying to change.

The multi-year collaboration between Volv Global, Takeda, the Cleveland Clinic, and Komodo Health on AATD – operational since 2021 – has worked through each of these questions in practice. The model, trained on more than 30,000 confirmed AATD cases and six million negative controls in Komodo Health’s longitudinal claims data, achieves an ROC-AUC of 0.88 against both the general population and COPD mimics. It identifies predicted-undiagnosed patients with a median age 13 years younger than the currently diagnosed population, confirming that the model is finding patients earlier, not just finding easier cases.

In Phase 2, live in Epic at the Cleveland Clinic under IRB oversight, the model routes a prompt to the treating physician to consider ordering an AATD test. In this programme, the primary endpoint is test positivity rate. While guideline-only testing achieves approximately 1%, the study is powered to detect an improvement to at least 6%. Meanwhile, the number needed to test should drop from 100 to 17.

To make the distinction clear, these are clinical infrastructure metrics, not AI research outputs.

The conversation the field needs to have

The ASCO results announced in June 2026 represent genuine scientific achievement. The patients who participated in those trials, and the researchers who designed and ran them, have produced data that changes what is possible for patient care in oncology. That achievement deserves to be translated, fully and efficiently, into actual benefit for patients.

That translation is not automatic. It requires the same rigour, the same commitment to evidence, and the same willingness to ask difficult design questions that produced the trial data in the first place. It requires a move beyond improving the interpretation of tests that have already been ordered, and to serious investment in the infrastructure that determines which patients are tested, and which patients reach the right therapy at the right moment in their disease course.

The gap between the podium and the patient is a clinical intelligence gap which is solvable. It has already been solved, in one disease, at one health system, at scale. What remains is the will to scale it. That is a strategic choice for the entire healthcare ecosystem, and the data presented in Chicago in June 2026 have made it harder to defer.

About the author

Léon van Wouwe is Clinical Innovation Director at Volv Global SA, a clinical intelligence company applying real-world data science to diagnosis acceleration, care-gap identification, and treatment-trajectory prediction across pharma and health system partnerships.

References

Share the Post:

Press release: Cross-industry white paper calls for three paradigm shifts to unlock orphan drug development

Drive For Change white paper: three paradigm shifts to unlock orphan drug development

Jun 15, 2026 Read More

Drive For Change: a white paper on overcoming the barriers to orphan drug development

Approximately 95% of rare diseases have no approved treatment. Drive For Change identifies three structural paradigms...

Jun 15, 2026 Read More

Volv Global Poster: Real-World Endpoint Mapping and Identification of Evolving Phenotypes of Pompe Disease using Machine Learning

Machine learning applied to 3,549 confirmed Pompe disease patients – the largest machine learning study of the diseas...

Jun 2, 2026 Read More

The Gap Between the Podium and the Patient