APPENDICURE

APPENDICURE: A CLEARER PATH – AI IN APPENDIX CANCER, PART 6

For most appendix cancer patients, the question of clinical trials comes up sooner than expected.

Sometimes it comes up right at diagnosis, when the standard treatment path is not clearly defined and the options on the table feel extrapolated from other cancers. Sometimes it comes up after surgery, when the next decision is whether to pursue systemic therapy and, if so, what kind. Sometimes it comes up later, after a recurrence, when the question shifts from what is standard to what is possible.

In all of those moments, the same problem tends to surface. There are trials out there. Some of them might be right for you. But finding them, understanding whether you are actually eligible, and getting to a site that can enroll you is often a maze that patients are expected to navigate mostly on their own.

This gap is one of the most consequential barriers in rare cancer care, and it is one of the few where AI is starting to make a real difference.

Why Trial Access Has Been Such a Persistent Problem

Clinical trials are how new treatments become standard treatments. For common cancers, that process runs on a large and well-resourced system, with dedicated recruitment teams, established trial networks, and enough patient volume that trials can fill at a reasonable pace.

For rare cancers, the system works differently, because the math is different. Appendix cancer affects a small number of people each year, and those patients are spread across the country. A trial running at a single academic center may need to reach a national or even international patient population just to enroll enough participants. That reality shapes everything about how trials for this disease get designed, run, and populated.

On the patient side, the experience is often the mirror image of that logistical challenge. You hear about a trial through a support group, or you find something on ClinicalTrials.gov, or your oncologist mentions a study at a center across the country. You try to understand the eligibility criteria, which can run to several pages of highly technical language. You try to figure out whether your specific pathology, your prior treatments, your lab values, and your disease burden fit what the trial is looking for. And often you discover, weeks or months into the process, that a single exclusion criterion rules you out.

For patients with progressing disease, time spent pursuing the wrong trial is time not spent on the right one.

THE SCALE OF THE MISMATCH Fewer than one in twenty adult cancer patients enrolls in a clinical trial. For rare cancers, eligibility screening is one of the largest bottlenecks. Many patients never learn that a trial exists for which they would qualify and many trials close without meeting their enrollment targets.

Where AI Is Beginning to Change This

The core task of matching a patient to a trial is, in structural terms, the kind of problem large language models are well suited to. It involves reading long, densely written eligibility criteria, reading a patient’s medical record, and making careful, criterion-by-criterion judgments about whether the two align. Historically, that work has been done by research coordinators, often manually, and often under significant time pressure.

Recent research has demonstrated that language models can now perform this matching at near-expert accuracy. A framework called TrialGPT, developed by researchers at the National Institutes of Health, evaluates a patient’s note against a trial’s eligibility criteria one criterion at a time and then produces a consolidated judgment about whether the patient is likely to qualify. In published evaluations, the system reached criterion-level accuracy close to physician performance and reduced screening time by more than 40 percent in a real-world matching task. It’s important to note performance on rare cancers specifically has not been separately validated

That last number is where the clinical relevance starts to become obvious. Screening time is not an abstract efficiency metric. It is the thing standing between a patient and the knowledge that a trial exists for them. Cutting it nearly in half changes what is possible for patients with rare diseases, where the universe of relevant trials is small and the cost of missing one is high.

More recent work has extended this approach into real-world hospital systems, integrating trial matching directly into electronic health records so that eligibility screening happens automatically as a patient’s chart is updated. That shift, from a patient having to go find trials to the system proactively surfacing them, is the change that matters most for rare cancer patients.

How This Plays Out for One Patient

Imagine a patient with high-grade mucinous adenocarcinoma who has completed cytoreductive surgery and HIPEC and is now considering systemic therapy. Under the current model, finding a relevant trial typically means a combination of things. Their oncologist may know of one or two studies at their home institution. The patient may search ClinicalTrials.gov on their own, using search terms that are often too narrow or too broad. A patient advocacy organization may maintain a list, but it will not be exhaustive, and it will not be specific to the patient’s particular pathology, treatment history, and molecular profile.

Under an AI-assisted model, the workflow could look very different. A system reviewing the patient’s full record could identify that they have completed first-line surgical treatment, that their tumor carries a specific molecular feature, that they have a particular pattern of residual disease, and that their lab values and performance status are within typical trial windows. It could then surface a ranked list of trials for which the patient likely qualifies, with a plain-language explanation of why each trial is a potential fit and which criteria are borderline or require confirmation.

The patient and their oncologist still make the decision. The AI is not enrolling anyone. What it is doing is replacing a blind search with a guided one, and replacing a best-effort manual review with a systematic one that covers trials the patient and their clinician might not otherwise have heard of.

WHAT AI DOES AND DOES NOT DO HERE AI-assisted trial matching is a screening and surfacing tool. It identifies trials that a patient is likely to qualify for and flags the criteria that need verification. Final eligibility is always confirmed by the trial team and the decision to pursue any particular trial always belongs to the patient and their care team. The goal of these tools is to make sure that relevant trials reach the patient in the first place.

Why This Matters More for Rare Cancers Than for Common Ones

For common cancers, a patient at a large academic center has a reasonable chance of being connected to a relevant trial through their existing care team. The volume is high, the trials are familiar, and the infrastructure is in place.

For appendix cancer, that assumption breaks down. The trials are fewer. The eligibility criteria are more specific, often hinging on exact subtype and grade classifications that, as earlier posts in this series have discussed, are not always consistently assigned. The sites that run these trials are concentrated at a relatively small number of high-volume centers. A patient being treated at a community hospital may have no systematic way to learn about a trial across the country, let alone assess whether they would qualify for it.

This is exactly the kind of problem where AI can close a gap that human systems have struggled to close on their own. Not because physicians and coordinators are not doing their jobs, but because the scale of the search is beyond what any individual can realistically cover. A tool that can read thousands of active trials and match them against a specific patient’s history in minutes, rather than over the course of weeks, is the difference between a patient hearing about an option while there is still time to act on it and hearing about it after the window has closed.

The Registry Problem, and Why It Matters for Trial Design

There is a second piece of this that deserves attention, even though it sits one step behind the patient-facing experience. Clinical trials get designed based on what researchers know about a disease. For rare cancers, that knowledge base is often thin, because the patient data is scattered across institutions, stored in formats that do not talk to each other, and rarely aggregated into a form that can guide new trial design.

Patient registries, which are structured databases of de-identified patient information across many institutions, are one of the most important tools for fixing this. A well-built appendix cancer registry would make it possible to understand, at a population level, what patterns of disease exist, how they respond to different treatments, and where the most urgent gaps in current care are. That information is what allows researchers to design trials that ask the right questions for the right patients.

AI has a role here too, in making registries more useful. Extracting structured data from unstructured clinical notes, linking pathology across institutions, and identifying patient cohorts with specific molecular or clinical features is work that AI can now accelerate significantly. The end result is not just better trials, but trials that are designed to actually serve the patients who most need them.

What to Ask About Right Now

AI-driven trial matching is not yet a standard part of most oncology practices, but the tools are becoming available, and patients can ask about them. More immediately, there are concrete steps that can make trial access easier regardless of whether an AI system is in the loop.

A clearly documented pathology report with subtype, grade, and any available molecular profiling is the foundation of any trial match. So is a complete and accessible treatment history. The more precisely your record reflects your actual disease, the better any matching process, whether human or machine, is going to perform. Having a second opinion at a high-volume appendix cancer center is often the fastest way to get both of those things established.

It is also worth asking your care team directly whether they have searched for relevant trials, and whether they have access to a matching tool or service. Several cancer centers now offer concierge trial-matching support as part of their intake process. Some nonprofit organizations, including Appendicure, can help connect patients with resources that go beyond what an individual oncologist has the time or reach to provide.

The Direction Things Are Moving

Trial access for rare cancer patients has been a quiet crisis for a long time. It does not get the attention that treatment innovation gets, because it is not about a new drug or a new technique. It is about whether the drugs and techniques that already exist, or that are being tested right now, can actually reach the patients who might benefit from them.

AI is not going to solve that on its own. But it is changing what is possible in a way that matters. It is making it feasible to search a larger universe of trials, to screen more carefully against specific eligibility criteria, and to surface options that would otherwise have been missed. For a disease where options are already limited, that kind of expansion is not incremental. It is the difference between a patient having access to a trial and never knowing it existed.

For a cancer community that has spent years being told there is not enough data, not enough trials, and not enough patients to justify better answers, the arrival of tools that can change that equation is significant. Appendicure will continue tracking this space, and will continue pushing for the kind of infrastructure, including registries, subtype-specific trial design, and accessible matching, that turns the promise of this technology into real outcomes for patients.

Questions to Ask Your Doctor

• Are there any clinical trials currently open that you think could be relevant to my specific subtype, grade, and treatment history?
• Has my record been reviewed against active trials using a trial-matching tool or service, or is this something done manually?
• Is my pathology report detailed enough to establish eligibility for trials that require specific subtype or molecular criteria?
• Has my tumor been tested for molecular features that could qualify me for biomarker-driven trials, including basket trials open to multiple cancer types?
• If there is a promising trial at a different center, can you help coordinate a referral or a second opinion to establish eligibility?
• Are there registries or patient databases I should consider enrolling in, even if I am not currently pursuing a specific trial?

About This Series

This is the sixth and final post in Appendicure’s series on how artificial intelligence is beginning to intersect with appendix cancer. Earlier posts explored AI’s role in guiding surgical decisions, shortening the path to diagnosis, defining tumor subtypes, personalizing treatment, and monitoring for recurrence. The through-line across all six is the same. For a disease that has been held back for years by scarcity of data and scarcity of attention, tools that can make better use of what does exist are not a side story. They are part of how the standard of care actually improves.

Read Part 1: Appendix Cancer and Surgical Decisions: How AI May Help Guide the Hardest Choice

Read Part 2: Why Appendix Cancer Is So Often Missed, and What Could Change That

Read Part 3: Understanding Tumor Types in Appendix Cancer, and Why They Matter

Read Part 4: Personalized Treatment in Appendix Cancer: Why One Plan Does Not Fit All

Read Part 5: Watching for What Comes Back: How AI Could Change Recurrence Monitoring in Appendix Cancer