Why Relevance and Eligibility Aren't Enough: The Case for Evidence-Informed Clinical Trial Ranking
Moving from "you might qualify" to "here's what the evidence says"
The Problem with Matching Without Evidence
Clinical trial matching has come a long way. AI systems like TrialGPT can now read a patient's medical record, assess relevance, and determine which trials they're eligible for with impressive accuracy.
But here's the uncomfortable truth: relevance and eligibility aren't enough.
A patient might match dozens of relevant trials, but that doesn't mean those trials are equally good choices. Consider:
- A diagnostic imaging study that won't provide any treatment
- A trial testing a drug with poor results in similar tumors
- A Phase I dose-finding study with no proven efficacy
- A trial for a related but different cancer type
All of these might pass relevance and eligibility screening. But are they worth considering? That requires understanding the scientific evidence—something patient-trial matching alone can't tell you.
Case Study: Spinal Astrocytoma
To demonstrate this gap, we ran a benchmark patient case through both TrialGPT matching and evidence-informed ranking.
Patient Profile
- Age/Sex: 45-year-old male
- Diagnosis: Anaplastic astrocytoma of the thoracolumbar spine (unresectable)
- Complications: Severe lower extremity weakness, urinary retention
- Prior treatments:
- Radiation therapy (T10-L1)
- 11 cycles of temozolomide
- Irinotecan + bevacizumab (Avastin)
This is a heavily pretreated patient with a rare spinal cord tumor. Finding the right trial could be life-changing—or the wrong choice could waste precious time.
This case is from the TREC 2021 Clinical Trials Track, a benchmark dataset used to evaluate clinical trial matching systems.
We searched ClinicalTrials.gov (as of March 3, 2026) and found 23 active interventional trials for anaplastic astrocytoma—a typical number for a rare tumor type. After running them through eligibility screening using TrialGPT (an NIH-developed AI system for patient-trial matching), 10 trials passed the threshold.
Now comes the critical question: How should we rank these 10 trials?
TrialGPT-Only Ranking: What Eligibility Matching Shows
If we rank purely by TrialGPT score (which combines trial relevance and criteria matching, but not scientific evidence about treatments), here's what we get (showing top 7 by TrialGPT score):
| Rank | Trial | TrialGPT Score | Worth Considering? |
|---|---|---|---|
| #1 | NCT06241391 - PET/CT Imaging Study | 86 | No (diagnostic) |
| #2 | NCT01269853 - Bevacizumab Intra-arterial | 84 | Yes |
| #3 | NCT06264388 - DB107 Gene Therapy | 81 | Yes |
| #4 | NCT05698524 - Temodar + Abexinostat | 80 | Maybe |
| #5 | NCT04541082 - ONC206 Phase I | 79 | Maybe |
| #6 | NCT06975332 - Alpha Emitter Targeted Therapy | 54 | Maybe |
| #7 | NCT02800486 - Cetuximab Intra-arterial | 51 | Yes |
What's Wrong with the Top-Ranked Trials?
#1: NCT06241391 — PET/CT Imaging Study
The problem: This isn't a treatment trial at all—it's a diagnostic imaging study using a radiotracer. The patient would undergo a scan but receive no therapeutic benefit.
Why it wasn't filtered earlier: ClinicalTrials.gov classifies this as "interventional" (patients receive a radiotracer injection), so it passes the interventional study filter. But "interventional" doesn't always mean "therapeutic."
Eligibility screening sees "glioma patient" and matches. Evidence analysis sees "no treatment offered" and correctly drops the score.
#4: NCT05698524 — Temodar + Abexinostat
The nuance: This patient already received 11 cycles of temozolomide (Temodar)—and progressed. While the trial combines it with a novel HDAC inhibitor (Abexinostat), the primary drug is one the patient already failed.
Evidence analysis gave a moderate score (62) recognizing the novel combination, but not as high as trials with completely new mechanisms.
A Borderline Case: Phase I Trials
#5: NCT04541082 — ONC206 Phase I
The nuance: This Phase I trial scored 65—just above our 60 threshold. Evidence analysis noted "Phase I design limits efficacy certainty" and "no proven benefit yet," but recognized the novel mechanism.
For a heavily pretreated patient who has exhausted standard options, Phase I trials with novel mechanisms can be reasonable choices. This is a judgment call, not a clear-cut wrong answer. The evidence helps patients understand the trade-off: promising safety profile, but unproven efficacy.
Evidence-Informed Ranking: How Context Changes the Picture
When we incorporate scientific evidence—checking published literature, understanding drug mechanisms, and considering the patient's treatment history—the ranking changes:
| Rank | Trial | ClinTrialFinder Score | TrialGPT Rank |
|---|---|---|---|
| #1 | NCT01269853 - Bevacizumab Intra-arterial | 75 | was #2 |
| #2 | NCT06264388 - DB107 Gene Therapy | 72 | was #3 |
| #3 | NCT02800486 - Cetuximab Intra-arterial | 70 | was #7 |
| #4 | NCT04541082 - ONC206 Phase I | 65 | was #5 |
Why These Trials Rank Higher
#1: NCT01269853 — Bevacizumab Intra-arterial
Why it's promising: While the patient previously received systemic bevacizumab (Avastin), this trial uses a novel intra-arterial delivery method—injecting directly into cerebral arteries. This approach can achieve much higher drug concentrations at the tumor site than systemic administration.
Evidence analysis recognized this as a meaningfully different approach worth considering.
#3: NCT02800486 — Cetuximab Intra-arterial
Why TrialGPT scored it low: The trial requires "pathology confirmed EGFR overexpression" (not mentioned in patient record) and "Karnofsky ≥60%" (patient has severe weakness and urinary retention, suggesting lower performance status). These eligibility concerns dropped the score to 51.
Why evidence analysis boosted it: The treatment mechanism—EGFR-targeted therapy via intra-arterial delivery with re-irradiation—has moderate evidence supporting efficacy in gliomas. For a patient who has exhausted standard options, this novel approach may be worth discussing with their doctor, even if eligibility requires clarification.
This illustrates how the two scores complement each other: TrialGPT flags real eligibility concerns, while evidence analysis recognizes therapeutic potential.
Summary: How Evidence Changed Rankings
| Metric | TrialGPT-Only | Evidence-Informed |
|---|---|---|
| #1 ranked trial is worth considering | No (diagnostic study) | Yes (novel therapy) |
| Diagnostic study correctly filtered | No (ranked #1) | Yes (score dropped to 35) |
| Cetuximab trial rank | #7 (low visibility) | #3 (promoted by evidence) |
How Evidence-Informed Ranking Works
Evidence-informed ranking goes beyond "does this patient match the criteria" to ask:
- Is this a treatment trial? Diagnostic, observational, and sample-collection studies are filtered appropriately.
- Has the patient already tried this approach? Offering failed drug classes again is deprioritized.
- What does the literature say? Drugs with promising efficacy data in similar tumors are prioritized.
- Is this a novel mechanism? For heavily pretreated patients, new approaches may offer more hope than incremental improvements.
- What's the risk/benefit profile? Phase, study design, and known side effects are considered.
Behind the Scores: The Drug Evidence
To show exactly how evidence affects ranking, here's what the system found for each drug when searching PubMed and Perplexity:
| Trial | Drug | Evidence Source | Key Finding | Score Impact |
|---|---|---|---|---|
| NCT06241391 | Ga-68 PSMA | PubMed | Diagnostic imaging only—not a treatment. "No direct evidence for treatment efficacy." | 86 → 35 |
| NCT01269853 | Bevacizumab IA | PubMed | Intra-arterial delivery achieves higher tumor concentrations than systemic. Novel delivery method despite prior Avastin use. | 84 → 75 |
| NCT02800486 | Cetuximab IA | Perplexity | Moderate evidence for EGFR-targeted therapy in gliomas. Innovative intra-arterial delivery with re-irradiation. | 51 → 70 |
| NCT06264388 | DB107-RRV + FC | Perplexity | Gene therapy targeting DGM7+ patients (20-30% of HGG). Limited efficacy data available. | 81 → 72 |
| NCT04541082 | ONC206 | Perplexity | No clinical trial results for anaplastic astrocytoma. Phase I ongoing, preclinical promise only. | 79 → 65 |
| NCT05698524 | Temodar + Abexinostat | Perplexity | Novel HDAC inhibitor combination, but patient already failed Temozolomide. Moderate novelty. | 80 → 62 |
Implications for Clinical Trial Matching
This case study illustrates a fundamental limitation of matching systems that rely solely on relevance and eligibility criteria:
TrialGPT answers: "Is this trial relevant, and could this patient enroll?"
Evidence-informed ranking answers: "Is this trial worth considering for this patient?"
For patients with complex treatment histories—especially in oncology—the difference is critical. The "best" trial isn't always the one with the highest eligibility match. It's the one most worth considering given the patient's specific situation.
Conclusion
AI-powered eligibility screening is a major advance. But stopping there leaves patients with a list of "maybe eligible" trials without guidance on which ones actually make sense for their situation.
Evidence-informed ranking closes this gap by incorporating scientific literature, treatment mechanisms, and clinical context. In our test case, it surfaced novel therapies as promising options—trials that eligibility-only ranking had ranked lower.
For cancer patients searching for clinical trials, the question shouldn't just be "what am I eligible for?" It should be "what's worth considering given my situation?"
That's the question evidence-informed matching tries to answer.
This analysis was conducted using ClinTrialFinder, an evidence-informed clinical trial matching tool. The patient case is from the TREC 2021 Clinical Trials benchmark dataset.