How to Assess Biomedical Research Question Feasibility with AI
Linda
Learn how to assess biomedical research question feasibility with AI by checking novelty, measurement readiness, population access, statistical inputs, operational barriers, and fatal flaws.
A biomedical research question can be scientifically interesting and still be a poor study question. Before building a protocol, investigators need to know whether the question is sufficiently novel, measurable, statistically plan-able, operationally realistic, ethically manageable, and capable of producing decision-relevant evidence.
This guide shows a practical AI-assisted feasibility workflow using a real Noah AI case:
Can plasma p-tau217 replace amyloid PET for identifying cognitively unimpaired adults with amyloid pathology for Alzheimer's disease prevention trials?
The point of the exercise was not to force a yes/no answer. It was to determine whether the question should move forward as written, be refined, or remain not yet ready for formal research-plan development.
Testing note: We tested Noah AI directly using Deep Research. The workflow below reflects the actual structure of that run.
1. What Does "Research Question Feasibility" Actually Mean?
Research feasibility is the assessment step between an interesting idea and a formal research plan. It asks whether the question is ready to become a study.
- Novelty: Is the main question still unanswered?
- Scientific rationale: Is there enough evidence to justify the hypothesis?
- Measurement: Can the exposure, biomarker, comparator, and outcome be measured reliably?
- Population: Can the intended participants realistically be recruited?
- Statistics: Are the inputs needed for design and power calculations available?
- Operations: Can the workflow actually run across sites, labs, and participants?
- Ethics: Are the risks of classification, disclosure, exclusion, or confirmatory testing manageable?
- Decision value: Would the study produce new evidence that changes a scientific or development decision?
2. Research Feasibility vs Research Planning
| Feasibility assessment | Research plan |
|---|---|
| Should this question move forward? | How should the study be executed? |
| Checks whether the question is already substantially answered | Builds the protocol around an accepted question |
| Tests measurement, population, statistical, operational, and ethical readiness | Defines design, procedures, endpoints, analysis, milestones, and implementation |
| Looks for fatal flaws | Works around assumptions already judged acceptable |
| Ends with GO / REFINE / NOT YET READY | Ends with an executable study plan |
3. Real Case: Can Plasma p-tau217 Replace Amyloid PET?
The original question sounds simple, but one word creates most of the difficulty: replace.
"Replace amyloid PET" could mean several different things: diagnostic equivalence, noninferior eligibility classification, PET reduction, blood-first prescreening, trial enrichment, or complete elimination of PET. Those claims are not equivalent.
4. Step 1 — Clarify the Research Claim
The first feasibility task is to translate broad language into a decision that can actually be tested.
- the cognitively unimpaired population;
- the exact p-tau217 assay and platform;
- the PET or CSF reference standard;
- the intended use of the blood result;
- the false-negative tolerance;
- the PET-reduction objective;
- how indeterminate results are handled;
- what outcome would make the strategy acceptable.
5. Step 2 — Check Whether the Basic Question Has Already Been Answered
The basic scientific premise is already substantially supported: plasma p-tau217 can identify amyloid pathology in cognitively unimpaired populations and can enrich a population before confirmatory PET or CSF.
The more decision-relevant question is whether a specified assay and specified plasma-first workflow can reduce confirmatory PET while keeping the rate of incorrectly excluded amyloid-positive participants below a prespecified limit.
| Research claim | Feasibility interpretation |
|---|---|
| p-tau217 is associated with amyloid pathology | Low novelty; substantially answered. |
| p-tau217 can enrich a CU population before PET/CSF | Substantially supported. |
| A specific assay can safely exclude blood-negative candidates | Still unresolved and decision-relevant. |
| A plasma-first strategy can improve PET utilization and trial screening | Suitable for prospective implementation research. |
| p-tau217 can universally replace PET | Not established. |
6. Step 3 — Evaluate Measurement Feasibility
A biomarker name is not the same as a standardized test. "p-tau217" can be measured using different immunoassays, mass-spectrometry methods, calibrators, antibodies, units, and platforms.
A feasible study therefore needs to lock one primary assay, one reporting unit, a central laboratory or harmonized network, collection and shipping rules, quality control, and a prespecified cutoff strategy.
7. Step 4 — Test Population Feasibility
A prevention-trial population may differ from a memory-clinic population in age, amyloid prevalence, APOE distribution, comorbidity, geography, and recruitment source. These differences can change PPV, NPV, PET volume, screen-failure rates, and recruitment yield.
8. Step 5 — Choose a Decision-Relevant Endpoint
| Endpoint | Why it matters |
|---|---|
| Sensitivity / specificity | Shows classification performance at the chosen cutoff. |
| PPV / NPV | Shows the reliability of positive and negative screening decisions. |
| False-negative exclusion | Measures eligible amyloid-positive candidates lost because of the blood-first strategy. |
| PET scans avoided | Measures the main operational benefit. |
| Number needed to screen | Captures total screening burden. |
| Randomization yield | Shows whether the pathway actually improves trial recruitment. |
The most defensible primary decision framework is false-negative control + meaningful PET reduction.
9. Step 6 — Check Statistical Feasibility
At this stage, AI should identify the inputs needed for sample-size planning rather than invent the final sample size.
- amyloid prevalence;
- assay-specific sensitivity and specificity;
- PPV and NPV;
- allowable false-negative exclusion rate;
- target PET reduction;
- confidence-interval precision;
- assay variability;
- missing biomarker data;
- site heterogeneity;
- confirmatory-testing rate;
- recruitment and randomization yield.
A missing parameter should remain visible as an uncertainty, not be filled with an invented number.
10. Step 7 — Run a Fatal-Flaw Analysis
This is the most important part of the case. Noah identified which weaknesses could make the study uninterpretable if left unresolved.
- "Replace PET" is too broad.
- The basic classification question has already been substantially answered.
- Standalone PPV is insufficient as the main decision criterion.
- False negatives may exclude eligible amyloid-positive participants.
- Assay thresholds do not transfer between platforms.
11. Step 8 — Refine the Research Question
The feasibility assessment resulted in a REFINE decision. The original question was scientifically relevant but too broad to support a clearly defined study.
Original Research Question — Before Noah AI
Can plasma p-tau217 replace amyloid PET for identifying cognitively unimpaired adults with amyloid pathology for Alzheimer's disease prevention trials?
Main problems identified:
- "Replace" does not define the intended clinical or screening use.
- The p-tau217 assay and decision thresholds are unspecified.
- False-negative exclusion of amyloid-positive participants is not addressed.
- The expected reduction in PET use is not defined as a measurable outcome.
Final Refined Research Question — After Noah AI
In cognitively unimpaired adults being screened for an Alzheimer's disease prevention trial, can a specified plasma p-tau217–first pathway reduce confirmatory amyloid PET use while maintaining a prespecified acceptable rate of incorrectly excluding amyloid-positive participants?
What Changed?
| Original Question | Refined Question |
|---|---|
| Replace amyloid PET | Use a plasma p-tau217–first screening pathway |
| Unspecified biomarker test | Require a specified assay and cutoff strategy |
| No explicit false-negative safeguard | Define an acceptable false-negative exclusion rate |
| Broad replacement claim | Measure PET reduction while preserving screening accuracy |
Why the refined question is more feasible: It replaces a broad diagnostic-substitution claim with a testable screening strategy. The revised question identifies the target population, intended use, principal operational outcome, and key safety constraint.
The exact assay, numerical thresholds, confirmation pathway, and statistical assumptions still need to be finalized during research-plan development.
Final feasibility decision: REFINE — the research concept can proceed toward formal planning after these specifications are resolved.
12. Step 9 — Make the Final Feasibility Decision Explicit
REFINE — feasible after specific changes.
The refined blood-first research concept is feasible enough to justify formal development. The original unrestricted claim is not.
What supports feasibility:
- strong human evidence that plasma p-tau217 contains useful amyloid-related information;
- direct prevention-relevant evidence in cognitively unimpaired adults;
- existing blood-first and sequential confirmation models;
- technically feasible centralized biomarker testing;
- a clear operational reason to reduce PET burden.
What prevents proceeding as written:
- the original claim is too broad;
- one assay and cutoff strategy are not specified;
- the false-negative tolerance is undefined;
- the confirmation pathway is incomplete;
- population-specific prevalence and recruitment assumptions are not fixed;
- the main endpoint needs to reflect both screening safety and PET reduction.
Assess your biomedical research question with Noah AI
Use Noah AI to test novelty, evidence sufficiency, measurement readiness, population access, endpoint quality, statistical assumptions, operational barriers, and fatal flaws before committing to a full research plan.
Noah AI can be used for free, includes free credits, and requires no credit card.
13. GO vs REFINE vs NOT YET READY
| Decision | When it fits |
|---|---|
| GO | The question is sufficiently novel, specific, measurable, statistically plan-able, operationally realistic, and ready to enter formal study planning. |
| REFINE | The underlying idea is feasible, but wording, assay, endpoint, population, comparator, verification strategy, or operational assumptions need explicit changes. |
| NOT YET READY | Critical evidence, measurement validity, infrastructure, population access, or planning inputs are still missing. |
14. A Practical AI-Assisted Feasibility Workflow
- Clarify the research claim.
- Check whether the basic question is already answered.
- Confirm the scientific rationale.
- Test population feasibility and generalizability.
- Lock the measurement strategy.
- Choose a decision-relevant endpoint.
- Identify statistical inputs and missing assumptions.
- Assess technical and operational barriers.
- Review ethical implications.
- Run a fatal-flaw analysis.
- Refine the question if needed.
- Decide whether it is ready for Research Plan Development.
15. Common Mistakes
Starting with Protocol Design
If the question is not yet feasible, adding more protocol detail does not solve the underlying problem.
Treating "Interesting" as "Novel"
A topic can remain scientifically important even when the basic classification or association question has already been substantially answered.
Using Broad Verbs
Words such as "replace," "predict," "validate," and "improve" often hide multiple different estimands and operational decisions.
Choosing AUC as the Only Endpoint
Discrimination does not show whether a real screening pathway safely excludes candidates or materially reduces PET use.
Inventing Sample-Size Inputs
If prevalence, false-negative tolerance, assay variability, recruitment rate, or attrition are unknown, they should remain assumptions or pilot-data requirements.
Ignoring Platform Differences
Biomarker cutoffs are not automatically transferable between assays, laboratories, and populations.
Skipping Verification of Negative Results
If only biomarker-positive participants receive the reference standard, false-negative performance can be badly biased.
Forcing a Binary Yes/No Answer
REFINE is often the correct scientific answer.
16. When Is a Question Ready for Research Plan Development?
A refined question is usually ready when the team can answer yes—or partly, with a controlled plan—to all of the following:
- Is the question specific?
- Is the remaining novelty defensible?
- Is the primary endpoint measurable?
- Can the target population be accessed?
- Is the reference standard available?
- Can the statistical inputs be justified?
- Are major technical barriers manageable?
- Is the workflow operationally plausible?
- Are ethical issues manageable?
- Can generalizability be assessed rather than assumed?
17. Related Research Guides
- Best AI Tools for Medical Research Workflow Automation (2026)
- How to Find Sources for a Research Paper with AI: A Medical Literature Workflow
- How to Use Noah for Medical Literature Review
- PubMed Search with AI: How Life Science Professionals Find and Analyze Biomedical Evidence Faster
- How to Search PubMed with AI: A Step-by-Step Guide
18. FAQ
Can AI determine whether a biomedical research question is feasible?
AI can help retrieve current evidence, test novelty, structure feasibility dimensions, surface missing assumptions, and identify fatal flaws. Researchers still need to review the evidence and own the final scientific decision.
What makes a biomedical research question feasible?
A feasible question is specific, not already fully answered, measurable, statistically plan-able, operationally realistic, ethically manageable, and capable of generating decision-relevant evidence.
What is a fatal flaw in a research question?
A fatal flaw is a problem that would make the proposed study unable to answer its intended question if left unresolved.
Does REFINE mean the research idea is weak?
No. It means the underlying idea can move forward after specific changes make the question more interpretable and operationally realistic.
Should AI calculate a final sample size during feasibility assessment?
Not until the required quantitative assumptions are supported. Before that, the correct output is a list of the parameters that need evidence or pilot estimates.
Why is false-negative verification important in biomarker screening studies?
If biomarker-negative participants never receive the reference standard, the study may not know how many truly eligible participants were incorrectly excluded.
19. Final Takeaway
A good research question is not only interesting—it must be specific, sufficiently novel, measurable, statistically supportable, operationally realistic, and capable of producing decision-relevant evidence.
The p-tau217 example shows why feasibility assessment should happen before formal study planning. The original wording—whether plasma p-tau217 can "replace amyloid PET"—was too broad and implied a level of substitution that current evidence does not justify.
The underlying research concept remains feasible. By narrowing the question to a specified blood-first strategy, locking the assay and reference standard, defining acceptable false-negative exclusion, retaining confirmatory testing where needed, and measuring PET reduction and recruitment consequences prospectively, the question becomes suitable for formal research-plan development.
Try Noah AI for Research Feasibility Assessment
Use Noah AI to assess biomedical research questions before protocol development by mapping current evidence, novelty, measurement readiness, statistical inputs, operational barriers, and fatal flaws.
Free credits are available, and no credit card is required.