Keiji AI LogoKeiji AI

Real-World Evidence guide

Clinical trial feasibility with real-world data: patient counts, criteria and the tools that run them

How to test a trial's feasibility on claims and EHR data: the questions to answer, per-criterion attrition, common pitfalls, and how feasibility tools differ.

Real-world data feasibility answers three questions before a trial opens: do enough eligible patients exist, which criteria shrink the pool most, and where those patients are treated. The core output is an attrition table with one row per criterion, run on claims or EHR data. Federated networks such as TriNetX give fast counts across many institutions; TrialMind runs the same analysis on data you already hold, in your environment, and returns the code. Trial Pathfinder (Nature, 2021) showed that broadening restrictive criteria more than doubled the eligible pool on average with little change to the treatment effect.

The three questions feasibility answers

Protocol feasibility with real-world data (RWD) is a design step, not a study. It should finish before the protocol is final, while criteria can still change.

  1. Is the population there? How many patients meet all the criteria in a given data source and time window.
  2. Which criteria cost the most? How many patients each inclusion and exclusion removes, in order.
  3. Where are they? Which sites, regions or health systems treat the eligible patients, which feeds site selection.

Real-world counts are not enrollment forecasts. Many eligible patients will not be reachable, willing or near a site. Use the counts to compare protocol versions and data sources, and use site history for enrollment speed.

The attrition table is the deliverable

A single final count hides the decisions that matter. The attrition table shows each criterion in order, the patients left after it, and how many it removed.

StepWhat to record
Index populationDiagnosis codes and the date window that define the starting cohort
Each inclusion criterionThe code list or lab threshold used, and patients remaining
Each exclusion criterionSame, plus the share of the previous step it removed
Data completenessPatients excluded because a needed field is missing, kept separate from true exclusions
Final countWith the data release and run date

Keep missing data separate. A patient with no recorded ECOG score is not the same as a patient with a poor one, and mixing the two overstates how restrictive a criterion is.

Evidence that criteria can be broadened

Trial Pathfinder (Liu et al., Nature, 2021) emulated completed advanced non-small cell lung cancer trials on EHR data from 61,094 patients. Many common exclusions, including several lab-value thresholds, had little effect on the trials' hazard ratios. Broadening restrictive criteria more than doubled the eligible pool on average.

FDA's diversity action plan guidance and its real-world evidence guidance both reward this kind of analysis. FDA's 2023 guidance treats data-source feasibility as part of study design and asks sponsors to describe the sources evaluated, including feasibility results (FDA, 2023). Save each run with its definitions so it can go into the protocol appendix.

Pitfalls that make feasibility counts wrong

  • Claims miss clinical detail. Biomarkers, stage and performance status are often absent from claims. Use EHR or registry data, or note the gap.
  • EHRs miss care elsewhere. A patient treated at two health systems appears partially in each. Lookback windows need care.
  • Codes are not diagnoses. A single diagnosis code can be a rule-out. Require two codes or a confirming test where it matters, and say which rule you used.
  • Notes hold key criteria. Prior therapy lines and reasons for stopping are often only in notes. If the tool reads notes with AI, FDA asks that the method and its validation be described (FDA EHR and claims guidance).
  • Time windows drift. "Diagnosed in the last two years" must be anchored to a fixed date and data release, or reruns will not match.

How feasibility tools differ

The main difference is whose data you are counting.

Kind of toolExampleCounts patients inFits when
Federated EHR networkTriNetX LIVEEHR data from 14,200+ sites; data stays at each source (TriNetX)You want fast counts across many institutions
Network with daily-updated dataTruvetaDaily-updated US EHR and claims dataYou want a large, current US population
Specialty data providerFlatiron HealthCurated oncology dataOncology criteria that need deep clinical detail
Analysis on your own dataTrialMindClaims, EHR, registry or genomic data you already license, in OMOP or your own schemasYou hold the data and want the definitions and code to keep

Vendor facts come from the pages linked, checked in October 2026. See TrialMind vs TriNetX and the full real-world evidence comparison.

Feasibility in TrialMind

In TrialMind, a researcher states the trial's criteria in plain language. TrialMind maps each criterion to codes in the data, shows the mapping for review, runs it, and returns the patient count, the attrition table and the SQL, Python or R that produced them. Changing a criterion and rerunning takes minutes, so protocol versions can be compared side by side.

  • Result at a customer: at a precision oncology diagnostics company with twenty-plus data scientists, feasibility studies went from weeks to minutes. The team uploads the attrition table format it already uses, and the study runs in its environment.
  • Data: claims, EHR, registry and genomic data you already license. TrialMind sells no data.
  • Deployment: inside your environment. Raw patient records are not moved.
  • Next steps from the same cohort: eligibility criteria testing against the real-world population, and the other study types in AI for real-world evidence.

Frequently asked questions

What is the best software for clinical trial feasibility analysis using real-world data?
It depends on whose data you need. TriNetX gives fast counts across a federated network of sites, Truveta and Flatiron Health supply large or oncology-specific datasets, and TrialMind runs feasibility on data you already license, inside your environment, with the code returned.
How can AI speed up a real-world data feasibility study?
By turning plain-language criteria into code lists, a query and an attrition table in one pass, so a researcher can test protocol versions without waiting on a programmer. At a precision oncology diagnostics company, TrialMind took feasibility studies from weeks to minutes.
What should a feasibility analysis deliver?
An attrition table with one row per criterion, the code list behind each row, missing-data exclusions shown separately, and the data release and date of the run.
Do broader eligibility criteria change trial results?
Often less than expected. Trial Pathfinder (Nature, 2021) found many common exclusions had little effect on hazard ratios in emulated lung cancer trials, while broadening them more than doubled the eligible pool on average.

See TrialMind on your own question

A 30-minute walkthrough on a study like yours, with sources and code you can check.