Bringing Multi-Modal Intelligence to Early Clinical Recruitment

10 December 2025 | Wednesday | News | By Tom Stubbs, PHD – CEO of Hurdle.bio

Clinical trials often fail to recruit the participants they were designed for.

A protocol might specify a cohort defined by inflammatory markers, disease severity, a diagnostic pattern or a particular biological signature — but the moment recruitment begins, sites must make decisions using far simpler signals. Coordinators typically see a patient’s diagnosis code, age, medication list and whatever labs happen to be in the record. They rarely see the biomarker history. They cannot view longitudinal patterns. They cannot infer whether the patient is likely to meet a complex set of criteria. And they certainly cannot run new tests before screening.

This mismatch between what the protocol needs and what sites can observe is the single largest source of avoidable friction in early recruitment.

Where Recruitment Breaks Down

Take a trial targeting patients with a biomarker-positive subtype of an inflammatory condition.

Protocol requirement:

The biomarker must be measured centrally after consent.

Pre-screening reality:

Sites cannot run the biomarker test early. Most patients don’t have the biomarker measured in routine care. Many EHRs don’t even store historical test orders. So coordinators rely on guesswork — often sending forward patients who look clinically plausible but are not biomarker-positive.

The result is predictable:

Most candidates fail screening
Screening slots are wasted
Timelines drift.
Enrichment collapses

Across therapeutic areas, studies needing disease-severity bands, diagnostic subtypes, risk-enriched groups or longitudinal patterns encounter the same wall: there is no way to detect what matters using only the information sites have.

Sponsors compensate by widening criteria, adding sites or adjusting timelines — but the real issue is the information gap at pre-screening.

Hurdle is designed to make the richness of multimodal data available at the moment sites need it most, even when their EHR only provides a handful of fields. Innovate is central to this. It allows sponsors to understand, model and test eligibility logic using millions of real-world patient records, long before recruitment begins

This solution makes this practical in three ways:

1. Seeing the Real-World Consequences of Eligibility Rules Before Recruitment Begins

Using Innovate, a sponsor can simulate how millions of real patients flow through proposed criteria across multiple populations. This allows teams to ask:

If we tighten this lab cutoff, how many real patients remain?
Does this comorbidity exclusion accidentally remove half of our target group?
Do these criteria enrich the population the way we assume?
Are these criteria workable in Europe AND the U.S., or only in one?

In conventional feasibility, sponsors might get a count from one dataset or a site survey. With Hurdle, they can watch the entire population reshape itself under different rule sets — in advance. This prevents designing elegance into the protocol only to discover operational impossibility at recruitment.

2. Turning Minimal EHR Inputs Into Useful Clinical Signals

Imagine a site has a spare record:

an ICD-10 code
a handful of labs
a medication history
a brief clinical note

There is no biomarker result, no disease-severity index, no diagnostics subtype

Innovate can translate this into:

a probability the patient is biomarker-positive
a likely disease-severity band
a risk-score indicating whether the patient matches to the subgroup of scientific interest
similarity to multimodal phenotypes seen in richer datasets

Instead of guessing, a coordinator can make decisions guided by models trained of petabytes of multimodal patient data.

This is the difference between “send them to screening and hope” and “send them because they are likely to match the biological profile the study requires.”

3. Recovering Clinical Signals That Don’t Exist in the EHR

Many inclusion characteristics simply do not appear in routine healthcare data:

fibrosis severity
imaging-derived patterns
disease trajectory
diagnostic subtypes not coded in structured form

A site can’t observe these during pre-screening.

Innovate can infer these because the platform has learned how these phenomena correlate with more common signals. For example, in a respiratory trial, Innovate might estimate a patient’s likely lung-function trajectory even if the patient has never had a formal pulmonary function test recorded in the EHR — because similar patterns appear in multimodal datasets used to train the model.

This allows sponsors to find patients who genuinely fit the scientific intent, not just those with complete charts.

Why This Matters

When the earliest decisions in recruitment are made with shallow data, trials lose power, time and precision.

With Innovate:

fewer patients are sent to screening who ultimately fail
more patients who should qualify are identified
enrichment strategies hold up in the real world
diversity improves because relevant candidates are not overlooked
timelines shorten because coordinators work with better information
protocols can be refined using evidence rather than assumptions

Most importantly, the trial tests its hypothesis on the population it was designed for — not the population that happened to appear in the EHR.

Bringing Multi-Modal Intelligence to Early Clinical Recruitment

Stay Connected