Clinical Trials Need a Quality Copilot

Surface issues in protocols, validate created content, and handle amendments with ease

Validate Source
Consistency verification across clinical trial documents, starting with the protocol.
Create Content
Generate study documents from the protocol, trace back to source.
Handle Amendments
Analyze amendment changes, apply changes to created documents and validate output.
HumanTrue Quality Loop






We've studied thousands of real-world protocols

Using our AI pipeline, we've analyzed thousands of publicly available clinical trial protocols. We've found inconsistent language, ambiguities, and knowledge gaps. All of which compromise the ability to run a trial.

Unlike other approaches that treat the protocol as a simple input-output problem, we believe the complexity and stakes of clinical trials demand a quality copilot that can help human teams understand and improve their protocols.

INCONSISTENCY
Protocol p. 20

An estimated 15 to 24 patients will be enrolled in Phase 1b, depending on dose escalation and expansion.

Protocol p. 27

An estimated 32 patients will be enrolled in Phase 1b, depending on dose escalation and expansion.

How many patients, exactly?

The synopsis says 15-24 patients will be enrolled in Phase 1b, but the sample size section says 32.

INCONSISTENCY
§ 4.1.1.3.2

The dose ... in each chemotherapy combination will be 24 mg...

§ 4.1.1.3.2

An initial three participants will receive 80 mg ... in combination with chemotherapy

Inconsistent dosing for the same drug combo

The same safety run-in section gives two different starting doses, 24 mg and 80 mg, for the same drug combination.

AMBIGUITY
Protocol pp. 8, 20

Treatment with study drugs may be continued until ... disease progression. ... Added confirmed tumor progression according to iRECIST to the reasons for withdrawal from study treatments.

§ 10.3.1.1

... iRECIST requires the confirmation of progression and uses the terms iUPD (unconfirmed progression) and iCPD (confirmed progression). Confirmatory scans should be performed at least 4 weeks, but no longer than 8 weeks after iUPD.

Disease progression — which kind?

The synopsis uses "disease progression" as a treatment-stopping trigger, but that term can mean any radiographic progression or only iRECIST-confirmed progression.

INCONSISTENCY
Protocol p. 4

Patients who have discontinued study treatment without progression will continue to be followed for progression. Every attempt should be made to gather all information every 2 months (radiological scans and survival)...

§ 10.1

Patients who come off therapy and do not have objective progressive disease should be followed every 2 months after discontinuation from study treatment until death, including radiologic examinations every 6 weeks...

Imaging cadence collision

The synopsis says post-treatment imaging happens every 2 months, but the efficacy section says every 6 weeks.

INCONSISTENCY
Protocol p. 17

For participants enrolling in HNSCC dose optimization cohort the maximum treatment duration is approximately 2 years, or up to 17 cycles for both GSK3359609 (feladilimab) and pembrolizumab.

§ 4.2

The maximum treatment duration for participants enrolled in HNSCC Q6W dosing cohort is expected to be approximately 2 years, up to 18 cycles for both GSK3359609 (feladilimab) and pembrolizumab.

17 cycles or 18?

The synopsis sets the duration of appx. 2 years to be 17 cycles, but the treatment-duration section sets it at 18.

INCONSISTENCY
§ 12.2.7

Time to event variable: 6b. Time to first 50% reduction in BPI average pain score

§ 12.2.7

Derivation and Details: 6b. For the patients with a 50% reduction at a visit in the treatment phase, time = days from the date of the visit that the earliest 30% reduction is observed to the randomization date.

Named 50%, timed at 30%

The endpoint is labeled "time to first 50% pain reduction," but the calculation qualifies patients on their 50% reduction date while measuring the clock from their earlier 30% reduction date (i.e., the qualifying event and the timed event are different).

AMBIGUITY
§ 4.3

In Part 1A and Part 2A, if at any dose level a participant prematurely discontinues study treatment before the completion of the 28-day DLT observation period ... a replacement participant may be enrolled ... if <3 participants complete the DLT observation period at that dose level.

§ 4.3

In those Part 1A dose levels that have a single participant enrolled, a replacement participant will be enrolled if <1 participant prematurely discontinues study treatment before the completion of the 28-day DLT period.

Replacement rule, reversed

At a dose level with only one participant, the replacement rule says to enroll a replacement only when fewer than one participant drops out meaning a replacement is triggered when the participant completes the period, and no replacement happens when the participant actually drops out.

AMBIGUITY
§ 3.3.2

Women of childbearing potential ... and men able to father a child must be ready and able to use ... birth control ... until 6 months and 12 days after last dose.

§ 4.2.2.3

The contraception requirements for male patients ...3 months and 10 days. For female patients ... 6 months and 10 days

Contraception duration mismatch

The protocol enrollment criteria and amendment log give different durations for the required contraception rules during the study.

This problem is pervasive in the industry...

Tufts CSDD conducted a recent study and cited this quote from a survey participant:

We often are receiving protocols, lab manuals, IB, pharmacy manuals, ICF—that are not in agreement with each other or have glaring mistakes that are not corrected. Study procedure timelines don't match the lab processing for those study procedures, or data collection guides don't match the visit schedule. We often see inclusion and exclusion criteria that contradict each other. All of this adds confusion, unavoidable PDs (protocol deviations), missing data, out of window study visits, but most importantly it can impact patient safety and data quality.

Source: Harper, B., Ford, R. M., Krayem, R., Nomizu, R., Andrus, J., & Getz, K.
Characterizing the Protocol-Guided Source Preparation Process at Investigative Sites.


Trial documents now move at machine speed.
HumanTrue makes sure errors don't.

AI accelerates content creation and systems configurations, so that studies can be built in a fraction of time. But an inconsistency that once slowed a single site now spreads to every connected system at machine speed - unreviewed, everywhere at once.

76%

of Phase I-IV protocols now require at least one substantial amendment

Tufts CSDD
~45%

of those amendments are considered avoidable

Tufts CSDD, 2016
$141K-$535K

direct cost of a single substantial amendment, by complexity

Tufts CSDD
~3 months

average added timeline per amendment

Tufts CSDD

Amendments don't stop at the protocol. Neither do we.

On average, a phase 3 trial carries 5-7 substantial amendments with each one affecting dozens of downstream artifacts, decisions and regulated systems. HumanTrue is the verification layer that is with you end-to-end, propagating the needed changes with built in re-verification before final release.

HumanTrue provides the details that you need to operationalize an amendment.

Visual diff

Visual diff

A paragraph-by-paragraph delta of the protocol itself, with attribution to author and version.

Dependency map

Dependency map

Every downstream artifact, system, and site instruction the amendment touches visualized as a propagation tree.

Targeted regeneration

Targeted regeneration

Affected sections of every downstream document regenerated against the new source with citations.

Re-verification report

Re-verification report

Every artifact in the chain is re-checked against the new source, protecting against silent drift between the protocol and downstream documents.

Staff your trial like you have a blank check.

HumanTrue gives a lean team the roster of a global sponsor. Every protocol, site, and amendment runs through the same verified semantic model - so a handful of people get the rigor that used to take a department of specialists, on every protocol, ICF, IRB submission, and amendment alike. The expertise doesn't have to scale with your headcount. It's already built in.

  • Per-protocol QA at scale. AI-assisted verification built-in; human certified review available.

  • Operational services. No internal build needed. You get protocol verification, amendment governance, and startup-package generation from day one.

  • From submission to site. Everything from regulatory filings to site-executable materials is produced from one verified semantic model, with full provenance on every output.

  • Speed-to-activation. Coherent protocols mean faster activation and far less avoidable cost.

  • Partner-grade trust. CRO integrations, IRB references, and source in every output. The deliverable a sponsor partner expects.