MASH Trials Are Screening Out the Patients They Were Designed to Treat

Picture the enrollment coordinator at a hepatology center reviewing a stack of referrals for a Phase 3 MASH trial. Each patient has been flagged by their gastroenterologist using FIB-4 and transient elastography, the same tests recommended in clinical practice guidelines, the same thresholds used to route these patients toward pharmacological treatment. On paper, every one of them belongs in this trial. Then she opens the protocol’s inclusion criteria and starts crossing names off the list. The cutoffs are different. The acceptable test sequence is different. Some patients need a liver biopsy to confirm what the NITs already said. By the time she reaches the bottom of the stack, she has a referral pool of thirty and a screen-eligible list of four.

This scene repeats across MASH trial sites worldwide, and the numbers bear it out. The SYNERGY-NASH Phase 2b trial, investigating tirzepatide for MASH, faced screen failure rates severe enough that the sponsor was forced to amend its inclusion criteria mid-enrollment, introducing the FibroScan-AST (FAST) score (requiring FAST greater than 0.35) and raising the AST inclusion threshold from 20 to 23 U/L just to keep the trial viable. That kind of mid-stream protocol amendment carries its own cost: regulatory questions, timeline slippage, site retraining. The amendment is an admission that the original design did not map onto the patients who actually exist in clinics.

The Threshold Problem Nobody Wants to Name

The core mechanism driving this crisis is straightforward, but it is almost never stated plainly in the trial design literature. Clinical practice and clinical trials ask non-invasive tests to do fundamentally different jobs, and they have been handed incompatible toolkits to do them.

In clinical practice, the goal of fibrosis staging with tools like FIB-4, the Enhanced Liver Fibrosis (ELF) test, or vibration-controlled transient elastography (VCTE, commonly delivered via FibroScan) is risk stratification across a population where prevalence is high and intervention thresholds are forgiving. A clinician using FIB-4 wants to separate the patients who need a hepatologist from the patients who can be managed in primary care. The threshold is calibrated for sensitivity: catch as many true positives as you can, accept some false positives, because the cost of missing a fibrotic patient is higher than the cost of an unnecessary referral. A prospective cohort study of 3,378 participants using the ELF test, FIB-4, and NAFLD fibrosis score confirmed that these tools perform well as population-level filters, particularly for ruling out advanced fibrosis, but their positive predictive value for pinning down an exact fibrosis stage (F2 versus F3, specifically) is considerably weaker.

Trial protocols need something different. They need to enrich for a specific histological stage because the primary endpoint, usually fibrosis improvement by at least one stage without worsening of MASH, or MASH resolution without fibrosis worsening, depends on knowing where the patient started. A patient enrolled as F2 who was actually F1 on biopsy not only fails to contribute a valid efficacy signal; they contaminate the intent-to-treat population. So trial designers tighten the thresholds. They demand higher VCTE cutoffs, additional confirmatory biomarkers, and sometimes biopsy regardless of NIT results. Each of these requirements is individually defensible. Together, they build a wall between the trial and the clinic.

This is the central principle: the diagnostic precision a trial requires from a non-invasive test exceeds the diagnostic precision for which that test was validated in the population a trial wants to enroll. Every time a protocol tightens a NIT cutoff to reduce histological misclassification, it also excludes a portion of patients whose disease burden is real, clinically meaningful, and biopsy-confirmed if you bothered to look. The screen failure rate is not a recruitment execution problem. It is a design specification problem.

The regulatory machinery has begun to acknowledge this. On August 27, 2025, the FDA’s Center for Drug Evaluation and Research formally accepted a proposal to qualify liver stiffness measurement by VCTE as a reasonably likely surrogate endpoint for MASH trials, covering both accelerated approval and all-cause mortality or liver-related outcomes. Separately, the FDA accepted Echosens’ Letter of Intent to qualify FibroScan-based liver stiffness measurement as a surrogate endpoint that could eventually replace biopsy as a primary endpoint. Both actions signal that the agency wants non-invasive tests to do more work inside the trial, not just at the screening gate. But qualification as a surrogate endpoint is a separate question from whether enrollment thresholds map onto clinical reality. You can accept VCTE as your efficacy yardstick while still setting inclusion criteria that bear no resemblance to how hepatologists actually use FibroScan in practice.

When the Instrument Becomes the Obstacle

Consider what actually happens when a hepatologist orders a FibroScan for a patient with suspected MASH. She is using it to decide whether to escalate management, initiate pharmacotherapy, or refer to a specialist. Her working threshold for significant fibrosis (F2 or above) reflects the clinical practice guidelines she trained on and the patient population she sees: obese, often diabetic, frequently presenting with elevated liver enzymes without symptoms. The liver stiffness measurement she considers actionable in that context might be 8 to 10 kPa, depending on the inflammatory activity she suspects is confounding the reading. That same reading, submitted as part of a trial screening visit, might fall below the protocol-specified cutoff of 12 kPa or higher that the sponsor adopted to maximize histological confirmation rates. The patient goes home. The trial site counts another screen failure.

The mismatch is not hypothetical. The Lancet comment on non-invasive tests in MASH makes precisely this point: NIT strategies used in clinical drug trials diverge from those used in clinical decision-making in ways that have real consequences for which patients get enrolled and what the trial population actually represents. If trial participants must clear higher NIT hurdles than the patients who will eventually receive the approved drug, then the efficacy and safety signal generated in the trial belongs to a subset of the eventual treatment population, not the full population a label will cover.

That gap should concern every outcomes researcher working in hepatology. When resmetirom (Rezdiffra) was approved for MASH in March 2024, the approval was built on a trial population defined by biopsy-confirmed F2 and F3 fibrosis with specific histological activity criteria. The patients who will now receive that drug in clinical practice will be identified using NITs, not biopsies, by gastroenterologists who follow clinical practice guidelines, not Phase 3 inclusion criteria. The two populations are not identical. How large the gap is remains, at this moment, an open and uncomfortable question.

Adaptive Design as a Partial Antidote

The SYNERGY-NASH amendment is instructive precisely because it demonstrates both the problem and a possible corrective in a single trial. Tirzepatide’s developers did not wait for screen failure data to become catastrophic before acting. They incorporated FAST score greater than 0.35 as an additional inclusion pathway, which acknowledges that liver stiffness alone, without an inflammatory signal from AST, produces a noisier picture of who is truly F2 or above. That is an adaptive move in the operational sense, even if the protocol amendment itself was fixed once implemented. The amendment was designed to improve enrollment without compromising the histological enrichment the trial needed.

But adaptive enrollment criteria carry a burden that sponsors often underestimate. Every protocol amendment touching inclusion criteria triggers site retraining, updated informed consent, and in many cases an IRB review cycle that can consume substantial time at each site. For a large global trial with dozens or more sites, the cumulative delay is not trivial. And the FDA reviewer who eventually evaluates the trial has to satisfy herself that the population enrolled before the amendment is comparable to the population enrolled after it, a sensitivity analysis that rarely produces a clean answer when the defining characteristics of “comparable” are the thresholds you just changed.

The more durable solution sits upstream, in trial design, before a single patient is screened. Sponsors need to run NIT validation sub-studies in the target geographic populations before locking enrollment thresholds. A VCTE cutoff calibrated on a European cohort may perform differently in a Southeast Asian population where metabolic MASH phenotypes carry different anthropometric profiles, and liver stiffness measurements are sensitive to body mass index in ways that can shift the entire distribution of results. Building a staged screening algorithm, where a first-pass NIT screen at clinical-practice-level thresholds qualifies patients for a confirmatory second-pass NIT panel before any biopsy decision, would reduce both screen failures and unnecessary biopsies simultaneously. The FDA’s growing openness to VCTE and other NITs as surrogate endpoints creates the regulatory runway for exactly this kind of design innovation, but only if sponsors stop treating NIT thresholds as a plug-and-play specification and start treating them as a design variable that requires the same rigor as the primary endpoint itself.

The first MASH trial that closes the gap between its enrollment NIT strategy and clinical practice NIT standards will not just recruit faster. It will generate a trial population that actually predicts how the drug performs in the real world, and that is the evidence base a payer, a guideline committee, and an FDA reviewer can trust past the approval date.

References

1. The Lancet, “Non-invasive tests in MASH: clinical practice versus trials”
2. PubMed / SYNERGY-NASH, “SYNERGY-NASH Phase 2b trial: adaptive inclusion criteria incorporating FAST score for tirzepatide in MASH”
3. FDA CDER, “FDA Accepts Proposal for Reasonably Likely Surrogate Endpoint in MASH: All-Cause Mortality or Liver-Related Outcomes” (August 27, 2025)
4. HCPLive, “FDA Accepts FibroScan Letter of Intent for Surrogate Endpoint in MASH Trials”
5. Journal of Hepatology / ResearchGate, “Using the ELF test, FIB-4 and NAFLD fibrosis score to screen the population for liver disease” (3,378-participant cohort study)