Two Approved Companion Diagnostics Validate Pembrolizumab Eligibility Across dMMR/MSI-H Tumors

One of the most striking ideas in modern oncology is that a cancer drug can be approved not for where a tumor grows but for what it is made of. Pembrolizumab, an antibody that blocks the programmed cell death protein 1 (PD-1) checkpoint and thereby releases the brakes on T cells, holds a tumor-agnostic approval in the United States for patients with unresectable or metastatic solid tumors that are mismatch repair deficient (dMMR) or microsatellite instability high (MSI-H) and that have progressed after prior therapy with no satisfactory alternatives. That approval, first granted on the strength of clinical trial data, rested on assays used inside those trials. Now a team of researchers at Merck & Co. has reported the results of post-marketing bridging studies designed to confirm that two commercially available companion diagnostics can reliably identify the patients who benefit, and the findings, published in BMC Cancer, offer an unusually detailed look at how the diagnostic backbone of tumor-agnostic medicine is validated.

Companion diagnostics, or CDx, are tests that pair with a specific drug to determine which patients are candidates for treatment. In the case of pembrolizumab and dMMR/MSI-H tumors, the biological logic is elegant. The mismatch repair system, built from proteins that proofread DNA during replication, corrects small errors that accumulate as cells divide. When that system fails, stretches of repetitive DNA known as microsatellites grow or shrink unchecked, producing microsatellite instability. The resulting flood of mutations generates abnormal neoantigens that the immune system can recognize, which is precisely why tumors with defective mismatch repair respond so well to checkpoint blockade. Detecting this state, however, can be done in more than one way: immunohistochemistry (IHC) can reveal whether the four mismatch repair proteins are present in tumor cells, and polymerase chain reaction (PCR) or next-generation sequencing (NGS) can directly interrogate microsatellite length.

The original clinical trials that supported pembrolizumab’s tumor-agnostic approval used a clinical trial assay, or CTA, based on IHC or PCR to determine MMR/MSI status. Once a drug is on the market, regulators typically require that the diagnostic tests actually used in routine practice be clinically validated against the evidence base, a process known as bridging. In the new analysis, the researchers evaluated two approved companion diagnostics: the Ventana MMR RxDx Panel, an IHC-based test, and FoundationOne CDx, a next-generation sequencing assay. The goal was twofold. First, they asked whether patients whose tumors were dMMR/MSI-H by the companion diagnostics showed the same enrichment of response to pembrolizumab that had been seen with the CTA. Second, they measured how well the tests agreed with one another and with the CTA at the level of individual tumor samples.

The efficacy analysis drew on two pivotal studies. KEYNOTE-158 enrolled patients with many different advanced, previously treated solid tumors, and its cohort K comprised 321 participants with noncolorectal cancers that were dMMR/MSI-H by the CTA. KEYNOTE-164 enrolled 123 participants with previously treated metastatic colorectal cancer, a tumor type in which dMMR/MSI-H status is less common than in some other malignancies but where checkpoint blockade has proven especially transformative. Together, 444 participants with dMMR/MSI-H tumors per the CTA entered the efficacy analysis. The researchers estimated the objective response rate, the proportion of patients whose tumors shrank measurably, using statistical methods appropriate for bridging a CTA to a commercial CDx, accounting for the fact that the companion diagnostics classify some samples differently than the trial assay does.

The results showed clear enrichment of response among patients confirmed positive by the companion diagnostics. Among participants whose tumors were dMMR/MSI-H by the Ventana MMR RxDx Panel, the objective response rate was 34.7 percent, and among those positive by FoundationOne CDx it was 43.0 percent. Across all 444 participants with dMMR/MSI-H tumors per the CTA, regardless of what either companion diagnostic showed, the objective response rate was 31.8 percent. Those numbers tell a coherent story: the companion diagnostics do not dilute the signal that justified the approval, and in the case of the sequencing-based test, they identify a subgroup with an even higher response rate than the overall CTA-positive population. For clinicians, that means the tests used in routine practice can be expected to select patients who resemble those who responded in the trials.

Agreement analyses formed the second pillar of the study, and here the sample sizes were much larger because tumor samples could be drawn from KEYNOTE-158 cohorts A through K, KEYNOTE-164, and KEYNOTE-177, the latter a trial of pembrolizumab versus chemotherapy as first-line treatment for MSI-H colorectal cancer, along with commercially procured tumor samples evaluated by both methods. The researchers computed three standard metrics: overall percent agreement (OPA), positive percent agreement (PPA), and negative percent agreement (NPA). When the CTA was compared with the Ventana MMR RxDx Panel across 934 samples, OPA was 90.9 percent, PPA was 72.0 percent, and NPA was 98.5 percent. Against FoundationOne CDx across 1174 samples, OPA was 94.5 percent, PPA was 69.8 percent, and NPA was 99.3 percent.

Those asymmetrical agreement figures deserve attention because they reveal something fundamental about how different assay platforms see the same biology. Both companion diagnostics agreed strongly with the CTA when the CTA called a sample proficient, with negative percent agreement approaching or exceeding 98 percent, meaning that tumors the trial assay deemed mismatch repair proficient were almost never called deficient by the commercial tests. Positive percent agreement, by contrast, hovered around 70 percent, indicating that a meaningful fraction of samples positive by the CTA were not flagged by the companion diagnostics. This pattern is common when comparing IHC, PCR, and NGS approaches, because each platform interrogates a different proxy of the same underlying defect: protein expression, microsatellite length variation, or sequence-level changes. Discordant cases often sit near the biological boundary between proficient and deficient repair rather than reflecting outright error by any test.

Reassuringly, when the two companion diagnostics were compared head to head across 662 samples, they agreed with each other at a high rate: overall percent agreement of 95.9 percent, positive percent agreement of 79.8 percent, and negative percent agreement of 98.3 percent. The two commercially available tests, one reading protein expression under the microscope and the other reading DNA sequence on a sequencer, converge on the same answer for nearly all tumors. That convergence matters for the practical promise of tumor-agnostic therapy, because it suggests that a patient’s eligibility for pembrolizumab should not hinge dramatically on which approved test a hospital happens to run. The authors concluded that both the Ventana MMR RxDx Panel and FoundationOne CDx demonstrated high agreement with each other and, compared with the CTA, were able to identify participants with dMMR/MSI-H tumors likely to respond to pembrolizumab.

The significance of this work extends beyond a single drug. Tumor-agnostic approvals represent a regulatory philosophy in which shared biomarkers, rather than anatomical origin, define the treatable population, and the entire framework depends on diagnostics that can be trusted in every oncology clinic, not only in the trials that generated the evidence. Bridging studies like this one are the mechanism by which that trust is established and maintained as part of post-marketing commitments to regulators. With pembrolizumab now an option for patients with dMMR/MSI-H tumors of virtually any origin, from endometrial and gastric cancers to biliary tract and small bowel tumors, the reliability of the tests that flag those tumors becomes a matter of life and death for thousands of patients who might otherwise never be considered candidates for immunotherapy.

There are also honest limitations worth noting. The research was funded by Merck Sharp & Dohme LLC, a subsidiary of Merck & Co., and all authors are company employees who may hold stock or stock options, a disclosure that readers should weigh when interpreting the uniformly favorable conclusions. Positive percent agreement of roughly 70 percent between the CTA and each companion diagnostic leaves room for tumors that one platform would treat as eligible and another would not, and the study did not attempt to resolve which platform is definitively superior in discordant cases. Still, the core message stands on solid ground: two widely available companion diagnostics, validated across more than a thousand tumor samples and hundreds of treated patients, can reliably find the patients whose broken DNA repair machinery has loaded their tumors with the mutations that make pembrolizumab work. In the evolving story of precision oncology, this study is a reminder that the diagnostic test is not an accessory to the drug. It is, in a very real sense, half of the therapy.

Subject of Research: Clinical validation of companion diagnostics for pembrolizumab in dMMR/MSI-H tumors

Article Title: Companion diagnostics for pembrolizumab in patients with dMMR/MSI-H tumors

Article References: Lang, L., Cesario, J., Levitan, D., Mathur, S. K., Aurora-Garg, D., Card, D., Lunceford, J., Jin, F., & Wehn, A. K. (2026). Companion diagnostics for pembrolizumab in patients with dMMR/MSI-H tumors. BMC Cancer. https://doi.org/10.1186/s12885-026-16994-0

Image Credits: AI Generated

DOI: 10.1186/s12885-026-16994-0

Keywords: pembrolizumab, companion diagnostics, dMMR, MSI-H, immunotherapy, biomarker, Ventana MMR RxDx Panel, FoundationOne CDx, tumor-agnostic approval, checkpoint inhibitor, mismatch repair deficiency, KEYNOTE-158