The POLAR Trial Proves Pancreatic Cancer Trials Were Broken by Design

Picture a trial statistician sitting across from a principal investigator in 2019, reviewing enrollment projections for a gemcitabine-based pancreatic cancer study. The PI estimates 400 patients. The statistician needs 400 patients. Nobody in that room is asking which 400. That was the problem — and the POLAR trial, published in Nature Medicine, is the most direct argument yet that “which patients” was always the question that mattered most.

The POLAR trial, sponsored by Merck and AstraZeneca, evaluated pembrolizumab combined with olaparib in patients with homologous-recombination-deficient (HRD) metastatic pancreatic cancer — a molecularly selected Phase 2 population representing one of oncology’s most hostile indications. The trial did not attempt to rescue an immunotherapy that failed in unselected pancreatic cancer. It started, by design, with only the patients whose tumor biology gave the combination a mechanistic rationale. What POLAR demonstrates — and what the field needs to hear plainly — is that decades of failed pancreatic cancer trials were not evidence that the drugs didn’t work. They were evidence that the trials enrolled the wrong patients.

That reframing is not semantic. It has direct consequences for how sponsors write protocols, how IRBs evaluate feasibility, and how the FDA should be thinking about its guidance on biomarker-driven oncology trial design.

The Stratification Imperative

Pancreatic ductal adenocarcinoma has an approximately 5-year survival rate of 12%, making it one of the deadliest solid tumors in the U.S. For two decades, the standard of care moved glacially — gemcitabine in 1997, FOLFIRINOX in 2011, nab-paclitaxel combinations shortly after. Each of those advances came from unselected populations, which was scientifically defensible given what we knew at the time. But immunotherapy and PARP inhibition are not cytotoxic drugs. They depend on the molecular context of the tumor in ways that gemcitabine does not. Enrolling biomarker-unselected patients into pembrolizumab or olaparib trials was never a neutral design choice — it was a near-guarantee of diluted signal.

The FDA’s guidance on biomarker-based clinical trial enrichment strategies has been on the books since 2019. That document explicitly describes how prospective stratification by predictive biomarkers can “increase the probability of observing a clinically meaningful treatment effect if the biomarker correctly identifies the sensitive subpopulation.” The POLAR trial operationalized exactly that logic. And yet, for years, pancreatic cancer drug development proceeded with enrichment strategies borrowed from lung and breast oncology — or none at all — because the field lacked validated HRD assays applied prospectively at scale in this indication.

Here is the operational tension that POLAR exposes: prospective biomarker stratification is only as good as the testing infrastructure around it. HRD status in pancreatic cancer is not a reflexive pathology order. Determining which patients carry germline or somatic HRD alterations — including BRCA1, BRCA2, PALB2, and ATM variants — requires coordinated tissue collection, genomic profiling turnaround times compatible with trial enrollment windows, and site-level education that most community oncology practices simply do not have in place. That is a trial operations problem, not a biology problem. And it is exactly the kind of problem that sponsors routinely underestimate in their protocol feasibility assessments.

Consider a scenario where a sponsor designs an HRD-enriched trial and projects 18-month enrollment based on the population frequency of HRD alterations in published pancreatic cancer genomics databases — roughly 15 to 25% of patients depending on the cohort and the breadth of the HRD definition used. If sites are not pre-qualified for rapid genomic profiling and tissue adequacy is not built into the screening visit as a hard gate, that 18-month window becomes 30 months. The trial becomes underpowered not because patients don’t exist, but because the operational architecture failed to find them in time. POLAR’s publication in Nature Medicine gives future sponsors a proof-of-concept anchor, but the protocol design lessons embedded in its methods section deserve as much attention as the efficacy data.

What the FDA’s Enrichment Framework Actually Demands

The FDA’s biomarker enrichment guidance is frequently cited and infrequently interrogated. Open that 2019 document and you will find a framework that is conceptually sound but operationally thin. It describes what enrichment can accomplish. It does not prescribe minimum standards for assay validation, turnaround time thresholds, or the statistical handling of patients who screen positive for HRD but whose tissue fails quality control before randomization. Those gaps are not academic — they directly affect the integrity of the intent-to-treat population and the reproducibility of any efficacy signal.

The POLAR trial is a Phase 2 study, which means its primary role is signal detection, not registration-level confirmation. That distinction matters enormously for how sponsors read the results. A Phase 2 win in an HRD-enriched pancreatic cancer population is not a green light to file an NDA. It is a mandate to design a Phase 3 with a pre-specified HRD assay, a locked analytical cut-point, and a pre-agreed analytical validation plan reviewed in a Type B meeting before first patient in. The conversation with FDA about companion diagnostic co-development strategy is not one you want to have after your pivotal trial is enrolled.

The EMA’s guidance on co-development of medicines and companion diagnostics, last updated in 2020, is more prescriptive than the FDA’s equivalent on the question of assay lock. European regulators expect sponsors to specify the exact CDx platform in the protocol and to have analytical validation data in hand before Phase 3 initiation. The FDA’s parallel framework allows more flexibility — but in HRD testing, where different platforms produce meaningfully different HRD scores and classification thresholds, that flexibility creates a reproducibility problem that will surface at advisory committee review. The POLAR results should prompt the FDA’s Oncology Center of Excellence to issue updated thinking on HRD assay standardization in pancreatic cancer specifically, because the field is about to generate a wave of follow-on trials built on this data, and they will not all use the same testing platform.

Building the Trial That POLAR Demands Next

If you are a sponsor reading POLAR as a signal to advance into Phase 3, the first protocol decision is not about the drug combination — it is about the screening strategy. Retrospective HRD testing of archived tissue is not the same as prospective HRD testing at diagnosis. Tumor heterogeneity in pancreatic cancer, combined with the frequent inadequacy of biopsy material from pancreatic primaries, means that a significant proportion of patients who are clinically HRD-positive will not have evaluable tissue from their most recent procedure. Sponsors who do not build a re-biopsy pathway into their protocol will face a screening failure rate that makes enrollment projections fictional.

The decentralized clinical trial infrastructure that the FDA’s guidance on DCTs has been promoting since 2023 offers a partial solution here. Mobile phlebotomy and local laboratory networks can support liquid biopsy-based HRD screening in patients who cannot travel to academic centers for repeat tissue procedures. But liquid biopsy HRD concordance with tissue-based HRD in pancreatic cancer is not yet established at the sensitivity and specificity levels required for registration-level patient selection. Any sponsor who substitutes liquid biopsy for tissue HRD in a Phase 3 enrollment strategy without a pre-specified concordance analysis built into their statistical analysis plan is creating a regulatory vulnerability that a complete response letter will eventually find.

None of this diminishes what POLAR accomplished. A Phase 2 trial that demonstrates meaningful clinical activity in metastatic pancreatic cancer — an indication where the median overall survival in unselected populations hovers near 11 months with standard chemotherapy — is a genuine advance. The HRD-enriched design is what made that signal visible. The question is whether the sponsor community and the FDA will treat POLAR as a template for how to build biomarker-driven Phase 3 trials in difficult solid tumors, or whether they will treat it as an anomaly and return to the design habits that produced 20 years of failed pancreatic cancer immunotherapy studies.

Precision oncology earned its credibility in lung cancer and melanoma by insisting on molecular stratification before enrollment. Pancreatic cancer is overdue for the same discipline. POLAR showed that the signal is there when you look in the right population. The next failure in this indication won’t be the drug. It will be the protocol that didn’t bother to ask which patients.