Metastatic Cervical Cancer: Pembrolizumab Below the PD-L1 Cutoff
KEYNOTE-826 tested pembrolizumab in essentially everyone with recurrent or metastatic cervical cancer, but the FDA drew its approval line at a PD-L1 CPS of 1. Her biopsy came back just under that line, in the exact place a trial's own intent-to-treat population and its regulatory label stop agreeing with each other.
Renata S. is thirty-one, the older of two sisters, and her younger sister moved into her spare room the week of her diagnosis without being asked twice — someone had to help manage a two-year-old while Renata's chemotherapy started, and neither of them considered any other arrangement. She presented with vaginal bleeding six months after a normal Pap smear, and biopsy confirmed squamous cell cervical cancer with imaging showing para-aortic nodal and a single pulmonary metastasis: recurrent/metastatic disease at first presentation, not a local recurrence after prior treatment.
KEYNOTE-826 changed first-line treatment for this exact clinical picture by adding pembrolizumab to platinum-based chemotherapy with or without bevacizumab, and the trial enrolled essentially all comers — it did not require a minimum PD-L1 level to participate, testing its hierarchical primary endpoints first in the PD-L1-positive population, then across the full intent-to-treat group, and both showed a real, significant overall survival benefit. But approval followed the trial's own hierarchical testing structure rather than its broadest population: the FDA label restricts pembrolizumab to tumors with a PD-L1 combined positive score of 1 or greater, and Renata's biopsy came back at a CPS of 0 — PD-L1 negative by the specific assay and threshold the trial used. That number sits her outside the approved indication even though she was, technically, exactly the kind of patient KEYNOTE-826's intent-to-treat analysis included and whose aggregate result showed benefit; the honest complication is that the CPS-negative subgroup within that intent-to-treat population was 35 patients against 34, 11.2% of the trial, and its own overall survival hazard ratio came to 0.87 with a confidence interval running from 0.50 to 1.52. The trial's own authors wrote that no clear inference about that subgroup could be drawn and that the effect, if any, appeared small — which is a weaker thing to build an appeal on than "she was in the trial."
Renata has asked, more than once, whether a strongly worded insurance appeal might get her the drug faster than a trial-screening process that could take weeks she feels she doesn't have. The team's honest answer is that an appeal built on an underpowered subgroup analysis is a weaker case than either of them would like it to be — which is itself a reason to pursue the trial pathway in parallel rather than instead of it.
A CPS of zero, one point below the label
KEYNOTE-826 didn't require a PD-L1 floor to enroll, and its intent-to-treat population — which included patients with a CPS like hers — still showed a real overall survival benefit. The FDA drew the label at CPS 1 because that's where the hierarchical testing plan reached statistical confidence, not necessarily where the biology actually stops working. I don't think a regulatory line should be read as a biological one automatically.
That threshold isn't arbitrary, though — it's where the trial's own statisticians determined a subgroup could be characterized with confidence. The CPS-negative slice of the intent-to-treat population was small and not independently powered. Being included in a trial that was positive overall is not the same claim as having a demonstrated benefit in her specific subgroup.
"The biology probably doesn't have a hard cutoff at exactly CPS 1" may well be true, but that's a plausibility argument, not the trial's own finding — and we don't get to substitute plausibility for the specific evidentiary standard the label was built on.
I'd frame this differently than either of you. The real choice isn't pembrolizumab versus nothing — it's whether she gets it through an off-label argument built on an underpowered subgroup, or through a clinical trial specifically designed to test benefit at exactly her CPS range. I'd start chemotherapy and bevacizumab per the approved standard now, and refer her for trial screening in parallel rather than resolve this by arguing past the label.
Agreed: carboplatin-paclitaxel plus bevacizumab started this week per the approved standard for PD-L1-negative disease, with a clinical trial referral submitted the same day.
Not agreed, and the reason this case is left genuinely unresolved rather than smoothed into consensus: whether the FDA's CPS ≥1 threshold reflects a real biological boundary or an artifact of where KEYNOTE-826's statistical plan happened to reach confidence. The oncologist and the pharmacologist never reconciled that question — the trial-referral path let the team act without resolving it, not because it was resolved.