Disagree. The AI "slop" likely was poor human oversight/prompting. Used properly AI is incredible helpful research tool.

This one bothers me. When SAFFRON opened in Aug 2022 there was real equipoise, savo + osi only had single arm data and chemo was the post-osi standard. That changed in June 2025 when SACHI, a randomized phase 3 of the same combo, showed a big PFS benefit over chemo and China approved it. I can see the case for SAFFRON continuing to randomize, SACHI was China only and mostly after 1st or 2nd gen TKIs. I can't see the case for still denying chemo pts the combo once they progressed. Every one of them got in because a biopsy showed MET-driven resistance, and outside China savo was essentially unavailable. That includes pts randomized before June 2025 who progressed after it. SACHI, same combo and same PI, let chemo pts cross over and 43% did. KRYSTAL-12 allowed crossover to adagrasib, and GEOMETRY-E, capmatinib + osi in this same setting, added it by amendment. No SAFFRON amendment adding crossover has been reported, and that likely made the OS gap bigger than it would have been with MET-directed therapy available after chemo.

Pan tumor off one positive ph3 outside lung, from one country, at an interim? GI1 is a real result, first ph3 to beat an active PD-L1 arm on OS in BTC. But there was no PD-L1 plus bev arm, so we don't know yet what the bispecific itself adds. EA2251 should tell us.

Nice work by Dr. Powles and the VOLGA team. Durva/EV improved OS with EV given only before cystectomy, which makes sense for pts who struggle after surgery. I'd like to see EFS events by type in each arm, since 21% of controls never had surgery vs 15% on durva/EV.

DeLLphi-305 is a real OS win, HR 0.42, and since many SCLC pts never reach 2L, giving tarla early makes sense. But only 33 of 281 control pts got tarla later, so the trial can’t tell us how much of the gain is early tarla vs any tarla at all.

LITESPARK-012 is consistent with KEYMAKER-U03, where qmab/pembro/len had numerically lower ORR (71.3% vs 80.6%) and median PFS (18.0 vs 26.3 mo) than pembro/len. It adds anti-CTLA-4 to a PD-1/VEGF TKI doublet, a different question from ipi with nivo, which 8Y8 tests.

I agree these are impressive results. One apparent correction, though, since these numbers are being shared widely. The HR 0.66 you quoted is from the first-line patients only, nivo/ipi vs nivo alone. That comparison wasn't statistically significant: p was 0.043, and the trial needed p below 0.006. The OS result that was significant vs nivo alone is the one on the slide you posted. It includes patients treated in any line (about 40% had prior therapy for metastatic disease): HR 0.61, five-year OS 74% vs 62%. In first-line patients, the trial clearly shows nivo/ipi beats chemo on OS (HR 0.37, even though most chemo patients later got immunotherapy). It doesn't prove that adding ipi beats nivo alone; first-line PFS vs nivo also fell short of significance. That matters because ipi adds toxicity. Across all lines, grade 3-4 treatment-related side effects were 22% vs 14%, and adrenal insufficiency (any grade) was about 10% vs 3-5%, usually permanent. Agree chemo alone has no first-line role outside a true contraindication to IO, with one caveat: confirm the MSI result. In the first-line part of 8HW, 16% of patients called MSI-H locally weren't confirmed centrally, and the ones who turned out to be MSS/pMMR had a median PFS under 2 months on nivo/ipi (small numbers). Agree duration of therapy and oligoprogression are the next questions that are often neglected in ICI trials.

One caveat on ‘SoC’: the control arm got no pembro, which most PD-L1 CPS ≥1 US patients would get. In the PD-L1+ (TAP ≥1) subgroup, the triplet’s OS gain wasn’t significant (HR 0.82; exploratory). No head-to-head data vs pembro + trastuzumab + chemo.

Sadly I’m not surprised. Journals quietly signing off on inflated framing has been all too common for years. Calling a descriptive OS a “positive OS result” is a textbook case, and I agree it should have been caught.

Thanks for sharing. Agree. The OS p is nominal, and it's open-label with 7.9% of controls withdrawing consent at randomization vs 3.3% on ribo. A 2.1-point gap when most relapsed controls didn't get a 1L CDK4/6i isn't a reason to change US practice imo.

Agree access at recurrence is a problem. But if US patients get the CDK4/6i at relapse and OS barely moves, then persistent iDFS here mostly means delaying the same outcome, not changing it.

Sobering, agreed. It missed its own pCR bar, though with no chemo comparator and pCR a weak surrogate in HR+ disease, that's all it can tell us. Three cases of grade 2 ILD and a death are a lot to see in patients treated with curative intent. The 2022 interim data already suggested anastrozole added nothing. In HER2-low disease, T-DXd's proven benefit is still metastatic.

Agree, and that monopoly is already eroding. Editors still decide what gets printed. They no longer get the last word on what it means.

Two percentage points is small. We accepted it for taxanes, sure, but that doesn't make it the right bar. NATALEE's OS p-value is nominal, and we don't know how many controls got a CDK4/6i at relapse (in monarchE only 47% got one in 1L), so some of this could be CDK4/6i vs none rather than early vs late. Also open-label, and 7.9% of controls withdrew consent at randomization vs 3.3% on ribo.

Not ready to call NATALEE an OS win. The p=.019 is one sided and nominal, on the 5th look at survival. More than twice as many control pts dropped out right after randomization, and post-relapse CDK4/6i use in the control arm hasn't been reported .

HR 0.58 is right where the phase 2b was, and they used a real pembro control, which I give them credit for. Unfortunately this is RFS on top of a pembro backbone that was itself approved on RFS. COMBI-AD showed big adjuvant RFS and DMFS gains with no significant OS difference. Patients who recur now have good salvage options, so treating earlier isn't necessarily curing more people. Fever in 52% vs 8% means nobody was really blinded. I'd want OS before calling this a new standard.

Usage isn't survival, and popularity isn't clinical benefit. A pharma-funded tool that's wrong often enough to matter sounds more like a bioethics case study than a success story.

Travesty

Most patients are treated in community practices, not academic centers. Many of us would enroll if trials came with funding and research staff. The accrual problem is structural. Asking critics to hold off until they run an RCT doesn't fix it.

Well said. There’s no quality data that this cocktail is standard of care, and presenting it as settled is misleading. @ShrutiGkidney’s work tried to move us forward in a data vacuum, and that’s exactly what we should be encouraging.