Headline: AI: 83%. Doctors: 36%!
Published in Science. Reported everywhere as "AI is twice as good as doctors."
This AI model is sooo “good” by only reading opening 2 to 3 sentences of each case it already outperformed doctors! 😁🤦🏼♂️ its like it new the answers! 🤭
Is it true? I read the paper. 🧵
There aren't many diseases where you can say "11th line" with a straight face. In myeloma, some trials enroll pts with a median of 5 to 7 prior lines.
https://t.co/CWxja9zrLg
One mystery I will never quite understand in myeloma therapy: A patient resistant to almost every anti-myeloma drug who responds slowly but surely to weekly cyclophosphamide. I present to you 11th line CyBorD...
The problem in REGAL is the control arm. "BAT" ranged from observation to HMA/ven.
If many controls were just watched, 24 vs 14.7 mo mostly shows treatment beats no treatment. Against pts who actually got ven, that OS gap could shrink fast. https://t.co/ynMKgQQDCG
It looks like I was wrong about Sellas Life Sciences $SLS and its experimental blood cancer immunotherapy.
On Tuesday, Sellas said the therapy, called galinpepimut-S, reduced the risk of death by 41% compared to standard treatment for patients with an advanced stage of acute myeloid leukemia – an outcome that achieved the main goal of a Phase 3 clinical trial, the company said.
Story here:
https://t.co/oCTAcdnqru
5/5 Nutrition cannot reverse advanced cancer or guarantee survival benefits. But strength and comfort matter. Treating the tumor and caring for the person are inseparable. Su et al., JAMA Oncology (2026): https://t.co/AZViKAI7xV
Hospital air mattresses are a solution in search of data. The best trial couldn't show fewer ulcers than good foam.
What they do reliably: keep patients up at night and make it harder to get out of bed. Not the outcome we were going for.
FDA is right to insist on OS. We’ve already seen this with patritumab deruxtecan, sotorasib, and mobocertinib in lung cancer: early PFS or response gains that never translated to survival, leading to withdrawal or major scrutiny.
Ivonescimab’s PFS (HR 0.52) with a non-significant OS (HR 0.79) and initially flat Western results looks like the same thing.
JUST IN: the @FDA has declined to approve PD1/VEGF bispecific ivonescimab in advanced NSCLC with EGFRm, because of lack of OS benefit (via @Reuters) — Having seen approval in several cancers based on PFS, this is somewhat a puzzling rejection.
https://t.co/aLPJQm72Iw
I am so confused! 🤷🏼♂️🤨 please help me!
June 2025: FDA gives Dato-DXd accelerated approval in EGFR mutated mNSCLC after TKI. Decision was made on a pooled 45% ORR from single-arm phase 2 and a subgroup of a phase 3. Yes decision was made based on polled ORR from two different trials enrolling patients under DIFFERENT inclusion/exclusion criteria! 🤯
Nov 2026: randomized phase 3 trial meeting its primary end point - PFS. FDA says PFS without significant OS is not enough. And I agree!
But pooled single-arm response rates were enough for Dato-DXd.
Same disease. Double standard.🙅🏼♂️
FDA gave accelerated approval to Dato-DXd with a confirmatory trial pending. And that trial is TROPION-LUNG15 with primary end point being PFS!!!😵 so if no OS benefit FDA should withdraw approval right? In fact, let’s go back and remove all drugs that don't have OS benefit!
For Datao-dxd the label covers patients after an EGFR TKI and platinum chemo. Lung15 compares Dato-DXd against platinum doublet in patients with EGFR NSCLC and disease progression on prior osimertinib. So the trial would confirm benefit one line earlier than the approval is! 🤦🏼♂️
PS IMO neither of the drugs should be approved 😁
First and foremost, I have huge respect for you and all your work! 😅
However, I think calling modern AI a “parrot” significantly understates what these models actually do. Perhaps statements like this also highlight how much more education we need about what modern AI actually is and maybe we need more webinars and @ASCO sessions for healthcare professionals on how these systems actually work.
Yes, AI is trained on existing human knowledge, but so are we. Training on existing knowledge does not mean simply retrieving or repeating it (unlike USMLE or our board exams). Modern LLMs can reason across an ocean of information in seconds, generate novel hypotheses, identify relationships, write new code, and propose solutions that were NEVER present in their training data.
AI enabled a breakthrough in the decades-old protein folding problem, resulting in 2024 Nobel Prize. That's pretty impressive parrot. 😅🦜
if reviews provided by AI today are too long or gibberish, we need to standardize them, make sure we use high standard AI engines, make sure reviewers are aware that they can violate privacy if they use AI, we need to come up with new guidelines for AI generated peer reviews etc.
@oncology_bg I rec'd you reconsider with oversight. AI does a much better job than any human can.
Presentations that are AI assisted I find when done well are exponentially easier to follow and learn from.
SAFFRON is positive, but I'm cautious on the OS. Chemo pts couldn't cross over, we don't know what they got after, and 13% of them never started chemo vs <1% on the combo. I'd also really like to see it against ami-chemo.
https://t.co/BaeszNfecR
🫁 Phase 3️⃣ SAFFRON findings show osimertinib + savolitinib significantly improved progression-free and overall survival vs chemotherapy in EGFR-mutated #NSCLC with MET alterations after progression on osimertinib.
▪️Shun Lu, MD | #ESMO26 | @NEJM
🔗 https://t.co/nu7C8ovtZV
We've been here with HER3 before. HER3-DXd won on PFS after osi too, and OS was no different. This PFS difference is much bigger, 8.3 vs 5.4 mo, so I'm more hopeful. But with only 66 deaths, OS is unproven.
https://t.co/aVQ2bLU1vl
🚨 Another HER3 ADC delivers in EGFR-mutant lung cancer!
Published in @TheLancet, the phase III HERON-Lung01 trial evaluated ruzaltatug rezetecan vs platinum-pemetrexed after EGFR-TKI progression.
🔬 498 patients | Key results:
📊 Median PFS: 8.3 vs 5.4 months
🔥 HR 0.47, a 53% reduction in progression/death hazard
⚠️ Grade ≥3 TRAEs: 41% vs 39%
My take: Another positive phase III ADC trial in EGFR-mutant NSCLC. Promising efficacy, but immature follow-up, pending OS data and an all-China population warrant caution.
The post-TKI treatment landscape is getting crowded!
@oncoalert #LungCancer #NSCLC #ADC
Full paper link in comment below ⬇️
PERSEUS settled D-VRd up front for me. The question I still get 3 months after transplant is whether to keep dara going with len. Honestly, I don't know, and PERSEUS wasn't designed to tell me.
https://t.co/3ozUbiUH4B
#Myeloma Paper of the Day: Modeling of long-term PFS in NDMM treated w/ Dara-VRd on PERSEUS (TE pts) finds 158-255 months (VRd 76-119 mos) while for CEPHEUS (TIE pts) projected median PFS range across distributions was 96-118 (VRd 52-54 mos): https://t.co/ClunRLQKXm. #mmsm
>3 yrs of 1L PFS with T-DXd/pertuzumab is real progress. Before it's everyone's default I want OS and the cumulative tox picture, ILD especially. In the community, more options = more sequencing questions we don't have data for yet.
https://t.co/zNuYzeJpkx
There was a time when treating HER2+ breast cancer was relatively simple. That time is over. And that’s great news!
Not long ago, we recommended TCHP for most patients with stage II–III disease, and THP (CLEOPATRA) for those with metastatic disease. Straightforward. Almost one-size-fits-all.
Within just a few years, that simplicity is gone.
And that’s progress. More opportunities to spare unnecessary toxicity, tailor treatments to patients’ preferences, and move beyond rigid algorithms. To be more doctors.
In early-stage disease, options now range from de-escalated THP to intensified regimens such as DB11. In metastatic disease, T-DXd/P has achieved unprecedented PFS (>3 years in the 1L setting!), with novel maintenance strategies (palbociclib or tucatinib) potentially pushing the bar even higher. Meanwhile, biomarkers are finally coming of age, from imaging to tissue and plasma.
What an era to be treating HER2+ breast cancer!
But more options also mean more complexity. For oncologists, and especially for patients. More knowledge needed, more discussions, more decisions to make.
How do we navigate this increasingly complex landscape and choose the right treatment for the right patient?
We tried to offer some guidance in our new review recently published in @NatRevClinOncol.
Read it here: https://t.co/kHrPzxsRjM @ValenzaCarmine
This is the right question, and it comes up all the time in community practice.
A control-arm OS win only matters if that arm is what patients actually get. If the comparator is outdated or underused, it can look decisive and still leave the sequencing question unanswered.
https://t.co/pnpuyAx36a
Are the OS benefits reported in AQUILA, DREAMM-7 & STARGLO misleading?
OS is the gold standard. But beating a trial’s control arm does not necessarily mean beating today’s best care.
Before calling these results “practice-changing,” ask:
• Was the comparator adequate?
• Did patients have access to effective treatment after progression?
• Do the findings apply to our patients today?
Let’s critically review these trials—including the latest #ASH26 updates—and examine what the survival headlines actually prove.
“Improves survival” should start the discussion—not shut it down.
🔹 AQUILA: ASH 2026 strengthens the OS evidence: HR 0.52; 7-year OS 87.8% versus 78.7%. But earlier FDA data showed that only 31/105 monitoring-arm patients who started myeloma treatment received a daratumumab-containing first regimen. The new abstract does not update that treatment breakdown.
🔹 DREAMM-7: An OS benefit was demonstrated. In the October 2024 dataset, 34/130 control patients receiving subsequent therapy received drugs in the reported BCMA-targeting category; cellular therapies were listed separately. Access to effective salvage remains relevant to interpretation.
🔹 STARGLO: ASH 2026 confirms a sustained overall survival advantage: HR 0.63; four-year OS 45.5% versus 27.1%. Yet US enrollment was only 9%, and earlier exploratory regional estimates differed substantially. Updated regional and subsequent-treatment data would help assess applicability.
These observations do not establish how much subsequent care contributed to the survival differences. They identify questions that an OS headline alone cannot answer.
A real survival advantage within a trial does not automatically establish superiority over the best contemporary treatment sequence.
Read beyond the hazard ratio.
#ASH26 #MultipleMyeloma #Lymphoma #HemeTwitter
I’ve been looking at LinkedIn recently. It seems to consist mainly of self-congratulatory posts – people congratulating themselves on a new role or another achievement and other people congratulating them back. It’s platform for self-congratulation rather than critical debate.