Vincent Rajkumar

I would like to share the story of how a patient with cancer came up with the idea for a randomized trial, & how listening to him saved a lot of lives. 1/ In 2002, I had just completed a randomized trial with the notorious drug thalidomide for the cancer, multiple myeloma.

@essamelfaham @MedwatchKate Yes. We add it in mSMART but not included in IMS IMWG. I will consider extra medullary added. And the double hit we defined in mSMART is the ultra high risk. PCL and EMD are also best considered ultra high risk. static1.squarespace.com/static/5b44f08…

@VincentRK @MedwatchKate Dear Professor. It’s a great idea but I would like to add the ultra high risk which can be incorporated according to the latest imwg definition or we will just have our limit to the high risk in the calculator and to have extramedullary added as well in the tick options

Want to counsel patients accurately on risk stratification and outcomes in MGUS, SMM, WM, Myeloma, and Amyloidosis? One stop website for risk stratification. Input simple variables. And we provide the risk stratification and an estimate on outcomes MyelomaRisk.com eatoure.github.io/#/

@drdibakarpodder Hard one. But in general no.. it has to be one for the high risk partners. It’s possible that the unknown translocation is with a high risk partner chromosome. But unless we are sure it is better to consider as standard risk

@VincentRK Sir, does IgH with unknown translocation with 1q gain/amplification makes it high risk?

@KarunNeupaneMD @AlexisNarvaezMD @GeminiApp @Google @grok @deepseek_ai @claudeai If you can please try UpToDate AI

@VincentRK @AlexisNarvaezMD @GeminiApp @Google @grok @deepseek_ai @claudeai OpenEvidence seems to be the only reliable model for medical facts. Other models just cant answer factual questions reliably. At Moffitt, we showed that training a generic model with synthetic pt notes enabled it to simplify real clinic notes with zero hallucinations-pending pub

Maybe you guys need to switch to @GeminiApp @Google @grok @deepseek_ai @claudeai #Ai

@Rfonsi1 @marianojarribas Amazing.

@marianojarribas @VincentRK He quietly accepted K is the way to go 😎. Right?

1) Pt yesterday treated with Velcade for AL - neuropathy. 2) A colleague patient on PERSEUS regimen - neuropathy. New patient, I want to start on Dara-KRD - Insurance denial -> "Use bortezomib!" *** Until when @NCCN! We need K as one of the preferred options for payers to cover.

@kansagraMD Try the strawperry question. You will get a laugh.

I like the challenging question you pose; but its tell me exactly reason to use AI in medicine and teach the community how to do it “correctly” esp when to trust and when NOT too. It’s like GPS showing to keep driving straight and you can see that in 100 ft your 🚘 might turn jnto a 🚤

I got the exact same answer! Trust AI with medical decisions??

BCJ is committed to publishing the most impactful research in hematologic malignancies. We are thankful for the support of top researchers in the field.

Yes exactly. And you can see with medical questions that patients ask AI chatbots how it can make the same mistake, and patients are not going to know when it’s telling the truth, when it’s making a blunder (no matter how infrequently) that a human doctor will never make, and when it hallucinates and makes stuff up (no matter how infrequently) Some are leaping from ability of AI to solve complex diagnostic puzzles better than doctors, even expert doctors, with real life medicine where diagnostic puzzles are rare (most of the time we know what we are dealing with) and it’s mainly management, judgment calls, and counseling that’s involved. Ability of AI in medicine is tremendous. I don’t deny that. I’m excited to use it.

@VincentRK @CharlesMBrenner Opus 4.7 explains why it made the wrong choice. its very much like a human, certain words trigger it, AI: short trip = walk. No other reasoning comes into play. similar to Democrat: Trump proposed it = bad, no other reasoning comes into play.

@krystofkajanek I got the same answer today from the Claude app. Don’t know about Opus. But this is what the public uses now.. I have found so many hallucinations with AI programs pertaining to medicine. Good to not get too carried away. Thats all.

@VincentRK It is fake, Opus is not falling for this for a very long time

From 3 years ago. With @jamestalarico advocating for lower cost of insulin along with @NSmithholt12 and the producers of Pay or Die film that highlighted the problem. @jamestalarico is the real deal. I was so impressed by him then, and is a transformational leader. The kind we need.

Here’s the doozer.

AI is not intelligent. It’s analyzing and stringing together content we generated and repackaging it. Yes, we can use AI wisely to our benefit, and make our lives easier. Like if I’m a doctor, and I need to solve a computer or coding problem, AI is immensely useful and incredibly fast and efficient. But I’m not kidding myself that it’s intelligent. It’s finding and linking and presenting to me information I don’t know that other intelligent humans who do know have generated before.

What it told me today.

@Rfonsi1 @RahulBanerjeeMD @NCCN I think this will only be solved if we meet in person today 😀

@VincentRK @RahulBanerjeeMD @NCCN Despite so many good things happening in MM, the issue of PN has been minimized. And it has a major impact on QOL. The recent BENEFIt trial with SQ and weekly shows ~50% PN with ~25% being grade 3-4 (which is probably high). I will keep trying to get K...

I’m really proud @BloodCancerJnl is doing so well. We are getting a huge number of high quality submissions. It’s gratifying to take a journal from inception to this level in 15 years. Our goal is to be #1. @SagarLonialMD @Mohty_EBMT @n_gangat nature.com/bcj/

More here.

It’s the confident hallucination that bothers me. It bothers me particularly when it comes to clinical medicine.

Here is an honest answer from Claude. I asked it why it failed to find the answer to my software problem (that I dug up the web and found later on my own).