MedWatch

And we wanted to make sure that we were getting in front of some of these more dicey issues like governance, like safety.

Do you know that today there still is no AI safety program at the national level in the U.S.? By that I mean, if you take a drug and the drug has an adverse reaction, do you know what happens? There's a program from the FDA called MedWatch.

And MedWatch basically has been legislated that if you find an adverse reaction, you must report it to MedWatch.

Now, does everybody report it? No, but many people do.

And if you work for a pharma company, you'll know about it because the pharma company enforces this one without fail.

But that's also a regulatory thing.

If someone goes to the hospital, say they go to the ER, and they have a serious adverse event, which is different than an adverse event, then technically that has to be filed with the FDA via MedWatch.

But if someone takes more than your suggested use, The onus is on the consumer.

We didn't tell you to take that much.

Yes.

Or the FDA's MedWatch program in the U.S.

I used to think of passive reporting like a giant national suggestion box.

But a better analogy is trying to map a city's potholes by waiting for drivers to call a city council and complain.

By integrating spontaneous reporting centers with active monitoring in targeted medical colleges, they are establishing a robust benchmark for drug safety in a massive, genetically diverse population.

But here is the logistical nightmare.

You have India running PVPI, the U.S. running MedWatch, the U.K. running the Yellow Card scheme.

Millions of doctors are generating data.

I'll get in- Let's, let's stay with Vamos for a second, because we have been sort of pushing for, um, some type of way for the, th- some type of reporting mechanism for, uh, errors that AI makes.

I mean, just like, y- you know, you know my background's in, in pharmaceuticals, and the MedWatch system lays out a pretty reasonable, uh, approach to reporting, uh, adverse events.

And as we've pointed out many, many times in the podcast, it doesn't exist right now in, in, uh, healthcare AI.

And that's where the governance comes in, that's where the work at DCI comes in.

Yeah, so Peter described it very nicely as kind of air traffic control for AI systems.

One interesting thing about what they're building, they're, they're, I mean, it- it's institutional effort, right? So they're doing it within an, within an institution.

Uh, and they are open sourcing it so others can use it and contribute to it, but it doesn't quite get to, you know, one of the th- ideas that you as-- we're pushing, which is MedWatch for AI, right? We need the equivalent of a MedWatch system that monitors industry-wide what's happening-

Right.

However, what we're going to be discussing today is for acute agitation, which is an off-label use, and we'll see doses of around five to ten milligrams intramuscularly or intravenous, with a max recommended, like dose at one time of ten milligrams.

So in two thousand and one, the FDA published a black box warning due to reports of fatal arrhythmias with droperidol, regardless of the dose, and this was based on two hundred and seventy-seven case reports submitted through MedWatch.

However, independent reviewers requested the data from both the FDA and the manufacturer, and they-- Upon evaluation, they found that only sixty-five cases actually had a cardiac symptom tag, and of those sixty-five cases, thirty-five patients passed away.

Twenty-five of those thirty-five deaths were from cardiac arrest.