That's why we built PaiX Navigator.

We knew from the start that diversity is king. We would have to look at different ethnicities for sure, which are not represented usually in the clinical trials as well.

Once you develop your algorithm on a certain amount of demographic population, your solution is working on that population. Once you go to another part of the world where you don't know the demographic of that region, your AI will definitely fail there.

One in every four Africans have adverse drug reactions to warfarin, the world's most prescribed anticoagulant. We already have algorithms that accurately predict overdose risk for Europeans. But if you are non-European, those algorithms don't work well.

PCSK9 inhibitors, a medication very effective in lowering bad cholesterol and preventing heart attacks, were found because some African ancestry individuals were included in a study. If those Africans were not included, today we would not have that medication. More people will have died. Maybe millions of people will have died.

My own cancer test came back with a variant of unknown significance. What that really meant was: I'm Latina, I'm from Colombia, and my genome isn't represented in international databases. The science simply didn't know what my result meant.

Most AI models perform beautifully in the lab, but they frequently fail in the real world. That could be a biomarker that looks bulletproof on an internal test, but collapses in external trials, or a tumor classifier that breaks when applied to slides from a new scanner or staining protocol.

Unless you take an equity-focused approach to genomics and the gathering of data, supported with the engagement of participants and with an awareness of ethical issues, you have data, but it has no value. You don't have anything if you don't do that.

We do not get adoption unless we have trust. We don't get trust unless we have understanding and explainability of these systems. And if we don't get adoption, we don't have impact. There is a chain of consequences we have to set up to get to actual clinical impact.

If you're not putting diverse data in AI, you're not going to get diverse outcomes, and this is where the last mile of AI fails. We have to start collecting the data from all those populations, not just from the subset of people that walk through the door in a clinical trial site.

There needs to be a way to do basic research on those so-called rare diseases, create venues for researchers, positions, clinics, hospitals, universities, to find pathways for their research to get into translation. That could be through feeding a centralized database, which then could become an AI.

Most clinical trials, because it has always been done that way, are done with white, middle-aged subjects. These clinical trials are completely neglecting that a drug that works in a white, middle-aged man might not work in a woman in the Caribbean who is of African descent.

The sad story is that doctors have learned to silently adjust. Especially in the Global South. They've learned to silently, in their own practices, behind doors, adapt reference ranges, adapt treatment norms, adapt dosages based on their own trial and error experience.