MIL-OSI USA: Assistant Administrator Atul Gawande Remarks at the World Health Assembly

Recommended Sponsor - Buy Original Artwork Directly from the Artist

Source: USAID

Strategic Roundtable on Artificial intelligence for health: opportunities, risks, and governance

Geneva, Switzerland

May 30, 2024

[embedded content]

Thank you so much for the chance to speak here. Director General Tedros, Excellencies, Ministers, distinguished guests, and delegates. It’s a great pleasure. I want to start by just saying, USAID is deploying AI tools with country partners now, and that has given us a deep understanding of the great opportunities and great risks and how we are beginning and just beginning to manage them.

First thing to understand is that AI is not going to replace humans and health care in the near future. It is, right now, providing clear potential for augmenting human capabilities for health, so that human beings can do more and manage with less complexity and less cost. Many examples of human conditions that have required much higher levels of expertise. As a case in point, Director General Tedros referred to the deployment of chest X-rays, digital chest X-ray systems, for TB.

In TB, a great difficulty is that it is often latent and unseen without symptoms. So screening with chest X-rays is very important, but the capacity with radiologists to have same day testing of chest X-rays and readout and diagnosis and action makes that very difficult to deploy. So we have seven countries around the world, including places like Vietnam and Nigeria, where we are now deploying AI-based, computer aided detection software systems that instantly read chest X-rays and give people diagnoses. Using it in high burden countries, we are seeing that these tools have increased, for example, in Nigeria, case detection of TB cases by more than one-third in the course of a single year when coupled with molecular diagnosis.

Now, one of the critical things is there are many applications that are potentially coming down the pike. AI-based ultrasound for assessment in pregnancy, AI-based skin lesion detection. Those tools are out there and available. We went with this tool first for chest X-ray in TB. Why? Because WHO did an evaluation and provided an initial set of recommendations that could give us some normative guidance? Because, based on that guidance, countries we’re working with had confidence that they could begin to have their regulatory systems drive these capabilities forward. And then further, we have a test in case the AI is wrong. We are not relying on this coupled with a molecular testing capability for TB that can diagnose at the same time, we can confirm whether someone has TB or not, and then proceed to treatment and not base it solely on an AI reading for a chest X-ray. It gives us either a test or a human assessment that helps us proceed to the next use.

There are other uses besides diagnostic interpretation. We know it could help reduce the burden on health care workers in Eswatini. We are supporting partners to explore the use of generative AI in speech to text models to reduce their clinical documentation and monitoring and evaluation forms and reporting that they must do the types of systems we’re actively deploying at the Mass General Brigham, where I used to be employed before I came into the government. And we’re seeing these systems move around the world, potentially in equitable ways. There are many, many more applications.

The critical components of being able to address these needs are where USAID investments in the digital health space are going. We have long had a set of work around how to enable better digital solution use in countries. The critical architecture of systems for digital health are the same architectures that are required for use of AI tools. They’re not fundamentally different. You have to have ways to identify people and connect their data between different kinds of data sets. You have to enable clinicians to access that data and use it, and then you have to ensure security and privacy in the process. Same with the AI tools. And then you have to ensure accuracy along the way.

That has required more and more that countries have a digital strategy at a country level, and that that strategy is overseen by a digital leader who can make sure that all of the components that have to be executed are being executed. And we’re supporting training of country officials, people completing these plans, and enabling this to move out in the world. So my final point, I’ll just say is I’m seeing how these tools are deploying in the real world. There is great opportunity and there are great risks, but health is the space where we can demonstrate that the good will outpace the bad.