Opinion | AI could impact health on a ‘planetary scale.’ Here’s how.


To fully appreciate the transformative potential of artificial intelligence in health care, one must think globally. When applied to people in extreme poverty, its impact could be as great as the discovery of penicillin.

That’s according to Karen DeSalvo, a renowned public health expert who is Google’s chief health officer. As she explained in a recent interview, AI could improve health on a “planetary scale” by greatly expanding access to health services.

Before speaking to DeSalvo, I was already convinced of AI’s potential to advance care in developed countries such as the United States. As I have written previously, such technology could improve diagnoses, help personalize treatment and reduce administrative inefficiencies.

But DeSalvo’s pronouncement brings my optimism to a new level. She knows what she’s talking about: A primary-care physician by training, DeSalvo helped lead recovery after Hurricane Katrina as the New Orleans health commissioner. She subsequently served as the acting assistant secretary of health in the Obama administration.

She told me about her current team’s initiatives to help American patients, such as using AI to monitor search queries for people who might be considering self-harm. Now, Google interrupts that “user journey,” as she put it, and provides a pop-up box for the 988 Suicide & Crisis Lifeline.

These are important efforts, but what really impressed me was how Google Health is using AI to address health-care access in low-income countries.

For example, such technology could make great strides in reducing maternal mortality. Every day, nearly 800 women around the world die of preventable causes related to pregnancy and childbirth. The World Health Organization recommends all pregnant individuals receive prenatal ultrasounds, yet about half in developing countries do not.

To obtain an ultrasound, patients must travel to a facility with a technician, who then must transmit the sonographic images to a radiologist or specially trained obstetrician for interpretation. These are major barriers in many parts of the world plagued by poor transportation infrastructure and a lack of trained providers.

In partnership with Northwestern School of Medicine, Google Health is trying to overcome some of these obstacles by piloting a low-cost, battery-operated handheld ultrasound device that community health workers can use with minimal training. The device’s images are uploaded onto a smartphone for AI to quickly interpret to estimate gestational age and assess fetal position.

The technology enables earlier and more frequent screenings so that patients can be referred to higher-level care before major problems develop. A study, published last year in the journal Nature, found that novice operators in Zambia with just a few hours of training were able to collect images that an AI algorithm assessed with a level of accuracy that’s comparable to existing clinical standards.

Another area of potential use is diagnosing tuberculosis, a highly infectious respiratory disease that kills an estimated 1.6 million people globally every year. Left untreated, TB is fatal in about half of infected patients.

Early diagnosis with chest X-ray is crucial for preventing TB’s spread and initiating lifesaving treatment. Unfortunately, many people live in areas that don’t have medical professionals who can interpret these images.

Here again, AI can help expand access. Research published last year in the journal Radiology found that an AI model performed as well as radiologists in identifying the telltale signs of TB on chest X-rays. In some cases, the model exceeded the World Health Organization’s performance standards.

Sign up for The Checkup With Dr. Wen Guidance on how to navigate public health challenges.

Of course, patients still need to get to a facility to obtain an X-ray. And once diagnosed, they still need to see a provider for treatment. But having an automated screening process removes one crucial barrier.

As DeSalvo explained, a good approach to AI is to identify “capacity or capability gaps” and then determine how AI might fill them. Done correctly, the technology can “help create tooling that raises the floor for everybody.”

What keeps her up at night is the possibility of “wasting” this opportunity. There has been some resistance about the adoption of AI in medicine, but whether people like it or not, the technology is already being widely deployed. “What I want is for this to happen not to medicine and public health but with medicine and public health,” she said.

This is an interesting perspective that I hadn’t fully grasped before. It shouldn’t be up to technologists to come up with solutions that the health-care sector must then adapt to; rather, health-care providers should be proactively identifying access gaps and working with companies to find innovative solutions. They should also be involved in implementing guardrails to ensure privacy, security and accuracy.

To do so, I believe, AI needs to be incorporated into medical, nursing and public health education. Future practitioners must be trained on this new tool that is quickly becoming standard practice.

“AI won’t replace doctors, but doctors who use AI will replace doctors who don’t,” DeSalvo said. The history of other major medical breakthroughs supports this: Antibiotics dramatically changed health care last century. The same will be true of AI in this one.


Leave a Reply

Your email address will not be published. Required fields are marked *