Can Autonomous Vehicles Save Lives?

I consider myself a deliberative realist. My takes are usually more skeptical or nuanced than the academic consensus. At times, I flatly reject a study or a mainstream position. But there are two things in healthcare that I am particularly bullish on: GLP1s and AI.

I’ve long recognized the value of GLP1s, dating back to 2016 when the first large outcome trial was published, which I discussed in a State-of-the-Art Circulation review on applying an evidence-based framework to treating hyperglycemia in diabetes. I was an early adopter, and more recently expanded my research lab’s portfolio of work studying how Medicaid formulary policies shape GLP1 drug use, a potential policy lever to increase the dismally low use of these drugs.

With AI, the writing is also on the wall. Unlike with GLP1s, there are no large definitive trials confirming its benefits. But with AI, I don’t need an RCT to appreciate its value. I live it. I see how it makes my own ideas stronger. I see it compete with master diagnosticians. And I can already see AI outperforming humans in the physical world through autonomous vehicles, which serve as a potential solution to road traffic injuries, one of our most underappreciated public health problems.

I love living in San Francisco, despite its shortcomings. My one genuine day-to-day worry is my family getting hit by a car while walking. So when Waymos first arrived on the scene in 2022, I was suspicious. Anytime a Waymo was stopped at an intersection, I would jog across the street to avoid getting smushed if the sensors went haywire. Now, they’re so commonplace that I don’t think twice. It’s the human drivers I’m worried about. On more than one occasion a human driver has attempted to turn into the intersection while I was still crossing. These near misses are more than hypotheticals. Nationwide there are over 40,000 deaths each year from car crashes.

In 2014, San Francisco committed to Vision Zero, a 10-year policy initiative aimed at eliminating traffic deaths by 2024. While the city has implemented a range of interventions, such as street redesign, speed management, and enforcement, fatal and severe injury rates have proven stubborn, with more traffic deaths in 2024 than in 2014. On average, someone is brought to the trauma hospital where I work every 15 hours because of a car crash, accounting for roughly 1 in 5 hospitalizations requiring trauma surgery care. More recently, a car blew through a red light near my home, an intersection I’ve crossed countless times, crashing into a GoCar, ejecting its occupant and crushing the pedestrian, breaking his hip. Brutal.

So when I saw the Waymo Safety Report, I was both impressed and curious. Compared to human drivers, Waymo vehicles reportedly reduced crashes by nearly 90%. But company safety reports, while encouraging, aren’t enough. The question that mattered to me wasn’t whether Waymos crash less, it was whether autonomous vehicles reduce injuries. If Waymo’s report was true, trauma centers should see fewer patients. Then it hit me. Our UCSF team based at Zuckerberg San Francisco General Hospital and Trauma Center – The Hospital Outcomes and Policy Evaluation (HOPE) Lab – is uniquely positioned to answer this exact question. San Francisco is one of the few places where autonomous vehicle use is substantial, and our hospital is the city’s only Level I trauma center.

That’s why, in partnership with Waymo, our multidisciplinary team of investigators in hospital medicine, trauma surgery, emergency medicine, epidemiology, and biostatistics is launching the AVERT Study: Autonomous Vehicles’ Effect on Reducing Trauma. Rather than asking whether Waymo vehicles crash less, we are asking a more important question: are traumatic injuries from car crashes lower for autonomous vehicles than human drivers?

One obvious question is whether a study funded by Waymo can be trusted. I’ve spent enough of my career thinking about biases from industry-funded research to take that concern seriously. To strengthen confidence in our findings, we will have exclusive access to the data, retain complete independence over analyses and publication decisions, preregister our protocol, and establish an independent study advisory board.

I’m bullish on AI. But I’m also a scientist. Evidence matters. If autonomous vehicles truly reduce traumatic injuries, they could become one of the most important public health interventions of our generation. If they don’t, we should know that too. That’s the question we’re setting out to answer. Stay tuned to a publication near you.