AI and the Safe Surgery Checklist

In this exclusive video, Jeremy Faust, MD, editor-in-chief of MedPage Today, chats with surgeon, author, and public health leader Atul Gawande, MD, MPH, about his book The Checklist Manifesto: How to Get Things Right. The two also discuss whether patients are actually safer now and how to bridge the equity gap in medicine.

Gawande is currently assistant administrator for Global Health at the U.S. Agency for International Development (USAID) in Washington, D.C.

The following is a transcript of their remarks:

Faust: Hello, it's Jeremy Faust, editor-in-chief of MedPage Today. Today, we are going to be joined by Dr. Atul Gawande.

Dr. Gawande is a surgeon, a writer, and a public health leader. He was founder of Ariadne Labs, a joint center for health systems innovation. He served as a member of the Biden Transition COVID-19 Advisory Board, and from 2018 to 2020 he was CEO of Haven, the joint venture of Amazon, Berkshire Hathaway, and JPMorgan Chase.

Let's talk about The Checklist Manifesto. I love that book, and I want to know how you see that book in our new context. A reader asked about this with the advent of AI [artificial intelligence] -- this collision course between the humanism in medicine that I know you and I both advocate for, but also the regimented perfection of AI and checklists. And they sort of are related.

I actually think about one of my favorite parts of that book, which is your description of the "Miracle on the Hudson." They open the checklist, and the first thing it says is: fly the plane. I love that, because everyone's getting down to the nitty gritty and they just forgot to do the big thing.

I feel like with checklists and with AI, I worry that we will forget to do the first thing that should be on every checklist for us, which is: treat the patient, do the right thing for them. If you do that, the rest of the list you can ignore.

Given where we are today, how do you feel about checklists?

Gawande: Well, it's all about how you design it. You pointed out that the checklist when Sully [Chesley "Sully" Sullenberger] was flying the plane over the Hudson, and you turned to it and it says: fly the plane. It's about enabling you to manage algorithms and recipes, but also your brain.

Some checklists turn off your brain and some checklists turn on your brain. The safe surgery checklist, the story I described [in the book] of applying it in that context, in a way that has reduced mortality from surgery, it's designed to turn everyone's brain on by saying, "Have you discussed in the room the goals of the patient and anything non-routine about this patient's past medical history before you proceed? Have you made sure that your team tells you the instruments that you need and anticipate are ready? That you've addressed how much blood volume and blood loss everyone should be prepared for?" So, it's turning everyone's brain on instead of turning them off.

In the world of AI, we are now entering a space where we have more and more tools that can enable us to be performing at a higher level than we would otherwise perform. If we use them in ways that turn off our brains for the spaces where we can add value, then we're in trouble. That whole conversation we just had about mortality is a conversation about something only a doctor can do, which is having a conversation with someone about what their goals are for their care and how to match the treatment algorithm to their goals.

I can get from AI the treatment algorithm, I can get the perspective and the diagnosis, but the combination of the two together is what the magic is.

Gary Kasparov has written beautifully about getting beaten by IBM Deep Blue as a world champion chess player and being humbled by a computer. Now, laptop-level computers are beating him at his game. But nowadays, there are chess competitions where it's the laptop computer plus the human being together that beat any computer on its own, because of the combination of a human being learning how to work with the strengths and the weaknesses of the algorithm. I see this exactly in our case.

I am experiencing some of the fighting at USAID where I am doing work in assistance in global health. We are bringing computer-aided detection, AI-driven chest x-ray interpretation, a chest x-ray system in a backpack, to countries like Nigeria. That is allowing us to screen in Nigeria; we've had a bump up of a 40% increase in diagnoses by having a machine that can look at an x-ray and give you a rating on the likelihood this person has TB [tuberculosis].

And it's in a place where you have fewer radiologists. Now, the radiologists still want to stay looking at these chest x-rays, and they can do things that this machine doesn't do. At the present time, it doesn't tell you if there are broken ribs or a pneumothorax or a lung cancer, but it tees them up so that they're able to very efficiently run through hundreds of screens for TB, put their brain on what's actually needed for the human component, and add more value with a handful of radiologists for the whole country -- well, not a handful, but it's such a small number compared to what you need.

And that is, I think, the future of our being able to be effective in this profession. It is going to reward those who are doing specialized functions outside of what algorithms can do and actually strengthen the generalist who is able to create that capacity to deliver for the patient with the arsenal of assets that are out there.

Faust: During your career, you've seen a lot of changes with respect to patient safety. Do you think that we are safer than ever before in the healthcare industry? Are patients safer? Because a recent New England Journal of Medicine article on inpatient care, including at our own institution, found that it isn't.

So I'm curious, what progress do you think has been made? And if so, why don't the data reflect that?

Gawande: We are manifestly improved. For example, medication errors are down enormously from the error of writing handwritten prescriptions. We're able to catch many ranges of problems that we simply weren't on top of before, [like] avoidable catheterization errors. We've seen our protocols making surgery markedly safer and getting better outcomes.

But you can still have a high rate of errors in the system. We are far from Six Sigma; we're like Two Sigma in terms of our rate of defects in the complexity of practice that we have. I bring a public health lens to this and at the same time a clinician's lens.

The public health lens is: our biggest gaps are failures to deliver consistently the arsenal of medicine across our income spectrum. The top 1% of Americans in income have an 87-year life expectancy, and that was before the pandemic, and the bottom 1% have a 72-year life expectancy. So, a 15-year gap. That is an enormous number of lost years left on the table. If we are able to deliver what we already know how to do now for that [group], we would close enormous gaps.

We are all learning how to deliver on a massive ambition here, which is we have 6,000 drugs, 4,000 medical and surgical procedures, and several hundred high-value public health interventions. We're trying to deploy that capability town by town to everybody alive, tuned to their individual goals and priorities like we just discussed earlier.

We have not learned how to do that, right? That is a high-complexity, not a high-reliability, entry.

We're delivering better than we ever have and [we also] have major errors, missing components, and a direction to go.

Comment