Robert S. Warren, MD is a Professor of Surgery and a specialist in gastrointestinal and liver cancer. Dr. Warren joined UCSF Medical Center in 1988. Highly respected by his peers, Dr. Warren was named to the list of U.S. News “America’s Top Doctors,” a distinction reserved for the top 1% of physicians in the nation for a given specialty. His research focuses on the biology of colorectal cancer and how it spreads.
Synced sat down with Dr. Warren to discuss his experiences, MORE Health, and the role of AI in health.
Please tell us about your role at MORE Health
I became involved out of my developing interest in global health. As I progressed in my career, I started thinking about ways that I can contribute in a much broader way. So rather than seeing twenty-five patients in the day, I put my thinking cap on to figure out how I can treat 2,500 patients — and in a better way.
From a doctor’s perspective, what do you think of AI’s role in health care?
In the United States, AI is sort of a silent process according to most physicians — unless they’re part of the organizations that have a considered effort to do AI. A good example is the Memorial Sloan Kettering Cancer Center which has been seeing patients with cancer for decades and collecting data. They’re starting to analyze it in a machine learning environment.
It really comes down to medicine to say: Can we make a diagnosis better? Can we treat a patient in a better way or more effective way? Those are still the same questions we ask ourselves every day when we see a patient in the office. And so those are really the same questions we have to ask when it comes to AI.
What do you think of the recent media stories about AI replacing doctors?
I don’t think that will happen at all. I think the kind of work doctors do may change. For example, we might need fewer radiologists, as they could develop novel approaches for new imaging methods rather than spending their time in a darkroom reading X rays.
Physicians’ roles are already changing, even without AI. More precise methods of describing a tumor is just another way of characterizing molecular features. But we still need the oncologists to really put it in context.
Sometimes you need to lay a hand on a patient, talk to them, look at their eyes, talk to the family — to really come up with the right decision. I really don’t see AI replacing physicians. I see it as another tool in our little doctor bag that will make us better.
What are specific issues MORE Health is addressing?
The average amount of time a patient spends with a doctor in Beijing at a new patient visit is five minutes. So how can we improve the quality of life from a medical standpoint for as many people as possible. That’s our goal.
In China, a patient might be seen at a community hospital and have some tests, medical imaging and possibly a biopsy done. Then the patient goes to the provincial hospital because they’re not confident in their doctor, and everything is repeated.
It’s been a challenge to get all of the records going back to the original diagnosis. That’s where our boots on the ground in China has made a huge difference. Our folks in China have made a tremendous effort and gathered up those pieces of information, translated and uploaded them to the platform.
What information do you look at before diagnosis?
What I learned in medical school in terms of diagnosis, that 90 percent of the diagnosis comes from looking at the patient, talking with the patient and examining patients. The other ten percent are the blood tests, X rays, and everything else that goes into the big expense of starting to see a doctor.
The challenge is how do you improve that 80 or 90 percent from the patient – doctor interaction when it’s remote? The platform’s telling us and we’re getting better at that.
What is your focus at the UCSF?
I’m involved in products that use big data to try to understand biology. I’m a specialist in cancer, that’s what I study, and I have a laboratory and a clinic. But what’s nice about being at an academic institution like UCSF is we have access to databases from around the world that correlate characteristics of tumors, mutations, gene expression, protein expression, and clinical information with a clinical outcome.
Most AI applied in medicine is still used for predictive analytics, for mining, for big data, for medical imaging, pathology, for the evaluation. Ultimately there will be other people around the country who have developed algorithms for treatment suggestions.
How can AI be implemented in patient care?
I can speak to the issue of cancer. I think both the diagnosis and the treatment of cancer will be dramatically helped by AI. The term that’s used in cancer is “precision oncology.” And it’s a misnomer, because right now it’s not very precise. We identify a mutation. We have a drug we think will target that mutation. We can say, yeah, that’s what you should do. But I can tell you at least half the time that drug doesn’t work, even though it should.
We keep requiring more and more data to understand the basis for sensitivity or resistance to a therapy. It’s an evolving field, so how do you keep up with all that information? I think machine learning will help synthesize that evolving landscape of diagnosis and treatment in an individual patient.
The Atom project is a collaboration with UCSF, GSK, Frederick National Laboratory for Cancer Research, Lawrence Livermore National Laboratory, which has the largest and fastest computer in the United States right now. And so that’s where we’re focused right now. I’m taking available data just from a single pharmaceutical company and applying AI principles to try to understand the biology of cancer better. But there’s no reason that every large pharmaceutical company couldn’t contribute their data. Obviously, some proprietary information would have to be withheld.
The idea is that if you combine all of the massive amounts of data with the massive computing power at Lawrence Livermore, it will somehow advance the field. We will find out exactly how, it’s happening as we speak, and ultimately will feedback to individual doctors everywhere.
What are your thoughts on AI interpretability or “black box” issues?
It’s easy for patients to say: ‘I’m going to this doctor, he’s been practicing for thirty years, knows what he’s doing, I trust him.’ Trust is the key. So how do you trust a computer? I think what you have to say is ‘the doctor is being advised by the computer, not being told what to do’.
If the doctors are convinced over time that AI is a helpful tool, then that’s what it will be. Just like a stethoscope, it’s a tool.
Journalist: Tony Peng |Editor: Michael Sarazen