Algorithms Will Drive Future Health Gains, Dean of Stanford Medical School Predicts

“We are now at the point where innovation is at the algorithmic level,” says Dr. Lloyd Minor.

This piece appeared in The Wall Street Journal on April 21, 2017.

BY STEVE ROSENBUSH

Medical science blazed new paths of innovation with the development of vaccines and antibiotics, and new ways of treating heart disease and cancer. Going forward, continued advances in medicine and health increasingly will be tied to the ability to interpret massive amounts of data, according to Dr. Lloyd Minor, dean of the Stanford University School of Medicine. “We are now at the point where innovation is at the algorithmic level,” he said.

Dr. Minor, a former provost of The Johns Hopkins University, is known for discovering a disorder in which sound or pressure induces dizzyness in patients. He developed a surgical treatment as well. Since his arrival at Stanford in 2012, the School of Medicine has established a department of biomedical data science and has struck a data science partnership with Google. It’s part of an effort to move beyond personalized medicine to the concept of personalized health, which focuses more on prediction and prevention, according to Dr. Minor.

Here are edited highlights of his conversation with Wall Street Journal editors, which focused on the influence of data science on health care and medicine.

You talk a lot about how data science and medicine are coming together. In practice, how does that work?

Lloyd Minor: We know that a lot of hospital infections are transmitted by our hands and hand washing can be tedious, but it is extremely important. So we began a project with the folks in computer science to develop an AI-based system that can look at the hand-washing stations outside of an ICU room and determine when a person has washed their hands. The interlock on the patient’s room is only opened and triggered when a person has gone through the appropriate hand washing. Now, it can be overridden if there is an emergency. But if it is overridden it gets recorded and there needs to be an explanation of why it was overridden. I was skeptical at first. I thought you could fool it by … not actually rubbing  your hands. Well, it turns out to be pretty darn good.

We are validating the technology. It is being piloted in the ICUs but not yet driving the opening and closing of doors. The technology is validating that the hand washing is taking place and we are getting feedback from others on how it works.

When do you expect the technology to be fully deployed?

LM: The timetable is yet to be determined. But it is just one example of AI leading to a very practical solution to what has been a great dilemma. It turns out that our hands emit a very powerful infrared signal that is quite easy to distinguish. You can also distinguish when soap and water touch the hands and when the hands are rubbed together.

The point here is to prevent infection, not treat it. How does this project relate to your view of how health care and medicine ought to evolve?

LM: Everyone has heard of precision medicine. It was a major initiative, and hopefully it will continue to be a major initiative. We of course do precision medicine, applying genomics and Big Data science to the treatment of severe and acute diseases like cancer and heart disease. Every active medical center does that and will do it a lot more in the future. But we have a different vision and a vision we hope will be significantly more impactful in the intermediate to the long run. And that is precision health. The goal is to be predictive, preventive, and to cure precisely when disease occurs. It begins really with prediction.

That is genomics but not just genomics. It is also understanding the interplay between genomics and lifestyle and behavioral factors. I am excited about a project called Baseline. It will enroll 10,000 people, including 5,000 in the Bay area and 5,000 in North Carolina. We brought Duke into this because we wanted a demographically balanced population. We will have everything we can imagine about their health measures .. whole genome sequencing, a complex immune panel. There will be a new generation of wearables that the volunteers in this study will use. One third will be as best we know, healthy;  one third who are or have been treated for cancer; and one who are or have been treated for heart disease. We can track them longitudinally and for the first time, have enough baseline data so that when something comes up later on, when can look back and say well there might have been an early predictor of that. Rarely today are we able to  make those sorts of statements.

What role does AI play in personalized health?

LM: This goes back seven or eight years. We formed a de-identified database of all the clinical data in the Stanford health-care system. It has provided the resources for a number of advances. And about four months ago we signed a business affiliation agreement with Google to work with them on genomic analysis. Google Cloud will be the repository for our genomic data and they will have their data scientists working with our data scientists on the next generation of algorithms to interpret that data.

It encompases decision support. By and large most of us today can get more information from doing a Google search on the symptoms that our patients are coming in with than we can from electronic medical records. It’s because the electronic medical record companies … haven’t focused on developing those sorts of decision support algorithms … I hope … there will be alternatives that will be more focused on the delivery and quality of care.

We saw this as being such an important area that we established a new department in the School of Medicine, about 18 months ago, in biomedical data science. And we have first and second year medical students taking courses in computer science and advanced statistics. In one case, we have a medical student teaching an undergraduate course in the history of science. It enables us to leverage the interactions with the rest of the university in really meaningful ways.

Do you see a shortage in trained computational biologists and related specialities?

LM: There is a huge shortage. Those people are hired long before they finish their degrees.There is a huge demand for data scientists. We are now at the point where innovation is at the algorithmic level. There has been a lot of predictive modeling on readmission, based on the parameters of discharge. That modeling is now driving decisions about what kind of monitoring a patient needs when they leave the hospital. And that type of modeling is also extended to the neonatal ICU. Those patients can be doing great and then crash. Without data analytics you get little warning.  There are algorithms now that provide a better indication of when the crash might occur.