Every month, the CTSI Stories Blog will post excerpts from ongoing conversations with the institute’s co-directors.
Martin, you’re not totally new to the CTSI, but I’m sure some people are more familiar with you than others. Could you tell us a little about your background?
Certainly. I started out as a transplant nephrologist — a medical kidney doctor who takes care of people with kidney transplants — and my clinical practice has been taking care of folks who have complicated immune issues for kidney transplantation. My laboratory, meanwhile, studies B cells — which make the antibodies that respond to vaccines and can cause kidney rejection. We are interested in learning more about how that biology works, how the cells differentiate, how many you need to create a rejection episode, what the molecular signals are for vaccine responses, and so on.
The tools we use in the laboratory involve many of the same mathematical methods used in data mining, which is the process of looking at complex data sets to reveal relationships between elements. We then go back to the bench and try to figure how these relationships work to create an immune response. And the mathematics used for that are, at their root, some of the same mathematics used in social network analysis. I realized that about two and a half years ago after reading a humorous article about finding terrorists in a network — the example was Paul Revere as the central revolutionary in his network — and I became very interested in clinical problems and population health problems that could be studied with the same methods.
Very interesting. So that system approach is what appeals to you most now?
Well, I’ve had, for a very long time, an interest in what people call computational biology — mathematical and computer analysis of data, and creating models of biological systems. I’ve always been fascinated by these models, because they give us a way of understanding how nature works, and, sometimes, how naïve we can be as scientists in terms of our theories.
So let me explain that. When you model a biological system — whether it’s vaccine response or kidney transplant rejection or development of B cells or the healthcare outcomes of a population of people — you begin with an idea of how the world that you’re studying works. The model you build forces you to create an image of that world. If your models are quantitative and predictive, they provide a reality test for your ideas. What is really interesting is that the models are most useful when they’re wrong, because it tells you that the way you thought the world works is not the way that it really does. It tells you that you’re missing something. And then you can go look for that something. If you’re lucky, you find something really interesting.
Can you give a real-world example of this type of system-wide work?
One of our projects is looking at patients who go to the ICU, then get better and recover and are discharged to a medical floor, but then come back to the ICU because they’ve gotten sicker again. People who have that kind of pattern have a rather high chance of dying during their hospitalization. So we took some of the same methods used to look at patterns of gene expression and applied to the hospital admission and transfer data, and lo and behold, two things popped out.
The first was that we could graphically identify the patterns of who was returning to the ICU. They were a small fraction of those admissions, but accounted for a very large portion of our cost per patient. We are now using informatics to ask what medical conditions they had, what was going on right before they went back to the ICU. We want to put together a risk profile, an early warning system, that would tell us “This person has a high probability of ending up back in the ICU.” Then we can try to change the outcome. That’s exciting, because we may have a chance to use data to save lives.
The second, really amazing thing, was that we were able to create a map — a flux diagram — of all the transfers within the hospital between the different floors and units. That kind of diagram is used in basic science to look at how organisms metabolize things by looking at all the nutrients and chemicals that go into a system, and how they’re shunted to various chemical reactions and come out as products. It’s also the same mathematics that FedEx uses to figure out how many planes they need and how much they need to load in each one, and so on. So with this map of the hospital, one project we’re working on now is to say “OK, we’ve got flu season here, and we’re going to overload this part of the hospital. What other parts are going to get stressed, and how could we creatively move patients around to provide better care, shorter ER stay times, and better outcomes?” The beauty of these approaches is they look at things as systems. It’s not just one part. Everything is connected to everything else.
This type of science, systems analysis, is very exciting right now, and we have a chance to see the world as a connected network using these tools. My health informatics group, the Rochester Center for Health Informatics, is collaborating with the Institute for Data Science at the University of Rochester and Tim Dye’s CTSI Informatics group on these types of projects. Rochester is exactly the place where we can do this work.
Previous director’s updates:
February 2015 – Nana Bennett discusses the CTSI’s Seminar Series on population health.
January 2015 – Harriet Kitzman reflects on her time as a CTSI co-director.
December 2014 – Karl Kieburtz offers his takeaways from the CTSI all-hands retreat.
November 2014 – Nana Bennett speaks to the expansion of the role of the CTSI’s Community Advisory Council.
October 2014 – Harriet Kitzman discusses the science of team science.
September 2014 – Karl Kieburtz talks about why the CTSI is beefing up its informatics team.
August 2014 – Nana Bennett discusses the new Population Health pillar.
July 2014 – Harriet Kitzman offers her takeaways from the Mini Summer Research Institute.
June 2014 – Karl Kieburtz gives an overview of the CTSI’s six pillars.