Big data paving the way for medicine
This week, a new way to study diseases and the best ways to manage them has been published in Science Translational Medicine. The work is based on what's called "big data". Put simply, as health records are translated to computer systems, opportunities are opening up to use that data to study and treat diseases in ways that we couldn't before. Joel Dudley and his colleagues are developing the medical equivalent of a social networking site for diseases. They can look for links between an individual's genes, clinical measurements and their lifestyle factors, and then plot where a person is on a "disease landscape" map and that map can reveal a wealth of new things about a person's prognosis. The team have already discovered that there are actually three different forms of type 2 diabetes, as Chris Smith found out...
Joel - The overall goal of the study was really to start to realise this idea of precision medicine which is a big term at least in the US where President Obama mentioned it in his State of the Union which is taking a new look at medicine with the very data rich lands. We have things like genetics and electronic health records, and things like that and can we really take a new look at medicine and disease and understand it with those greater precision. That's what we're trying to implement here. The approach we took actually borrowed some ideas from social networking. We represented each patient by all the pieces of information we had on them in the health records for example blood tests and height, and weight, and things like that and we connected patients up in a network. An analogy in the social networking would be if people have the same friends or the same interests in movies or books, you would say they're more closely connected in a social network. That allowed us to sort of map out the whole clinical network if you will of the patient population on Mt. Sinai which really in effect represented a map and allowed us to really start to understand almost like Google maps like, "where is your GPS data on this map and where do you fit in this patient population and where people have various diseases?"
Chris - Well, let's consider the diabetic side of it first because that was the thing you used as your example here. What emerged when you started to draw together these trends in all people who were judged to have type 2 diabetes?
Joel - As you may know, diabetes is a big health concern worldwide but especially here in the United States and then here in New York. So, we focused on diabetes initially. When we built this map, the question we asked is, "Where are the patients with type 2 diabetes? Where do they live in this map?" What we found in fact is that there were three distinct subgroups that were emerging in the data and we're able to show that the differences between these groups were clinically meaningful. So for example, all type 2 diabetics have increased risk of cardiovascular complications. But there's one group in our result that actually had increased risk even relative to type 2 diabetics. Another interesting group was the group that had increased cancer risk.
Chris - One can therefore presuppose that what we used to regard as a single condition may actually be more than one disease. If you can group them like this, does that mean then that not just their risk of certain disease complications but their likelihood of responding to certain treatments is going to be different? So, by being able to analyse them in more detail, we can give better prognostic and diagnostic, and treatment to each group individually.
Joel - That's absolutely the goal and the opportunity. So, I would like to point out, we had genetics on these individuals as well. So, we're even able to find genetic factors that were unique to each group. Those genetic factors are also important because they might give us an opportunity to serve as biomarkers for example. So, if you were a newly diagnosed diabetic, the opportunity could be that, now these genetic factors allow us to say, "Okay, well you're actually maybe a type 4 diabetic" in the future if it turns out that there is that many types of diabetes. And while we're worried about cancer for all diabetes patients but now, these genetic factors may be indicating that we should screen you much more frequently.
Chris - That's diabetes. Does the technique though mean that you could take the same strategy and apply this to a full constellation of illnesses?
Joel - Absolutely. We're excited about this study because the approach is generalizable. So, it wasn't designed to only study type 2 diabetes. It could be used to study any common complex disease that has many factors. It could be rheumatoid arthritis, multiple sclerosis and cancer. Countries such as Denmark or even the UK and National Health Service where you have such huge volumes of data really collected and centralised, we have huge opportunities to apply these methods. And hopefully, these maps will continue to get better because we can incorporate information such as digital health and wearables, and data coming from apps. I think eventually in the future, we will have a much, much higher resolution and maybe get to the point where we have a tool that's sort of like Google maps for health and for complex, and chronic diseases.
Chris - And for the people who have diabetes who you've considered here, are there any immediate repercussions for your average type 2 diabetic based on what you've discovered?
Joel - With research, once we've discovered something, we got to test it and test it again, and then make sure it's real and the next step here is to do a prospective study where we're really designing a study to replicate these findings, and the methods are quite scalable. So, we could even join forces with other health centres to really test this at a much larger scale and hopefully, translate these findings and to improve diabetes care more rapidly.