Rock Health Panel: The Power of the Practice Fusion Dataset
The following panel discussion with the Practice Fusion data team took place during the Rock Health Health Innovation Summit earlier this month:
Q: Let’s start with a quick introduction. What brought you to Practice Fusion and what you do on the team?
Lindsay Jorgensen: My name is Lindsay Jorgensen. In my prior life I was an epidemiologist, worked in Africa for two years, and then decided I wanted to get more into the health tech end of things. San Francisco’s the place to do that, so I got a job here as a data scientist.
Jake Marcus: My name is Jake Marcus and I’m a data scientist. I first heard about Practice Fusion when I read this article about EHR data. Practice Fusion was mentioned as having so much data that it was basically framing the discussion around using data for healthcare. I got really excited, went to the Practice Fusion website and found there was a data job right up my alley.
John Schrom: My name is John Schrom and I’m also a data scientist here. A few years ago I actually entered and won a Practice Fusion data competition. The thing I built I spun off into a RockHealth-incubated company, and I eventually ended up here as a data scientist.
Q: Data is a huge topic across Silicon Valley, particularly in health. What makes Practice Fusion’s data set unique in how data is gathered and what we’re able to look at?
LJ: The size, for one. And the fact that we have e-prescribing within the EMR, we have labs, we just started imaging—it’s all there and all centrally located within the same de-identified data set, opening up big possibilities for analysis.
JM: The fact that it’s all integrated doesn’t appear like it’s a big deal at first, but traditionally EHRs have been very siloed, even within the same company, hospital or clinic. There has been a ton of effort to link up data and get data to the point where it’s all sitting in the same place and you can start to analyze it. Since we’re web based and have one centralized database, we don’t have to do that.
JS: On top of that, Practice Fusion is a much more agile environment than a lot of hospitals, where I spent a few years working in analytic roles. The attitude of the data team and the engineering team here is totally different than what you would see in a hospital. Every couple of weeks we do hack days where we don’t really work on what we would typically work on, we just build something using data that we find interesting. That’s not something I’ve ever done in previous jobs. Also, the questions that are asked in a hospital tend to get very driven by the administration, like how many diabetic patients are there. We’re able to move beyond that into a whole different realm of questions.
Q: Lindsay, given your background in epidemiology, what you see as the public health potential of the data?
LJ: It’s unlimited if you ask me. I graduated with a lot of people now in academia and they aren’t capable of doing what we can here because of data access, the number of patients, and the variety of outcomes. I read a stat that most academic research conclusions take ten years to become common practice. Being able to access the data in real time, I think that we have the ability to revolutionize that process from ten years to ten days.
JM: For example, we’ve done some work around public health surveillance on obesity. A lot of the current data comes from surveys called the behavioral risk doctor survey, billed as the world’s largest telephone survey—really exciting now in 2013, right? They basically call up people asking how much they weigh and how tall they are. Not surprisingly, that has some bias. A dating site called OK Cupid did an analysis of their own data and found that men typically over-report their height and women under-report their weight. So we tried to see what we could do with our data. We pooled nurse and physician-reported weight and height. With four million observations of weight and height from patients across the country, we were able to calculate down to three decimal places the rates of overweight and obese patients. And all this comes out of ordinary physician practices across the country.
JS: On the infectious disease side we can also know before the CDC when flu and other things are breaking out. We get data directly from doctors as it is happening, and this reporting is free of typical bias in surveillance and reporting.
Q: On the other side of things, our data source is an incredible asset to Practice Fusion as a business. Jake, can you talk about the Insight product and some of its goals with our partners?
JM: One of our big projects is for pharmaceutical and finance companies to be able to explore aggregate trends for products and medications. There’s a long tradition of institutions going and scrapping together data from different hospital records to create registries of these trends, often by manually looking through patient records and doing surveys of physician prescribing behaviors and deviations from typical treatment guidelines. What we are trying to do with the Insight product is come up with these same registries out of electronic health records, that can explore drug profiles and prescribing behaviors while simultaneously keeping patient information more secure by stripping out any identifying information.
Q: As a final question, what has been your favorite project so far and what are you looking forward to working on in the future?
LJ: Lately I’ve been working on different stats to share with the media in addition to a larger project focusing on gender disparities and cardiovascular disease and whether women are being tested and treated less for cardiovascular disease even though they are at higher risk.
JM: Academic collaborations have been really interesting for me. In one I was looking at a drug that was widely prescribed, but was known to increase the risk of cardiac events. People thought this was because it interacted with a drug given to patients after a heart attack, but some lab data suggested it might increase risks without drug interactions. So we compared this against our own data set and the data matched up almost precisely showing that this drug did have a separate increase in heart disease. There are this moments of real suspense in data analysis when you are waiting for results like this, even though it doesn’t seem like a suspenseful field.
JS: People think of health data science as a lot of really in-depth analytics but there’s a lot of very familiar things that can be achieved with it. Facebook and LinkedIn are pioneering in the way they use data for social networks, like telling me all the people that I actually do know but don’t have a connection with already. Our data team can build algorithms that replicate things like this that have never been done before in healthcare, like building connections between different health ontologies and medical concepts to infer information that is missing.
