6 thoughts on population health with Allscripts Analytics CMO Dr. Fatima Paruk

About three years ago, Allscripts decided to ramp up its position in the population health space, acquiring health analytics firms like dbmotion and hiring public health experts like Fatima Paruk, MD, who serves as CMO of what is now Allscripts Analytics.

Since its founding, Fatimathe Allscripts Analytics division has built its data science team, researched public health crises and deployed its findings into Allscripts' client-facing tools. "We're utilizing all of the fancy buzzwords you see in every article out there, with AI, machine learning, neural networks," Dr. Paruk says. "That's what we are up to in our office."

Dr. Paruk sat down with Becker's Hospital Review at the company's annual user conference Allscripts Client Experience in Chicago to discuss first steps, everyday challenges and interesting findings Allscripts Analytics has explored in recent years.

Editor's note: Responses have been lightly edited for length and clarity.

On the messy first steps in data science

Dr. Fatima Paruk: "Our payer life sciences unit has data rights to a small percentage of our client lives from across the United States in the ambulatory space, so we have a data lake, which is about 50 million longitudinal, de-identified lives, which we are able to use for research and development of predictive models. We got this fantastic data and we thought, 'Oh, it's going to be super, super easy' — and low and behold, it was not super easy. Things like lab values were stored thousands and thousands of different ways, so we spent about a year and a half normalizing and cleaning this data up. We used things like natural language processing to mine our problems and really figure out what's hiding in the data. It was taking the long way around in terms of the pre-processing and staging of the data, but from there, we could start to research things like chronic disease management."

On how murky raw data slows research

FP: "We started to do some outcomes research and some chronic disease work, concentrating on six chronic diseases at first, including prediabetes as one of our focus areas. If you can stop prediabetes before it gets to diabetes, you're not only saving a ton of money, but impacting patients' lives. We started to look at those prediabetes patients; we pulled out about 3.5 million of them out of the 50 million. The problem is, and this is across the board, what physicians are entering are generally not coded data. It's not standardized data in the way you need it for analytics. If you realize there's, say, 5 million diabetics in the population, what you'll find is half of them are not coded. We looked at labs and medications and used the CDC criteria for prediabetes to determine who's truly in the population, based on indicators like lab criteria, BMI, age groups and blood sugar in the chart."

On using data to fuel meaningful partnerships

FP: "When we did this analysis of the 3.5 million patients, we found about 80 percent of them were converting to diabetes in a five-year period. That was alarming. It's a huge, huge problem. Being able to demonstrate that progression is where our relationship with the American Diabetes Association started to get a lot stronger. Now I sit on their digital innovation committee, and I'm working with them on larger initiatives where we're trying to take the ADA's 2017 guidelines for prediabetes management and put them into all of our products. We got the leaders from all of our different products to agree this is a huge problem, because of our research in the field and because we managed to get the backing of the ADA."

On the opioid epidemic

FP: "The CDC is struggling because you have these state-level databases, the prescription drug monitoring programs, but they can't share it among all the states. That is something we could do, because our data is 50 million lives, it's longitudinal, and, on top of that, we're getting it with monthly refreshes. We can talk about the footprint of what's happening with the opioid epidemic, because we can see the prescriptions going up. We decided to plot these patterns against external data like county death rates and drug treatment programs. We see, as the opioid prescribing rates go up — and up, and up — the death rates do, too. That correlation is just one way we'll be able to go back and show our docs, 'Hey, by the way, your opioid prescribing rates impact people in these counties.' The next step is taking the analytics and throwing it into clinical decision support to deliver best practices."

On the 'holy grail' in data analytics

FP: "Our work has led to a lot of predictive modeling, where we're taking this big dataset, determining who's most likely to have an outcome and correlating with external data sources to create this picture of what's happening. That, to me, is the 'holy grail.' We can paint that picture, show that we could be saving hundreds of thousands of millions of lives, and turn that into relevant clinical decision support at the point of care.

Our predictive model for opioids basically says if you're on the fourth medication, you're 42 percent likely to get fentanyl next. If, when my colleagues are writing a script through e-prescribe, if at that fourth script I'm showing them this patient has a 42 percent chance of ending up on fentanyl, and then on the street, that's going to change my colleagues' management. That's what this is all about: delivering that predictive analytics experience at the point of care."

Advice for hospitals: 'Predictive analytics is not hard'

FP: "Getting involved in predictive analytics is not hard. We're already using predictive analytics in aspects of our lives — Amazon knows what you're going to buy before you buy it! So why are we not taking this approach to managing our patients? If I know a patient is going to end up addicted to fentanyl, why am I not managing that patient? It's bearing out on our end, on the vendor side, how to seamlessly embed this in our workflow.

On the CIO side, if you're looking to develop your own algorithms, looking to develop in-house, I told you we spent the first year and a half just cleaning data — so you've got to do all that. The second is ensuring you have the technology to balance those predictive models up and down, from your point of care, right up to where the magic is happening in the background. If you're looking to bring in outside algorithms, you look at what's out there, who's done it well, and then you've got to validate the algorithm in your own population."

More articles on health IT:
Paper records are here to stay: 4 questions with Illinois Pain Institute's Dr. John Prunskis
UChicago Medicine and Google — a data-driven duo to watch
Startup Insider: 3 questions with Cricket Health CEO Arvind Rajan

© Copyright ASC COMMUNICATIONS 2017. Interested in LINKING to or REPRINTING this content? View our policies by clicking here.

 

Top 40 Articles from the Past 6 Months