Minnesota Nursing magazine
Using big data to take the error out of trial and error
For people with elevated cholesterol, a growing array of medication options promises to more effectively reduce risks for heart attacks, strokes and related problems.
These drugs, known as statins, are prescribed to people for whom diet and lifestyle changes have not been effective in lowering dangerous cholesterol levels. Since no two patients are alike, selecting the most effective statin and dosage for each person can be a lengthy process of trial and error.
A team of health care and data experts is studying the experience of more than 37,000 cardiovascular disease patients with the goal of creating a tool to help providers identify the optimal individualized treatment plan for each patient. Analyzing health claims data as well as clinical and demographic characteristics of these patients over time, the multi-disciplinary team is identifying patterns leading to answers.
“By applying a computational approach to the data, similar to approaches used by Amazon or Netflix when they suggest products or movies you might like, we will be able to predict with a very high degree of certainty, which treatment can be most effective,” said Chih-Lin Chi, PhD, MBA, assistant professor at the School of Nursing and principal investigator on the first phase of this multi-year study.
“The goal is to help cardiologists and primary care providers be more proactive, instead of reactive, identifying the statin agent and dose that will have the greatest cholesterol-lowering benefit, with the least risk of adverse side effects and best possible outcome for the type of patient they are serving,” said Chi.
Too much for one brain to know
Today providers are inundated with information as discoveries are made and new drugs are introduced. Those who care for people with cardiovascular disease today have seven different cholesterol-lowering statin medications and dozens of dosage plans to consider. Challenged to stay current on new drugs and guidelines, many rely heavily on their own experience in treating their patients to determine the best treatment plan for each patient.
“ Using predictive modeling, this study creates a pathway to precision medicine, or individualized care, to treat patients.” – Jennifer Robinson, MD, cardiology researcher at the University of Iowa and study investigator
“We have current guidelines for treating patients based on their risk for a cardiovascular event, like heart attack or stroke, and their potential to benefit from treatment,” said study investigator Jennifer Robinson, MD, professor, cardiology researcher at the University of Iowa. “However, we currently have no way to factor in certain patient data, nor can we factor in the risk for harm. That is part of what this first phase of the study does.”
While statin use has increased dramatically over the last 20 years, many patients discontinue their statin medication within the first year, in part due to adverse reactions such as muscle, kidney or liver damage perceived to be caused by the medication.
Studying true cause and effect with precision
Working in collaboration with OptumLabs, which curates one of the largest health-related data warehouses in the world with de-identified data on 160 million lives over 20 years, the team began by carefully selecting its participant criteria. The initial research cohort includes more than 37,000 de-identified patient records.
The first phase of the project involves a study of adverse reactions to statin treatment. Subsequent phases will assess and weigh the benefits, risks, outcomes and costs of statin agents and dosage plans for individual types of patients.
“We seek to cluster different types of patients, analyzing the data hundreds of ways, to identify how each type of patient responded under different types of treatment plans,” Chi said.
The computer processing power needed to analyze tens of thousands of detailed records through a complex series of algorithms requires computational capacity equal to more than 500 times that of a standard office computer, said Chi.
“Chih-Lin’s work is very exciting for its methodological sophistication and application of machine learning methods to the analysis of health care data. The application of machine learning methods in health care is in its infancy,” said William Crown, chief scientific officer for OptumLabs, which is based in Boston, Massachusetts. “This kind of project can potentially find its way into translation and care very quickly by being imbedded into the tools that doctors and nurses are already using.”
Chi says the findings of this big data science work will be rigorously tested, and the results refined, through clinical trials. The end product will be a software application that may ultimately be integrated into the electronic health records system that provides individualized prescription recommendations to providers in real time, as they see patients and review their blood cholesterol levels in the clinical setting.
“Using predictive modeling, this study creates a pathway to precision medicine, or individualized care, to treat patients,” said Robinson, who co-authored the national blood cholesterol guidelines currently used by providers.
As the principal investigator and computational expert on the team, Chi says this work cannot be done effectively without a diverse team of people who bring essential clinical, research and entrepreneurial perspectives to the work.
The current research team is comprised of Thomas Clancy, professor, University of Minnesota School of Nursing; Jennifer G. Robinson, professor, University of Iowa College of Public Health and director, Preventive Intervention Center; Peter J. Tonellato, professor, University of Wisconsin-Milwaukee School of Public Health and senior scientist, Harvard Medical School; Terrence J. Adam, associate professor, University of Minnesota College of Pharmacy, associate director and core faculty, University of Minnesota Institute for Health Informatics; Chih-Lin Chi, assistant professor, University of Minnesota School of Nursing and core faculty, University of Minnesota Institute for Health Informatics; Jin Wang, PhD student, University of Minnesota School of Nursing; and the team at OptumLabs.
Finding the right records