Doctors, Decisions, and Data: Analytics for Better Health Care
By Scott Barsotti
A Heinz College faculty member and alumna recently teamed up to improve decision-making and care delivery for patients with chronic conditions. Dr. Rema Padman is a Professor of Management Science and Healthcare Informatics at Carnegie Mellon University’s Heinz College. Dr. Yiye Zhang (Ph.D. ’16) is an Assistant Professor of Healthcare Policy and Research at Cornell University’s Weill Cornell Medical College.
The power of data analytics already guides your purchasing behavior on Amazon and helps you find the next show you’ll binge-watch on Netflix. But what if those same algorithms could help you and your doctor choose the best strategy for treating your health problems?
Such is the motivation for Rema Padman and Yiye Zhang’s new study published in the American Journal of Managed Care. Their collaboration—which began while Zhang was a Ph.D. student of Padman’s—sought to utilize statistics and machine learning algorithms to better predict future clinical pathways and best practice treatments.
Padman explains that—in addition to their own knowledge and experience—physicians rely on the results of clinical trials to guide their choices about how to treat a particular patient. These large-scale randomized studies have provided information on the treatments for many major diseases, and are the gold standard for guiding health care decision-making.
But according to Padman and Zhang, these trials cost a lot of time and money, and they simply can’t be used to study every single condition or complication a physician might encounter. Meanwhile, the results of the studies that have been conducted can only tell us about the population in which that experiment was performed, and won’t apply to every patient in every setting.
Yet, even without good information to guide clinicians in every setting, people are still being treated for a variety of complex chronic conditions on a day-to-day basis. And as those patients receive care, health care facilities are collecting incredible amounts of data about the standard treatments doctors are providing.
“Every visit, every prescription, every lab result—it’s all being recorded,” said Padman. “We have data on thousands of patients being treated in multiple settings. Why can’t we learn current practices from this vast repository of data, associate them with outcomes to learn what works and what doesn’t, and make recommendations on best pathways for individual patients?”
An advantage of this data-driven approach is that it may help physicians assess better routes of care for individuals suffering from multiple chronic conditions—something even the best clinical trials aren’t able to do.
Padman and Zhang’s research was able to expose variations in costs among patients who are clinically similar, as well as variations in clinical complexity among patient with similar costs. They were able to group patients based on how their condition was progressing and identify trends in order to predict how a group of patients’ conditions will evolve and what their next clinical visit might entail, including treatment options.
“The reality is that few people have just one chronic condition; it’s typically a multiplicity of these,” said Padman. “Understanding the evolution of their condition, their compliance with medication and treatment regimens, and how physicians might best treat them requires a combination of analytical approaches—not just machine learning, but operations research, statistics, and behavioral economics as well.”
Padman and Zhang examined decision-making challenges that arise in managing Chronic Kidney Disease, a condition that currently affects more than 26 million Americans and carries a cost in excess of $23,000 per patient per year.
Partnering with Teredesai, McCann & Associates (TMA), a local nephrology practice, Padman and Zhang were able to examine the care different patients received over the past two decades, grouping them into different populations based on paths of treatment. (TMA has been recording its patients’ information in an electronic database since 1994.)
“Using data analytics to help manage patients with multiple chronic conditions—like ours—empowers doctors to spot trends, both good and bad, in a patient’s plan of care,” said TMA Chief Information Officer Linda Smith. “It enables the delivery of a quantitative form of medicine that can detail an exact measurement of the health risks of a condition and the benefits of compliance to the plan of treatment.”
Padman says that this line of research would be impossible without the cooperation of health care organizations that are willing to share their specialized knowledge, time, data, and human resources to help improve the quality of health care delivery.
Zhang praised the TMA team, remarking that they were extremely responsive to queries in addition to giving access to data.
“The staff and physicians have spent many hours to help us,” said Zhang. “Without that kind of supportive collaboration, it’s very difficult to understand the complex nuances of the health care delivery context well enough to achieve any progress.”
Padman also credits the collaborative environment at Heinz College for making this kind of research possible. The school’s Center for Health Analytics and the Health IT thrust of iLab represent some of the many ways that Heinz College researchers, students, and faculty members innovate and drive improvement in the health care sector.
“This type of research and analysis will have a tremendous impact on improving the care process while managing and reducing the costs and improving the quality of care,” said Smith.
But Padman is quick to point out that this work is just the beginning. The problem-modeling and solution techniques that she and Zhang have proposed will require significant evaluation in the actual decision-making setting before they can be translated into practice. The ultimate goal is to use this information to create computer-based visualizations and automated tools that will help doctors and patients quickly and effectively decide on appropriate treatments in a shared decision-making, patient-centered, care delivery environment.
Not only that, but by embedding cost information into clinical pathways and exposing inconsistencies in chronic care delivery, Padman is optimistic that this research can pave the way for improved treatment options that will combine with innovative payment models.
“You want to be able to manage the patient holistically,” said Padman. She gives the example that if a patient has kidney disease and in managing that disease their corresponding diabetes, high blood pressure, and other co-morbidities are also managed together, there is a potential upside for both costs and health outcomes.
Zhang, who was recently given the prestigious title of Walsh McDermott Scholar in Public Health by Weill Cornell Medicine, added, “I am excited that this work has received considerable attention from both academic and industry collaborators, indicating its potential value. I look forward to implementing and evaluating this research in real patient care settings.”
Clinicians are already interacting via electronic health records, whether it is entering data for a current visit or to look at a patient’s history. Because of this, Padman suggests this innovative analytics approach would fit easily into current practice.
“If pathway recommendations or associated risk assessments are automatically displayed as a result of the data that’s being collected and entered into the electronic health records, that would be the seamless way to enable the delivery of this complex information for timely consumption,” said Padman. “The possibilities for application are endless.”
You can read Padman and Zhang’s publications on this topic by following the links below.
Y. Zhang, R. Padman, “Data-Driven Clinical and Cost Pathways for Chronic Care Delivery”, American Journal of Managed Care, special issue on Health Information Technology, 2016.
Y. Zhang, R. Padman, Innovations in Chronic Care Delivery Using Data-Driven Clinical Pathways, American Journal of Managed Care, special issue on Health Information Technology, 2015;21(12):e661-e668.
Y. Zhang, R. Padman, N. Patel, Paving the COWPath: Learning and Visualizing Clinical Pathways from Electronic Health Record Data, Journal of Biomedical Informatics, 58 (2015) 186–197.