Identifying Spending Patterns Using Data-Driven Approaches

^{By Julie Gould}

A recent study in JAMA Network Open shows that the use of a data-driven approach helped identify spending patterns with high accuracy.

“Current approaches to predicting health care costs generally rely on a single composite value of spending and focus on short time horizons,” the researchers of the study explained. “By contrast, examining patients’ spending patterns using dynamic measures applied over longer periods may better identify patients with different spending and help target interventions to those with the greatest need.” 

To better understand the study design, we spoke to Julie C. Lauffenburger, PharmD, PhD, Brigham and Women’s Hospital. She highlights how health care organizations may find this application of trajectory modeling beneficial in their populations to identify groups of patients with similar spending and isolate those patients who may have progressively rising costs. 

What existing data led you and your co-investigators to conduct this research? 

Most prior studies of costs focus on total costs or spending outcomes over the course of one year, which does not necessary reflect real-world patterns of spending which can change over time. We had previously conducted some pilot work in commercial claims data over the course of 1 year, where we used trajectory modeling to identify patterns of spending and predict membership in those patterns. We expanded this work subsequently to Medicare patients and were particularly motivated to see how these might apply to longer-term patterns of spending, as the ability to identify and predict patients with progressively rising spending, even a year out, might be more amenable to intervention. 

Please briefly describe your study and its findings. Were any of the outcomes particularly surprising?

We used trajectory modeling and machine learning to cluster patients by their 2-year spending patterns and predict membership in these patterns using baseline characteristics, including potentially modifiable characteristics, measureable in Medicare claims data. Health care spending patterns over 2 years were best described by: minimal users (11% of pop), low- (15%), moderate-(25%), and high-cost (41%) patients, and patients w/ progressively rising costs (8%). These groups could be predicted well using boosted regression. Using fewer medications, having fewer office visits, seeing more physicians, and using tobacco were potentially-modifiable baseline predictors of being in the rising-cost group versus those with similar baseline spending. Adherence to medication was also an influential predictor. Adherence to medication is one of my core research areas, where I study the extent of non-adherence and develop interventions to try to address non-adherence; that said, I was surprised at how influential that particularly predictor was on membership in trajectory groups, most notably this rising-cost group. 

What are the possible real-world applications of these findings in clinical practice? 

First, health care organizations may find the application of trajectory modeling beneficial in their populations to identify groups of patients with similar spending and isolate those patients who may have progressively rising costs and intervene upon them. Second, these potentially-modifiable predictors may represent important levers for interventions to prevent escalating costs, such as coordination of care and ensuring access to necessary medications. These should be greater points of emphasis for interventions.

Do you and your co-investigators intend to expand upon this research? 

Definitely, we are considering expanding this work into different clinical areas and applying these methods to intervention development specifically. 

About Dr Lauffenburger:

Julie C. Lauffenburger, PharmD, PhD is an Assistant Professor at Harvard Medical School and Division of Pharmacoepidemiology and Pharmacoeconomics at the Brigham and Women’s Hospital. She is also the Assistant Director of the Center for Health Care Delivery Sciences and a practicing pharmacist at Brigham and Women’s Hospital. Her research focuses on improving medication use and health care delivery for patients with common chronic diseases, particularly through the design and conduct of pragmatic trials, studies using predictive analytics, and evaluation of interventions. 

Reference:

Lauffenburger JC, Mahesri M, Choudhry NK. Use of Data-Driven Methods to Predict Long-term Patterns of Health Care Spending for Medicare Patients. JAMA Netw Open. 2020;3(10):e2020291. Published 2020 Oct 1. doi:10.1001/jamanetworkopen.2020.20291