Machine-learning expert recommendations can influence Medicare Part D choice
Americans know that choosing a health insurance plan can tough. And once you’re retired and possibly on a limited or fixed income, it can become downright brutal.
Stanford Health Policy’s M. Kate Bundorf and Maria Polyakova and their colleagues set out to develop an online decision-support tool to test whether machine-based expert recommendations would influence choice among Medicare Part D enrollees — and make it easier.
“The use of technology seems like a natural way to address the challenges of choosing among plans,” they write in their study published in Health Affairs.
Medicare beneficiaries have been choosing among Medicare Advantage and Part D prescription drug plans for years, and more recently the Affordable Care Act established health insurance marketplaces for those who are younger than 65.
All that choice is supposed to create incentives for plans to offer a variety of low-cost, high-quality products that allow people to choose the plan that best meets their needs.
But sometimes too many good choices can lead to bad outcomes.
“Health insurance is a complex financial product with complicated cost-sharing rules, and the implications of different benefit designs for out-of-pocket spending and health care use vary across consumers depending on their needs,” wrote Bundorf, chief of the Department of Health Research and Policy and an associate professor of medicine at Stanford Medicine.
Another researcher in the study was Albert Chan, chief of digital patient experience and an investigator at Sutter Health, in Palo Alto, as well as an adjunct professor at the Stanford Center for Biomedical Informatics Research. Ming Tai-Seale, a professor of family medicine and public health at University of California San Diego, was also a principal investigator of the study.
Choosing Health Plan is Complicated
“Consistent with these challenges, researchers have documented that many consumers, both young and old, do not understand the characteristics of their plans,” they wrote in the March issue of Health Affairs, which is holding a public briefing on patients-as-consumers at the National Press Club on March 5th. Bundorf will present their research at the briefing in Washington, D.C., which will be streamed live and will be posted here once it has aired.
“(Patients) often make decisions that may signal inaccurate evaluation of the costs and benefits of coverage — such as staying in their plan when better options are available, not enrolling in the plan that provides the best coverage for their drugs, or enrolling in plans that are objectively inferior to other available choices,” the authors wrote.
The Centers for Medicare and Medicaid Services (CMS) offers a tool to help beneficiaries choose among plans, but older adults — even those with high levels of formal education — find it difficult to use.
So, the research team developed a decision-support software tool called CHOICE to assist Medicare beneficiaries in choosing a Part D prescription plan. The software automatically imported the user’s list of current drugs from their electronic medical records (allowing users to adjust the list if desired); the algorithm would then crunch the numbers to come up with three recommended plans which were likely to be the least expensive for the user.
The team then conducted a randomized trial of this software tool among 1,185 patients of the Palo Alto Medical Foundation (PAMF), a large health-care provider in Northern California. Fifty-four percent of those patients were women, 65 percent were white, and 54 percent were married. Living in the Bay Area, their income and education levels were fairly high: They lived in areas in which the median income is $106,808 and 54 percent of the population has a college degree or more education.
While not representative of the general population of seniors in the United States, the researchers emphasized that it was important to conduct this study among these potential users, who are more likely to respond positively to an interaction with a computer. If these users didn’t find this software helpful or user friendly, it would not likely be a useful tool to roll out across the country as a whole.
The study participants received access to one of two versions of the CHOICE tool: expert recommendations or individual analysis. Both versions automatically imported information on patients’ prescription drugs from their electronic health records and combined it with information on plan benefit design to provide individually customized information on users’ likely spending on both premiums and prescription drugs in each of the stand-alone Part D plans available in their area. The version of CHOICE that offered expert recommendations combined this information with an explicit recommendation on which plans were best for the user.
Willing and Able
The researchers found that providing an online tool not only increased older adults’ satisfaction with the process of choosing a prescription drug plan, but they also spent more time choosing that plan.
“The most significant finding of our trial is that individually customized information alone didn’t seem to be enough,” Bundorf, who is also a senior fellow at the Stanford Institute for Economic Policy Research (SIEPR), said in an interview. “The tool we developed was most effective when individually customized information paired with a clear-cut algorithmic expert recommendation that highlighted three plans that the computer thought were the best for the user based on total spending for prescription drugs.”
She said she was surprised to see that people spent more time choosing a plan and were more satisfied with the process when they had access to the CHOICE tool.
“Prior to our trial, I thought people might spend less time choosing a plan when they had access to expert recommendations because it would make the process easier,” Bundorf said. “But taken together, these results suggest that people are more engaged in decision-making when they have access to a patient-centered tool.”
Polyakova, who is also a faculty fellow at SIEPR, said a key takeaway from the trial is that people who are likely to use sophisticated tools are already more likely be more sophisticated shoppers of health care and prescription plans.