Improving the Performance of Health Plan Payment Systems
Sherri Rose illustrates ways to improve payments to health-care plans, making them more efficiently and fairly distributed.
American taxpayers lose billions of dollars each year to Medicare and Medicaid fraud. The formulas used to pay health-care systems from Medicare and the ACA marketplaces are also inefficient and can further marginalize patients with many health conditions.
In the end, the federal programs, taxpayers and patients all lose.
New research by SHP’s Sherri Rose, PhD, illustrates ways to improve payments to health-care plans, making them more efficiently and fairly distributed.
“By changing how we design and evaluate the formulas — broadening how we define what improving the performance is — we can prioritize marginalized groups in the health care system, such as those with mental health and substance use disorders,” said Rose, an associate professor of medicine at Stanford Health Policy and co-director of the Health Policy Data Science Lab.
Her paper, recently published in the American Journal of Health Economics, lays out potential payment formulas that could reduce demands on medical data while also curbing the risks of upcoding — purposely using additional billing codes to increase reimbursement — and ordering unnecessary services. By improving performance at the group level, Rose said, “we can increase access to care by reducing incentives for insurers to discriminate against groups currently undercompensated in the risk adjustment system.”
Three Design Elements
Rose, a statistician, and her co-authors lay out ways to simplify payment formulas with the introduction of three elements in plan payment design: reinsurance based on high spending or plan losses, constrained regressions, and powerful machine learning methods for selecting which variables should be used in payment formulas.
“We illustrate their power by showing, in the case of the complex payment formula applied in the Marketplaces, that using these novel tools permits radical reduction in the number of variables used in the risk adjustment formula, while maintaining or improving performance in terms of fit,” Rose and her co-authors write, where fit refers to how well the formula represents the data. “The best mix of these three tools is likely to depend on the empirical and policy context.”
In one example, the authors used constrained regression to improve fairness for four currently undercompensated chronic health conditions, balancing performance for the overall population with performance for these groups in the formula’s optimization process. This was in addition to deploying budget neutral reinsurance for high-cost enrollees and reducing the number of variables in the formula—dropping all prescription drug variables and using data-adaptive machine learning to exclude further variables.
The results went beyond what was anticipated. Not only did overall fit metrics and group fit metrics demonstrate improvements for the four chronic conditions — cancer, diabetes, heart disease, and mental health disorders — but 88% of other undercompensated groups studied also had improved fairness measures. Thus, they found that improving fairness for some groups can improve fairness for other groups with no meaningful loss in the formula’s performance.
The paper co-authors are Thomas McGuire, a professor of health economics at the Department of Health Care Policy, Harvard Medical School, and Anna L. Zink, a graduate student in the PhD program in health policy at Harvard University. Their research was supported by a grant from the Laura and John Arnold Foundation and Rose’s NIH Director’s New Innovator Award.
The authors note that risk assessment formulas and health plan payments in the last 30 years have done much to improve the fit of payments to plan costs at the individual and group level.
“The magnitude of reduction in undercompensating for persons with cancer falling from $2,000 per year to $1,000 per year, for example, can be readily understood and appreciated,” they write. “At the same time, researchers and policymakers recognize that gains in fit come at a cost in terms of data demands, administrative complexity, potential opportunities for gaming the system by ‘upcoding’ … and inducing provision of extra services to achieve higher risk scores.”
An Unconventional Approach
Their paper takes an unconventional approach to improving the performance of health plan payment systems. “Rather than treating fit as an objective, we treat the level of fit in the existing payment system, at both the individual and group levels, as a constraint,” they write.
“The main purpose of our paper is to introduce a counterpoint to the conventional wisdom that improvements in plan payment models should be pursued by adding variables,” the authors write. “Our starting point is the observation that improving fit by the addition of risk adjustor variables has hit the ‘flat of the curve’ in many countries and sectors.”
The researchers argue that the converse is also true, that removing variables from formulas does little to reduce fit.
The current formula has over 100 different health condition variables, however, Rose and her co-authors show that reducing that number by more than half has a trivial impact on overall fit and no impact on group fairness.
“We show that considerably more parsimonious risk adjustment models, when paired with improved estimation methods and small amounts of high-cost risk sharing, can attain fit as good as or in many cases better than the current system.”
This work then paves the way for reducing the administrative demands of the health plan payment system while decreasing incentives for upcoding and discrimination.