It’s time to take a fresh look at how the Centers for Medicare and Medicaid Services (CMS) designs its initiatives to test new models of provider payment and care delivery. As highlighted recently in the Health Affairs Blog by Tim Gronniger and colleagues, the new administration faces important choices about imposing requirements that support more rigorous and informative evaluations of new models on providers.

With the recent widespread implementation of alternative payment models (APMs), strong designs are needed more than ever to provide evidence for policy decisions about model expansion, modification, or termination for Center for Medicare and Medicaid Innovation (Innovation Center) initiatives. However, policy decisions to pursue designs with mandatory participation or random assignment can be difficult when providers resist participating in studies for which their requirements, financial incentives, and risks are not fully known in advance. The good news is that the tradeoffs between accommodating provider interests and CMS’ ability to identify worthy innovations using strong research designs may not be as stark as some have previously assumed.

In a Health Services Research article recently made available online, we argue that designs of CMS payment initiatives must effectively accommodate the changing payment and delivery system environment. Accordingly, we advocate for use of factorial experiments (randomized designs that test multiple versions of a model simultaneously) as the best prospect for producing definitive evidence on future APMs. This approach stands in marked contrast to that of William Shrank, Robert Saunders, and Mark McClellan, who have endorsed continuing to base policy decisions on a mix of quantitative analysis comparing similar populations, qualitative analysis of other populations, and other contextual evidence. They largely dismiss randomized designs for CMS, saying “momentum and timelines would be lost with too much focus on experimental design and … traditional rigorous evaluation methods.” Both papers share the objectives of improved evidence and accelerated learning, but we reach different conclusions about what methods will produce the best and quickest evidence to guide CMS policy in the years ahead.

Widespread Use Of APMs Creates Problems For Traditional Research Designs

At the heart of our argument is the anticipation that the widespread and growing use of APMs will soon make it virtually impossible to find a credible comparison group unaffected by any of the current payment reform initiatives. This development will make it very difficult to reliably distinguish the overall effects of a new APM from those of other contemporaneous initiatives with similar objectives. CMS can, nonetheless, produce reliable estimates of the causal effects of each of the separate APM features and incentives by using factorial experimental designs. These designs can be implemented on a voluntary or mandatory basis. Providers would be randomized to one or another variant of alternative payment—based, for example, on combinations of the size of rewards for quality, the share of savings allotted to the provider, and the degree of financial risk borne by the provider.

Randomized Factorial Designs Can Assess APMs

The essence of the Shrank and colleagues argument is that few providers will be willing to apply to participate in a demonstration if they do not know in advance their exact risks and rewards—how much they will receive and what requirements they must meet to earn rewards. They also suggest that conducting randomized trials is difficult and costly. Our suggested method of randomizing providers to different combinations of program incentives and requirements does not require a control group that does not receive incentives for participation. This approach could be designed to provide sufficient incentives for voluntary participation in all tested variants or be pursued on a mandatory basis (for example, to test acceptably small changes in payment methods without specific requirements for altering care delivery).

Factorial designs would enable simultaneous learning about effects of multiple incentives and model requirements. These designs have advantages over recent CMS randomized designs such as those of the Million Hearts Cardiovascular Disease Risk Reduction Model or the Medicare Care Choices Model as well as those that test a single form of a model. Yet the factorial approach would differ little in cost or complexity from current Innovation Center evaluations, which are based on quasi-experimental designs. Continued reliance on traditional designs would come with increasing risks to well-informed policy, as available non-APM comparison groups become less representative of what would have occurred without the intervention. Factorial experiments can solve this problem while simultaneously testing multiple payment model variations. Thus, they offer a surer and faster path to definitive learning about APMs. The focus shifts from “Does this specific APM with pre-specified parameters work?” to “How should an APM be designed to maximize its effects?”

Factorial Designs Can Accelerate Medicare Innovation

As Medicare payment policy evolves over the next decade, decisions on research and evaluation designs will determine the quality of the evidence used to reshape our health care delivery system. Some of the strongest designs for producing necessary evidence might include factorial experimental methods. These designs can be implemented in demonstrations or initiatives regardless of whether provider participation is mandatory or voluntary. It is not too early to begin planning for the Innovation Center’s next round of initiatives and demonstrations. Careful attention to using the best designs will help the Innovation Center initiatives yield timelier, more useful, and stronger evidence for CMS as it seeks to improve health care quality and lower costs in the years ahead.