Understanding The Latest ACO “Savings”: Curb Your Enthusiasm And Sharpen Your Pencils—Part 2

Editor’s note: This post was originally published on Health Affairs Blog on 11/13/2020. I

In part 1 of this two-part blog post, we provided a rigorous and nuanced interpretation of Medicare Shared Savings Program (MSSP) performance data and described problems with the program’s design. In this post, we lay out key considerations for improving the MSSP over the long haul. Our objective is to help get a repeatedly derailed conversation back on track.

Reforming The MSSP—The Conversation We Should Be Having

To achieve the MSSP’s long-term goals as the centerpiece of payment reform in the traditional Medicare program, significant design changes are needed. Below, we discuss those we believe will be most decisive in achieving success. We assume that the program will remain voluntary but note that the distinction between voluntary and mandatory is not as dichotomous as the debate over alternative payment models might suggest. Short of a mandate, participation in the MSSP becomes de facto mandatory when the fee-for-service alternative becomes sufficiently less attractive. Accordingly, our recommendations consider effects on participation among the factors CMS must weigh as it walks the regulatory tightrope strung by the MSSP; we also assume that striking the right balance in tradeoffs between savings and participation incentives should get easier with each step, as fee-for-service becomes increasingly less appealing.

1) Eliminate rebasing based on historical spending. To create meaningful incentives for accountable care organizations (ACOs) to save, the link between an ACO’s benchmark and its prior savings must be severed. One concern about such a policy when initial benchmarks are based on ACOs’ historical spending is that it could permit successive windfalls for ACOs with initially high spending by perpetuating wide differences in benchmarks, thereby advantaging providers with initially wasteful practices over their more efficient competitors. This concern, however, can be addressed without undermining incentives to save and participate. Benchmarks based on historical spending can be slowly blended with regional or national spending averages, or benchmark growth can be set administratively at a slower rate for initially high-spending ACOs than low-spending ACOs (for example, gross domestic product growth -1 percent versus +1 percent), thereby gradually converging benchmarks toward a common target. The key to preserving incentives is to ensure that an ACO’s own efforts to lower spending do not result in a lower benchmark. Currently, the MSSP violates this principle by periodically rebasing the historical component of an ACO’s benchmark to its most recent spending level.

2) Increase shared-savings rates. Although lower shared-savings rates allow Medicare to capture more of the amount saved, total savings are likely larger when the shared-savings rate is higher. Even if the shared-savings rate is 100 percent, Medicare benefits indirectly through spillovers, including savings from more efficient treatment of non-ACO patients and reductions in Medicare Advantage (MA) payment rates (which are based on fee-for-service spending). Also, if benchmark updates are based on regional or national fee-for-service spending growth, benchmark growth slows as ACOs achieve savings, thus producing eventual returns to Medicare as the actual (real) savings become larger than the benchmark-based savings given back to ACOs. We advocate for higher shared-savings rates in general and particularly high savings rates for ACOs with initially lower spending to account for the higher costs of reducing wasteful spending when there is less waste. For example, shared-savings rates could range from 65 percent to 85 percent depending on initial spending levels, reaching closer to 85 percent as high-spending ACOs lower spending.

Reticence to increase shared-savings rates stems from the calculus under the status quo. Even without actual savings or losses caused by the MSSP, there is always random variation of ACO spending around historical benchmarks that puts Medicare at risk of net losses when actual savings are minimal and shared-savings rates are greater than shared-loss rates. Low shared-savings rates limit these losses by limiting bonuses for “savings” generated by chance. But we believe it is self-defeating to design a program to not generate savings out of fear of it not generating savings. If we are to realize the gains from payment reform, we must reform payment.

3) Limit downside-risk requirements for now. Although the independent effects of downside risk on participation and selective exit are unclear, at this juncture we believe that limiting downside risk, and perhaps eliminating it for certain types of ACOs, would be prudent. With higher shared-savings rates and a plan for benchmarking without historical rebasing, downside risk can be gradually and progressively introduced without costly consequences for participation, particularly as fee growth is slowed. Downside risk may be important to elicit savings from large health-system ACOs, as they have particularly weak incentives in one-sided contracts due to fewer opportunities to lower wasteful spending without providing less care or less expensive care (and thus reducing fee-for-service profits).

4) Establish a long-term vision and plan. Conceiving an initial phase for the MSSP is relatively straightforward: base benchmarks on historical spending, update them at a preset or concurrent regional or national growth rate, do not rebase them based on performance, and allow ACOs to keep a substantial share of savings. We advocate for such an initial phase to last at least five years. This encourages participation by providers with relatively high spending and gives them strong incentives to lower spending, thereby fostering convergence in spending that in turn facilitates eventual implementation of a steady-state model that permanently replaces the traditional fee-for-service payment system for all providers. The design of a steady-state model, however, has not been specified. One vision for the MSSP is that there is no steady-state model. Rather, the initial model (which is inherently an improvement model) could be made permanently available. As this model is most attractive to providers with high spending, this approach would focus only on tamping down fee-for-service spending when and where it gets high—a type of whack-a-mole model. Another, more ambitious, vision for the MSSP is a direct contracting version of MA, with benchmarks that are risk-adjusted functions of a regional or national payment rate. This is the vision that most stakeholders have in mind for the MSSP.

Such a steady-state model is also relatively straightforward to conceive, although some key decisions need to be made and challenges overcome. One key matter to be decided is the basis for setting and growing a regional or national benchmark rate. A rate that is based on fee-for-service spending helps account for the costs of new technologies and introduces an element of yardstick competition into the payment model, whereby an ACO’s profits depend on its performance relative to other providers. A rate that is instead set administratively forces conversations about desirable health care spending levels and growth rates, avails a more direct mechanism for Medicare to control spending, and allows for ACO benchmark growth to be set above fee growth, thus making the MSSP a more viable option for providers. The major challenge facing a steady-state model is to develop adequate methods of risk adjustment (discussed below).

The transition phase is trickier. The MSSP must converge benchmarks in a way that expands the program and encourages progress toward a steady-state model. This will require monitoring and regulatory flexibility. Over its eight-year history, the MSSP has had an initial phase and embarked on a transition phase, the design and pace of which have halted progress. A reboot may be necessary.

Defining the long-term vision is vital to informing decisions by providers and policy makers. Providers with high spending cannot consider measures to reduce wasteful spending without knowing future returns. They may be more willing to incur short-term losses in exchange for larger long-term gains, but only if they can anticipate those gains. From a policy standpoint, a model offering subsidies in the short term (for example, from selective participation) should be tolerated if it produces long-run savings that outweigh its short-run costs. Similarly, a long-term vision puts short-term savings for Medicare in perspective, revealing tradeoffs between clawing back dollars immediately and ultimately positioning the program to control spending growth.

5) Simplify the Medicare payment model portfolio. Along with a long-term plan, the MSSP needs to be situated in a coherent broader payment reform agenda. The proliferation of alternative payment models is dizzying and has proceeded without much regard for interactions. The resulting overlap creates conflicting incentives, and the lack of coordination distorts incentives to participate, thus allowing providers to choose the model that best suits them under the status quo. The complexity has also increased the administrative costs of payment reform, spawning a cottage industry of consulting services marketed to providers scrambling to comprehend, choose, and comply. A population-based payment model, such as the ACO model, should serve as the centerpiece of a reformed payment system in the traditional Medicare program. Additional models should be conceived only to address limitations of that core model.

6) Improve risk adjustment. Risk adjustment plays a critical role in prospective payment systems. Evidence suggests that there was minimal systematic patient-level risk selection by ACOs in the first three years of the MSSP. But as the program transitions away from ACOs’ historical spending as the basis for benchmarks, ACOs have new opportunities to gain from favorable risk selection. Specifically, they develop incentives to make participation decisions at an ACO or practice level based on patient risk factors that are not adjusted for (but would be reflected in ACO practices’ historical spending). As benchmarks converge to a common basis, the program must rely more heavily on risk adjustment to ensure fair allocation of resources to providers serving relatively high- or low-risk patients.

There are many opportunities to improve prediction in risk adjustment, both through statistical techniques and additional sources of data on patient characteristics, such as survey data supplied by the ACO Consumer Assessment of Healthcare Providers and Systems (CAHPS) survey. But risk adjustment will always fall short; therefore, additional measures to mitigate incentives for risk selection and unintended consequences for vulnerable populations are needed. By limiting ACO risk for the high tail of the spending distribution, incentives for ACOs to avoid high-cost patients can be weakened. Currently the MSSP truncates per-beneficiary spending at the 99th percentile in calculating benchmarks and savings. We recommend lowering the threshold to the 95th percentile. This change may have additional salutary effects by shifting organizations’ focus away from high-risk case management to systemic changes in care delivery. With risk selection incentives sufficiently minimized, disproportionate participation of lower-spending providers should reflect selection based on efficiency rather than patient risk. Whereas risk selection is wasteful, ACO selection of efficient practices—although costly in the short term—may be beneficial over time in a growing MSSP by exerting marketwide pressure to be more efficient.

Even when risk adjustment is sufficiently predictive of spending to discourage risk selection, risk-adjusted payments may entrench patterns of insufficient spending for groups with unmet needs. Use of observed spending data to improve model predictions presumes incorrectly that current spending is efficiently and fairly distributed across disease groups, geographies, and individuals. In new payment models, additional adjustments will be needed to protect against underspending for populations with high levels of unmet need; these adjustments can be implemented in various ways and inevitably rely on normative judgments about desirable levels of spending. Conversations about risk adjustment in population-based payment models need to consider the limitations of conventional approaches and address social goals such as reducing health care disparities.

7) Limit pay-for-performance. The utility of the pay-for-performance component of the MSSP is unclear. To those who believe that improving health outcomes is the key to savings, the quality-linked incentives are complementary. Improving health through better care coordination and disease management, however, may not be a conceptually sound strategy for achieving major savings and has not been a primary driver of savings in the MSSP. Moreover, if improving outcomes were the path to savings, a population-based payment model with strong incentives to save would not need quality-linked incentives to elicit better outcomes. To others, the purpose of the quality measures in the MSSP is to prevent stinting on valuable care. However, it is not clear that we should expect stinting on net in a population-based payment model. An underappreciated potential benefit of such a model is that removal of interfering fee-for-service incentives may facilitate improvement in quality and outcomes. As decoupling revenue from service provision gives providers more flexibility in selecting care inputs, their intrinsic motivation to maximize patient health and care experiences may be more likely to prevail. Put another way, stinting on unreimbursed activities may be rampant under fee-for-service payments that divert providers’ attention to reimbursed activities.

At the very least, population-based payment models better position providers to enhance quality of care by more effectively harnessing professionalism. Including a complex pay-for-performance component in these models may limit such gains by substituting one set of distracting incentives for another. Finally, there is no reason why the inclusion of measure-linked penalties and rewards in the MSSP should be immune to other unintended consequences demonstrated in stand-alone pay-for-performance, such as wasteful gaming behaviors and inequitable financial transfers that threaten to exacerbate disparities.

Evaluating Success Moving Forward

Staggered implementation of MSSP reforms across randomly selected geographic areas would support evaluation of true savings from program changes in the short term. Ultimately, however, the long-term savings cannot be estimated although quasi-experimental evaluation, for the reasons summarized in part 1 of this post. Although dissatisfying, lack of evaluation does not negate the ability to judge program success. If a goal of the program is to control spending growth and if the MSSP gives Medicare a mechanism to set spending growth to desirable rates, the program will be a success; it would fulfill the task at hand, even if precise savings remain unknown.


At this stage of Medicare payment reform, progress is not served by applauding illusory savings or overselling simple solutions such as downside risk. Payment reform in Medicare needs reform. Conversations need to happen (and continue). Slapdash analysis and slogans should not be tolerated. Instead, sound evidence and theory should inform vision and set a course. Resolve will be a must and patience a virtue. Guided by these principles, designing a payment system that improves the efficiency of care delivery, meets fiscal goals, and makes patients better off should be doable. We hope that our analysis and recommendations serve to promote productive dialogue and advance successful change.