A fascinating analysis of the trade-offs involved in different forms of incentivization appeared late this spring in the Health Affairs Blog, concerning prospective and retrospective financial incentives. Could the difference between the two make a real difference in how the payers and purchasers of healthcare stimulate the best behavioral outcomes from the providers of healthcare? The question is a very complex and nuanced one, and ultimately, the blog’s authors reach no firm conclusions either way. But pondering the different elements and possibilities is certainly a worthwhile exercise.
Under the headline, “When Designing Bundled Payments, Don’t Ignore the Lessons of Behavioral Economics,” Jeroen Struijs, Arthur Hayen, and Koen van der Swaluw, begin their analysis by stating that “In health care, all bundled payment models shift financial and clinical accountability to a single provider-led entity that then is responsible for a budget and the quality of care delivered for any given episode. The entity receiving the bundled payment earns a higher margin if a patient has fewer expenditures but also bears the financial risk of (re)admissions and complications. This payment model therefore produces an incentive for providers to coordinate care across settings and not to stint on needed care. When implementing bundled payment models, payers and providers can choose two main strategies regarding the payment flow, namely a prospective payment (which is made before services are rendered) or a retrospective payment (which is made after all services are rendered).”
Continuing on, they write, “Currently, there is ongoing debate about which payment strategy to follow. In a June 2017 Health Affairs Blog post, François de Brantes and Suzanne F. Delbanco stated that the difference between retrospective and prospective payments is not significant in practice. Paraphrasing their line of reasoning, the difference between the two is believed to vanish as retrospective payments are often based on upfront, negotiated benchmarks as well. The authors continue by saying that people’s “aversion to losses” would make a case for including downside risk to retrospective bundled payments, as a way of mimicking the incentives that follow from prospective bundled payments.”
But, these authors beg to differ on that point. “The universal lessons of behavioral economics have taught us otherwise, namely that people’s aversion to losses is precisely the reason why an upfront, negotiated benchmark is not the same as a prospective payment,” they write. “What is more, assuming downside risk in a retrospective payment arrangement will lead to different physician behaviors than when these physicians would assume full financial accountability for managing a prospective bundled payment. Understanding the difference in reference points under different payment strategies and their expected consequences for provider behavior is crucial in this discussion. As we will show, behavioral economics can be a useful paradigm for thinking about the behavioral implications of different payment strategies.”
For evidence to support their contentions, the authors dip into “prospect theory,” noting that prospect theory is “a descriptive model of decision making, which acknowledges, in contrast to conventional economics, that people respond to changes in wealth, instead of expected final assets. These changes are evaluated in relation to a neutral reference point (the status quo). Therefore, the same level of wealth can feel like poverty to a previously rich person and feel like riches to a previously poor person. In prospect theory, the status quo serves as the zero point of the value scale. Outcomes that are better than the status quo are perceived as gains. Worse than the reference point, they are losses.”
So how might prospect theory apply with regard to stimulating the “correct” behaviors and actions among providers, in bundled payment systems? Here’s an experiment with implications for healthcare payment. “Losses are more painful than forgone gains, and whether something feels like a loss or a gain depends on our point of reference,” the authors write. “As a consequence, people systematically assign higher value to things they own than to things they can acquire: the endowment effect. The primary effect of endowment is not the increased appeal of money in our pockets but merely the pain of giving it up. The endowment effect and accompanied loss aversion have been leveraged by behavioral economists to bolster the impact of incentives. For example, in one experiment, Mitesh S. Patel and colleagues paid a group of employees $1.40 each day they achieved the goal of walking 7,000 steps. Another group of employees was instead allocated upfront the maximum attainable monthly amount ($42) as a prospective payment; the researchers then took back $1.40 each day that the goal was not achieved. While the incentive was identical in economic terms, only participants in the loss-incentive group outperformed the control group and also had a 10 percent higher chance of achieving their daily step goal than participants in the gain-incentive group.”
Looking at the upsides and downsides of both types of systems, the authors write that, “While the financial risks may be equivalent for providers in a prospective or a retrospective model, we reason that these different responses ultimately will result in different strategies and polices to achieve value, for the good or the bad. For example, providers under prospective payment systems could adopt more risky strategies targeted at expected high-cost outliers or risk selection of patients. The ongoing debate about whether to implement retrospective or prospective payment models is a multifaceted issue, depending on many factors, but, conceptually, it comes down to a trade-off between implementation efforts on one hand (in favor of retrospective payments) and incentive strength (in favor of the prospective payment) on the other hand. The strength of an incentive and, consequently, the resulting value for patients, could be a function of the timing of the actual payment.”
It’s hard to know how this might play out in actual systems among physicians in practice, of course, given the high level of complexity of the processes around healthcare delivery and payment. I suspect that one very big barrier to implementing more aggressive prospective systems might developing the experience over time to know precisely how much to weight prospective incentive payments, and around what elements or factors. And that process would be complicated by the question of the extent to which individual physicians in practice would be aware of minute elements in care delivery and management processes, and be able to control or change decision-making in the moment.
And all of this presupposes some level of consistency of processes among the multidisciplinary care teams that are responsible for delivering and managing care under bundled payment-based contracts—and right now, there really is none.
And, not surprisingly, Francois de Brantes had something to say about all of this. In the comments section following the article’s text, he praises the authors for some elements of the blog. “The authors…bring up several important points related to the behavioral effects of prospective payments against which payments are debited, and there’s good evidence on those effects,” De Brantes wrote. “But,” he quickly adds, “the authors also make a central assertion that is far more debatable. They claim that the reference point in a model in which actual spend is reconciled against a target budget are the individual services paid. That’s untrue. The reference point is the target budget or price. Much like has been done in the Medicare Bundled Payment for Care Improvement (BPCI), these prices are negotiated up front and are the true reference point. Any claims incurred are simply debited against the target price and, once the episode of care closes, the net result is either a debit or a credit. If it’s a credit, then the provider is owed money. If it’s a debit, the provider owes money.”
And that’s exactly where the problem around this is, de Brantes notes. “As such, an episode of care payment model with downside risk and a prospectively set target price is an endowment of sorts even though it technically doesn’t function as such since the provider doesn’t simply get to keep the difference between the endowment and the total cost of the services rendered,” he notes. “And there are several reasons why such a model is preferable. First, as the authors point out, a true endowment model could lead providers to cherry-pick patients or skimp on needed services. Second, there are many operational issues that provide a counterweight for the potentially greater effect of a true endowment. Third, field experience has shown that when providers receive timely and frequent (e.g. once a quarter and within 45 days of the end of a quarter) reports showing, by patient, the total actual costs incurred against the target prices for the episodes, the behavioral effect is very similar, if not identical. Fourth, back end reconciliations also allow for some flexibility in how to account for quality scores, for example by mitigating losses when quality is higher than expected or vice versa.”
In short, de Brantes says, “We agree with the authors that, at some point, a natural experiment that would compare the effects over time of the two models would help clarify the differential effect of one over the other. The evidence to-date, however, doesn’t seem to indicate that there would be much.”
In short, we are very early along in this journey, and evidence based on practical experience, of which types of levers work best in bundled payments, remains lacking. It will be fascinating to see what the evidence of the next couple of years offers us. Until then, everything remains fundamentally experimental in nature.