Using the lessons of behavioral economics to design more effective pay-for-performance programs

Abstract

Objective

Although pay-for-performance (P4P) incentives are increasingly popular, the literature on health care has found that these incentives have had minimal impact. We believe a key reason for this finding is that current P4P programs are poorly designed and do not reflect what is known about the psychology of how people respond to incentives.

Methods

Using lessons from behavioral economics, we describe seven ways to improve program design in terms of frequency and types of incentive payments. After discussing why P4P incentives often have unintended adverse consequences, we outline potential ways to mitigate these consequences.

Conclusions

The lessons from behavioral economics can greatly enhance the design and effectiveness of P4P programs in healthcare, but future work is needed to demonstrate this empirically.

The use of pay-for-performance (P4P) incentives in health care is widespread. In the United States, P4P incentives are utilized by half of all commercial health maintenance organizations (HMOs), and are found in contracts with ambulatory physicians, hospitals, and nursing homes.^1-5 Numerous state Medicaid programs also use P4P incentives and the proposed Medicare hospital P4P program could include incentives totaling more than $3 billion annually.⁶ In the UK, almost 25% of family practitioner income is tied to P4P incentives.⁷ Despite the widespread application of P4P, most of the published literature on the impact of P4P has concluded that these incentives have resulted in small or no improvements.^{2, 3, 7}

There have been various interpretations of these results. Some have raised concern that the premise underlying P4P is flawed.⁸ Others believe the magnitude of the incentives have been insufficient.⁹ Another potential reason is that the current design of P4P programs does not reflect the psychology of how people respond to incentives. This is not surprising as there has been little literature on the effectiveness of specific design features of a P4P program to guide health plans or government sponsors.^{10, 11} Currently, program sponsors do what seems reasonable and there is great variation in the design of programs.¹² The behavioral economics literature can serve as a useful guide on how to structure provider incentives.

In this paper we discuss a number of design alternatives drawn from the behavioral economics literature that we believe could lead to greater provider response for the same dollar amount devoted to a P4P program. We start by describing the importance of design features and a prototypical P4P program being used today. We then discuss a number design features that could improve this program. Lastly, we discuss some potential unintended consequences of P4P and how design changes could minimize these unintended consequences.

What is the goal of P4P and why do design features matter?

The primary goal of most P4P programs is improving health care quality, but incentives have been applied for other goals including improving patient experience, implementing electronic prescribing, increasing patient safety, or decreasing utilization.¹² In this paper we focus on an illustrative physician P4P program that focuses on health care quality, but we believe our recommendations can extend to other care settings, types of providers, and other goals.

Most evaluations of P4P programs have measured change in performance on quality metrics.^{2, 3} But in designing a P4P program one has to consider the more proximal goal. To improve quality, a P4P program has to change the behavior of physicians and more specifically increase the time and resources they allocate to quality improvement. The goal is to make the desired behavior (e.g. taking the time to speak to patients about a mammogram) front and center in the physician’s mind during a busy day.

We recognize that quality improvement, cost reduction, or any other goal of a P4P program often requires more than physician behavioral change including system changes, such as implementing an electronic medical record. Keeping physicians engaged is still critical, however. It is notable that when P4P programs target physician groups, the physician groups often create P4P programs internally for their individual physicians.¹³

Current Pay-for-Performance Programs

To make our recommendations more concrete we begin by describing a prototypical physician P4P program which is designed to increase the number of women who receive a mammogram. After the claims from the previous calendar year have been processed, the incentive is paid out in the following steps. The health plan (1) determines the number of women who should have received a mammogram and how many did and (2) determines which physician is responsible for each patient’s care and calculates a physician-specific mammogram rate.

There is a significant heterogeneity in how health plans structure some aspects of their P4P programs.¹² Some health plans give incentives to physicians who meet a relative threshold (e.g. the top 25% of physicians in terms of mammogram rate) and others use an absolute threshold (e.g. physicians with a mammogram rate >75%). These top physicians can receive their incentive in a variety of ways. Most commonly they will receive an increase in their reimbursement for each visit for the following year (e.g. $106 vs. $100) or a lump sum incentive payment at the end of the year (e.g. $1000).¹²

We now discuss a series of potential design changes to these commonly used programs (listed in Table 1). We recognize that some of the design changes we recommend conflict with each other. We see them as a menu of options to be considered and are not meant to be applied all together.

Table 1.

Suggested Design Improvements to Increase Impact of P4P Programs

Commonly currently used design	Suggested Improvement
Incentive given as a lump sum	Divide the lump sum into a series of smaller incentive payments
Relative thresholds (e.g. top 25% of physicians)	Tiered absolute thresholds (e.g. 25%, 50%, 75%, 90%)
Long lag time between care and receipt of incentive	Shorten lag time to as short as possible
Use of withhold payments	Bonus payment or use of deposit contracts
Complex uncertain structure of program (e.g. shared savings program)	Simplify program so that uncertainty minimized
Incentive often given as an increase in fee schedule reimbursement	Decouple incentive payment so that it is given separately, consider a lottery
Dollar incentives	“In kind” incentives

Design Change #1: A series of small incentives are better than one large incentive

Why do people go across town to save $10 on a clock radio but not to save $10 on a large-screen TV?(cite) In both cases, $10 is saved, but $10 is not always viewed the same. Research has shown that an individual perceives the difference between $0 and $10 as being greater than the difference between $100 and $110. Similarly, ten payments of $10 may be more motivating than a single $100 payment.

For P4P program design, it may be more psychologically motivating to provide a physician with smaller, more-frequent incentive payments than a larger, single lump-sum incentive payment. As an example, consider that a total of $1,000 is available to give in incentives to the top physician performers. Applying this principle, a physician’s behavioral response is likely to be greater if the $1,000 is divided into a number of payments—say, one hundred payments of $10 each—rather than paid as a single payment. Each $10 is perceived as a new $10 gain.¹⁴

We recognize that a reward program with frequent payments is administratively more difficult. But the more frequent incentive can be symbolic and still be effective.¹⁵ For example, every time a physician’s patient receives a mammogram an e-mail is sent, “Your patient Edith Jones received a mammogram on this date. We will credit you with $10 at the end of the quarter.” The combination of this frequent symbolic reward as well as a larger separate check at the end of the quarter might be doubly satisfying as the incentive is reinforced.

Design Feature #2: A series of tiered absolute thresholds is better than one absolute threshold

An individual’s motivation and effort when faced with a goal greatly depend on that individual’s baseline performance. Economists and psychologists have described this phenomenon as a “goal gradient.”¹⁶ If baseline performance is far away from goal performance, the individual exerts little effort, because the goal is viewed as not immediately attainable. As baseline performance gets closer and closer to goal performance, the individual exerts more and more effort to reach the goal (e.g. 75% mammogram rate). However, as soon as the goal is achieved, the motivation to improve decreases significantly.¹⁷ A simple illustration of this phenomenon is a study of a coffee shop reward program in which the tenth coffee purchased was free. Participants in this experiment decreased the time between purchases of a coffee as they got closer to the free coffee.¹⁸

Goal gradient theory has several applications in a physician P4P program. In aggregate, there is likely to be a greater behavioral response if there were a series of quality performance thresholds to meet (e.g., increasing dollar amounts for achieving a 50 percent, a 60 percent, a 70 percent, an 80 percent, and a 90 percent performance threshold) rather than one (e.g., a 75 percent performance threshold). In a single-threshold system, physicians who at baseline have both either low performance (e.g. 25%) and high performance (e.g. 80%) have little reason to devote more resources to try and improve quality.

Some have proposed eliminating thresholds entirely and using a continuous gradient (e.g. physician receives $1000 × 76% performance = $760). Our own opinion is that such a continuous gradient may be less effective than a series of thresholds, because there is some benefit in having the clear “bright line” goal of a threshold. However, this needs to be proven empirically.

Design Change #3: Reducing the time lags between care and receipt of incentive increases the behavioral response

Money received right away is perceived as different in value from money to be received in the future—even the near future.¹⁹ This steep initial discounting is much greater than would be expected by “rational” economic discounting and has been termed hyperbolic discounting.¹⁹ In a typical P4P program, the time required to collect and validate the data, create physician scores, and make the payout often means that the incentive payment comes many months or even a year or two after the actual delivery of care. We believe this long delay undermines the behavioral response of physicians. Ideally, there would be little or no lag time between the behavior being rewarded and receipt of the incentive otherwise the competing incentives of a busy day might trump the P4P incentive. For example, a physician might have to make the choice between “If I spend 5 more minutes with Mrs. Jones discussing the advantages of a mammogram I might receive an incentive next March vs. if I skip this discussion I will catch up on my schedule for the day.” The immediate incentive of not being behind in their schedule will likely trump the P4P incentive in the physician’s thinking. On the other hand if the physician knows the discussion might result in an immediate $10, the cost-benefit equation might change.

Design Change #4: While withholds have more of an impact than bonuses one needs to be cognizant of the negative psychological response

Previous research has found that individuals are more sensitive to incentives when they perceive they are losing something as opposed to gaining something.²⁰ In one experiment, physicians were asked to make a choice of treatment—either surgery or radiation—for a patient with cancer. In some cases the choice was framed as a loss (probability of dying after surgery) or as a gain (probability of surviving after surgery.)²¹ Physicians were more likely to choose the surgical option when the surgical risk was framed in terms of the probability of living rather than the probability of dying. The difference in the behavioral response for a choice framed as a loss rather than as a gain can be significant, almost twofold in magnitude.²⁰

This loss aversion has implications for structuring P4P incentives. Incentive payments can be structured as a withhold (a perceived loss in income) or they can be structured as a bonus (a perceived gain). The theory suggests that if the goal is to drive physicians to make changes that improve quality, withholding dollars (i.e., framing the incentive as a possible loss) may lead to a greater behavioral response than framing the incentive as a “gain,” in the form of a bonus, even if the same amount of money is at risk.

While framing something as a loss rather than a gain may result in a larger behavioral response, experiments have shown that doing so generally causes a significant negative psychological reaction and violates what the parties exposed to the incentive believe to be fair.²² So while the behavioral response is stronger with a withhold, this benefit is likely outweighed by the risk of angering physicians.

One possible way to take advantage of loss aversion without the negative reaction is through the use of a deposit contract. The P4P program would have two options. The first option would be for the physician to receive $1000 if his score increases by 10 percentage points from the previous year. But the physician has a second option to enter a “deposit”. If the physician submits this deposit of $500 of his money and his mammogram score increases by 10 percentage points the physician receives $2000 (instead of $1000). If he does not increase his score by 10 percentage points, the physician loses the $500 deposit. Such a program has several advantages. First, it introduces loss aversion as the physician will be very motivated not to lose the $500. Second, it takes advantage of the fact that individuals are overly optimistic in predicting their success and the physician electively enters the program. Lastly, it will make it easy for the health plan to identify physicians who are engaged in the program. The downside for a health plan is that health plan’s fraction of the incentive has increased from $1000 to $1500.

Design Change #5: Reducing the complexity of incentive plan increases the behavioral response

When given a choice of potential rewards, most people are risk averse; they will choose an option with absolute certainty over an option involving an uncertain but likely more valuable outcome. This principle of risk aversion is illustrated in a study in which subjects were given a choice between a one-week vacation that was certain or a three-week vacation they had a 50 percent chance of winning. The vast majority of subjects chose the one-week vacation.²⁰ Even though the 50 percent chance of a three-week vacation might be considered a more “rational” choice in strict economic terms because the expected return of such a choice is 1.5 weeks of vacation, most people will choose the sure thing because they perceive it to be a better choice than the possibility of getting nothing at all.

A related phenomenon is that individuals often cannot process complex decisions that are tied to a financial incentive. Current P4P incentive programs are relatively complex for a physician. It is cognitively difficult to keep track of complex trade-offs such as, “If I spend 5 more minutes with Mrs. Jones discussing the advantages of a mammogram I could increase my overall mammogram rate to X% which might put me in the 75^th percentile for my peer physicians and possibly lead to an incentive at the end of the year vs. spending 5 minutes with Mrs. Jones might put me behind for my morning.”²³ Because the P4P program decision is complex while the concern of being late is clear and tangible, the physician is going to push off discussing the mammogram so he or she is not late.

How can P4P programs decrease uncertainty and complexity? As noted above some health plans use relative thresholds, such as paying those physicians in the top quartile of performance, as the basis for determining who “wins.” This type of payout scheme creates great uncertainty for the physician. The level of performance necessary to earn the incentive is unknown until after the fact, frequently 6 months to 12 months later when physicians can be sorted by rank order of performance. A relatively new form of incentive payment being used is a “shared savings” program. If the costs of care for a patient are less than what would be expected and quality measures are met, the health plan and physician group share the savings.²⁴ In a shared saving programs there is uncertainty about whether there will be any cost savings and the complexity of determining how much cost savings there will be to fund incentive payments. In contrast, absolute thresholds known in advance provide greater certainty to the physician trying to hit the target.

The least complex and most certain P4P program would likely be with the “sure thing” of a payment for each mammogram received. Typically, the primary care physician is not paid for a mammogram, but under such a system if one of her patients receives a mammogram, the primary care physician would receive an extra $10. In such an incentive system, the physician knows that if they convince the patient in front of them that they should receive a mammogram they will get an incentive.

Design Change #6: P4P program and incentive payments should be “decoupled” from usual reimbursement

As illustrated in one of our prototypical P4P programs, a common design feature is that the incentive payment is an incremental increase in usual reimbursement (e.g. increasing the per visit reimbursement from $100 to $106). We believe this percentage increase on existing payment undermines the behavioral response of physicians. First, as we noted above an individual perceives the difference between $0 and $6 as being greater than the difference between $100 and $106. Second, in making financial decisions, individuals use “mental accounting.” Mental accounting is a term used to describe how individuals organize, evaluate, and keep track of financial activities.²⁵ By linking the incentive payments to usual reimbursement for a visit, the incentives are therefore mentally linked to usual reimbursement. In that context the incentive payment is minimized because compared to usual reimbursement they seem miniscule.

If the incentive payment was “decoupled” from usual reimbursement, we believe the incentive will garner more of a behavioral response. Instead of incorporating the incentive into the usual fee schedule, the health plan should make the incentive payment separate and special. Practical means of decoupling is to keep correspondence related to usual reimbursement and P4P separate and make incentive payments using a separate check.

One other way to de-couple is to use a lottery which has been successful with patient incentive programs.²⁶. Every week the health plan might hold a lottery for a $10,000 payment. For every one of their patients who received a mammogram in the previous week, a physician gets a virtual “ticket” and the odds of winning are a function of how many tickets they have. Every week an e-mail or letter is sent to all physicians about who won the lottery and how many chances to win they had “earned.” Beyond de-coupling the payment, is that the perceived value of the incentives is higher ($10,000 is likely a significant amount of money for any physician) though in aggregate the health plan is paying the same amount per week. This magnifies the incentive for all participants. Maybe more importantly, we believe under such a system the physician will perceive the incentive to be a pleasant surprise.

Design Change #7: “In kind” rewards may be a stronger driver of change than a dollar reward of the same amount

Dollar incentives might be less effective in driving behavioral change than an object or service of equal value. This is illustrated in how the National Football League creates an incentive for its top players to play in the Pro Bowl. In the past, when offered a financial incentive to play in the game, many players declined. For players with seven-figure salaries, a monetary incentive of several thousand dollars was not enough of an incentive to play an extra game.²⁵ But then the NFL put the game in Hawaii and provided two first-class tickets (for girlfriend or spouse) and accommodations for the players. This “in-kind” incentive has been more effective.

In the same manner, an incentive of all expense paid dinner at a fancy restaurant (worth $250) would be more valuable to a physician than $250 in cash (presuming of course that the physician likes fancy restaurants). Because the physician sees spending $250 on the restaurant as a splurge, it makes the dinner that much more valuable. If fancy dinners appear unseemly, other options include a bagel breakfast for the practice or the latest and most expensive stethoscope. It could even be a choice of several options such as used by credit card reward programs. Ideally the object should be something that the physician would not normally buy themselves.

POTENTIAL WAYS TO MITIGATE UNINTENDED CONSEQUENCES OF P4P

We believe the design changes described above can be applied to a P4P program to maximize the response of a physician to an incentive. We fully acknowledge, however, that one of the major drawbacks of using financial incentives is the potential for unintended and negative consequences. For example, in a recent evaluation of a large pay-for-performance program in England, researchers found that there was a decline in performance on measures not included in the pay-for-performance program performance.¹⁷ Also after a period of rapid improvement there has been concern that the rate of improvement has slowed, because physicians had achieved most of their potential incentive and saw little reason to focus their energies on further improvement.

“Teaching to the Test”

Multidimensional output, or multitasking, refers to situations in which the responsibilities of an individual encompass multiple activities or outputs that may require different types of skills to accomplish. ²⁷ A physician’s “output” includes many different components, such as managing a patient’s chronic illness, the timely and efficient diagnosis of a patient’s new symptom, counseling and advice on how to prevent illness, and emotional support.

Multitasking is relevant to P4P programs because the performance measures in these programs typically address only a narrow slice of a physician’s outputs or the processes that contribute to outputs. For example, a P4P program may reward the receipt of a mammogram but not other processes or outputs that are difficult to measure, such as diagnostic acumen for a patient presenting with unclear symptoms. If a large incentive is applied to one type of output, other outputs might be neglected, and overall care might worsen.²⁷ Thus, a large financial incentive based on a narrowly focused set of measures may lead to the unintended consequence of having a physician “teach to the test,” devoting resources to those things being measured and neglecting other important outputs that are not being measured. Teaching to the test is why few private-sector corporations put a large fraction of employee income at risk with incentives.²⁸ There is mixed evidence on whether current P4P programs in healthcare have actually led to the adverse consequence of “teaching to the test”.^{17, 29}

One classic method of minimizing the likelihood of teaching to the test is to create an incentive program that addresses a broad array of a physicians’ output by applying a broad dashboard of performance measures. This approach has been taken by the primary care physician P4P incentive program in the United Kingdom, which has over 146 quality indicators.⁷ The challenge with this approach is to avoid creating a program that may be overly complicated and expensive. Collecting and auditing quality data is inherently expensive and might outweigh the benefits of the P4P program.

Intrinsic Versus Extrinsic Motivation

Meta-analyses of studies that examined incentive programs in non-health care settings show that while some programs have a positive impact, some have a negative impact.^30-32 The theory to explain these mixed findings is that incentive might cause a conflict between intrinsic motivation, which is a person’s inherent desire to do a task, and extrinsic motivation, which is the external incentive—such as might be provided in a P4P program. Researchers theorize that instead of supporting intrinsic motivation, an extrinsic incentive “crowds out” intrinsic motivation.^{30, 31, 33} Another explanation for this crowding-out effect is that when a task is tied to an extrinsic incentive, people infer that the task is difficult or unpleasant.³⁴ Similar concerns have been raised about the effect of P4P in health care and how it may violate a physician’s sense of professionalism.⁸ An alternative possibility is that a person usually concentrates on only the primary reason for a task rather than the sum of all possible reasons. This theory is used to explain why financial incentives for blood donation are ineffective: the financial incentive is less than the altruistic benefit of blood donation.³⁵

The intrinsic motivation theory implies that a small P4P incentive could either have no effect or actually lead to lower performance if it is tied to something physicians are intrinsically motivated to improve, such as quality of care. A potential way to address the crowding out of intrinsic motivation is simply to increase the size of the financial incentive. A very large external incentive will crowd out any inherent intrinsic motivation; but, in turn, it may create a greater behavioral response than would be obtained through intrinsic motivation alone. A study entitled, “Pay Enough or Don’t Pay at All”, illustrated this concept in a study of IQ tests. Each of four groups were given a different incentive, no incentive or a small, medium, or large incentive for each correct answer.³³ The group given no financial incentive outperformed the group given the small financial incentive (56% versus 46% respectively), and the groups given the medium and large financial incentives (68% both groups) outperformed both of the other groups.

CONCLUSIONS

Together, the theories that we reviewed suggest that the way in which P4P incentives are structured, or framed, can influence whether they achieve the desired behavioral response. We suggest that for a given amount of money, the greatest behavioral response will occur with more frequent, smaller payments. There is evidence that a number of stepped absolute thresholds and “decoupling” incentive payments from usual reimbursement may be more effective than current P4P design. We also discuss the use of lotteries and non-dollar incentives as other mechanisms to increase a physician’s behavioral response.

The potential unintended negative consequences discussed serve as a helpful counterpoint to our recommendations. They emphasize that P4P incentives could lead to the neglect of other important, but unmeasured outputs of a physician and could even have a negative impact on quality. Therefore, any program should closely monitor for these unintended consequences and we have suggested some potential mechanisms to mitigate these risks. If unintended consequences arise, then they will be have to be considered in the cost benefit analysis of a P4P program just as a physician considers potential side-effects before prescribing a medication.

There are several important limitations and caveats to our recommendations. The theories and studies we cite above were used to describe the behavior of individuals, not institutions. It is therefore unclear to what extent the design changes we describe are applicable to hospitals and physician groups. Physician groups are more likely to have the resources to implement changes to improve their performance on a given quality measure that do not require increased effort of individual physicians (e.g. having a nurse call women who have missed their mammogram). While these efforts are important, it is still important to engage individual physician’s in quality improvement. It is notable that even within physician groups such as the Palo Alto Clinic with numerous quality improvement initiatives, physician-specific P4P programs have led to an incremental improvement in quality.¹³

Another caveat is that there are often practical reasons for not choosing the options suggested by these theories. For example, it was noted above that a more frequent payout might lead to a greater behavioral response. Yet this result might be outweighed by the higher administrative costs to the health plan of more frequent processing of data and payouts. We’ve highlighted possible work-arounds to minimize these administrative costs.

Another issue is financial risk. An absolute threshold with an associated incentive with a fixed dollar amount might have advantages in terms of a behavioral response. Yet such an approach leads to greater risk for the payer, which could face the prospect of paying out much more in incentives than was budgeted. In England’s primary care physician P4P program, provider performance greatly exceededwhat was expected, so the cost to taxpayers was considerably more than expected.⁷ While design of incentive payments is important, we recognize that there are other aspects of P4P that impact how providers will respond. These include other P4P design elements (e.g. measures used) and the provider’s practice environment (e.g. availability of electronic medical records.) The variation across health plans in P4P program design also makes it more difficult for providers as they face multiple, and sometimes conflicting, incentives.

Lastly and most importantly, we believe that the lessons from behavioral economics could greatly enhance the design and effectiveness of P4P programs in healthcare, but future work is needed to demonstrate this empirically. It will be critical to test the P4P program design enhancements we propose on both their effectiveness and whether they cause unintended consequences.