5

Health

For decades, policy makers have sought to influence health-related behaviors that contribute substantially to both morbidity and early mortality. Significant successes in influencing health behaviors—such as the multifaceted public health campaign that yielded a 58 percent decline in smoking in the United States between 1964 and the early 2000s—have fueled research attention to influences on health behaviors (Institute of Medicine [IOM], 2007). It is not only the choices that individuals make about their own health that are important to public health goals: understanding of the behavior and decision making of health care providers can help in the design of interventions to reduce medical errors, promote the implementation of evidence-based medicine, and influence outcomes for patients in other ways.

Health researchers have been designing and testing interventions grounded in core behavioral economics principles to encourage positive outcomes for individuals and patients and support provider decision making for more than 20 years. A wide range of behavioral ideas have been tested in varied settings. The committee explored the research in three domains that highlight both landmark work and knowledge gaps that point to research priorities for the future: behaviors that can have substantial effects on cardiovascular health, healthy individuals’ engagement with preventive care, and clinicians’ behaviors.

CARDIOVASCULAR DISEASE RISK

Cardiovascular disease is the leading cause of morbidity and mortality in much of the world, and a great number of intervention studies have

explored ways to lower risk for this disease. Three behaviors—smoking, physical activity, and adherence to medication regimes—have a demonstrated impact on cardiovascular disease and are potentially amenable to interventions grounded in behavioral economics, such as behaviorally informed incentives and use of structured defaults.

Smoking Cessation

It is well known that smoking significantly increases the risk of cardiovascular conditions, including atherosclerosis, coronary heart disease, stroke, peripheral arterial disease, and abdominal aneurysm (Centers for Disease Control and Prevention, n.d.). Smoking is extremely habit forming, and there are numerous strategies people can use to try to break the habit. Policy approaches have included increasing the price of cigarettes by imposing taxes (a traditional economic intervention) or not allowing smoking in public buildings. Strategies that rely on individual behavior have primarily focused on the design, timing, and form of incentives to stop smoking.

Financial incentives to change behavior may be thought of as addressing present bias (see Chapter 3), as well as related internalities. (Economists define internalities as the costs people impose on their future selves, while externalities are the costs imposed on others, such as the effects on others of second-hand smoke.) A 2019 meta-analysis reviewed the long-term effectiveness of financial incentives on smoking cessation, identifying 33 mixed-population, randomized controlled trials (Notley et al., 2019). The studies that met the criteria for inclusion used incentives that ranged significantly in value, from programs in which people deposited their own money in accounts and would lose the funds if they did not meet their self-imposed goals, to rewards worth more than $1,000. Looking across the studies, the authors found strong evidence that even modest incentives could improve rates of smoking cessation and that those rates were sustained after the withdrawal of the incentive.

Notable studies covered in the meta-analysis included a large clinical trial showing that programs in which participants were required to put down a financial deposit in order to ultimately receive a reward were less appealing to participants than ones where they could receive an incentive without having to make an up-front deposit. Overall, the authors of the meta-analysis found that incentives for smoking cessation are effective, with several large-scale studies showing roughly a tripling in long-term smoking cessation rates. However, there are still important questions about optimal design and about whether carrots (reward-based incentives) or sticks (loss-framed incentives) are more effective. When individuals were given the option of participating in an incentive program in which they could put their own money at risk, which would be matched roughly four to one but

could be lost, it was highly efficacious for those who choose to participate. However, participation rates were only about 14 percent; thus, a standard gain-based incentive may be on balance more effective (Halpern et al., 2015). Loss framing of incentives tied to smoking cessation in the form of higher premiums for smokers may be more palatable and thereby more widely used in employer and other settings (Volpp & Galvin, 2014). This work overall is a good illustration of how behaviorally informed incentives are distinct from traditional economic incentives.

Medication Adherence

Numerous studies have shown that adherence to medications prescribed by a physician is well below 100 percent. For example, in the year following a heart attack, adherence to the recommended cardiovascular medications has been estimated at 39–45 percent (Choudhry et al., 2011). A wide variety of explanations have been suggested: forgetfulness, inattention, emotions, misinformation about the benefits and risks of the medication, side effects or fear of side effects, and present-biased preferences that result in decisions that are not consistent with what might be in a patient’s longer-term self-interest (Volpp & Pauly, 2022).

One approach to improving medication adherence is to change some of the underlying defaults to make it easier for patients to follow their physician’s instructions. For example, reducing the frequency with which refills are needed through 90-day prescriptions rather than 30-day prescriptions appears to significantly improve adherence (Rymer et al., 2021). Refilling prescriptions is a complex organizational task for patients taking multiple medications, so synchronized refills in which all medications are on the same refill schedule can also be helpful (Doshi et al., 2016).

Automatic refill programs also increase the ease of obtaining refills. Although entry into such programs can be set up as a default choice, credit card charges are involved, so this approach requires approval from participants before joining, unlike some other default arrangements (see Chapter 10). In one study, however, using an active choice approach in which patients were given the option of signing up every time they refilled a prescription, but in which the convenience of automatic refills was highlighted, roughly doubled the rate of sign-ups (Keller et al., 2011). Increasing social accountability may also be effective in improving adherence. For example, one study showed an increase of 33 percentage points in adherence for posttransplant patients who were told that their adherence would be reported to their transplant physician, although blood samples measuring immunosuppressant levels suggested that some of the measured difference was spurious (Reese et al., 2017).

A significant body of research has shown that requirements in health plans for out-of-pocket payments can discourage utilization of health care services (e.g., Manning et al., 1987). Unfortunately, patients are not always able to differentiate between lower-value and higher-value care, and patient cost-sharing may have detrimental effects in the context of medications and services that are of high value and low cost, such as those used to manage blood pressure or elevated cholesterol. Value-based insurance design, in which patient cost-sharing for high-value services varies on the basis of the particular health care service, was inspired by research that showed the use of higher copayments significantly reduced the use of medications: this approach resulted in saving money on pharmaceuticals without reducing overall health care costs while raising mortality because lower rates of medication adherence led to higher rates of emergency department visits and adverse outcomes (Hsu et al., 2006).

However, lowering copayments has had only modest effects on medication adherence, typically increasing it by 3 to 6 percentage points (Tang, Ghali, & Manns, 2014; Volpp & Pauly, 2022). Even among patients who had recent heart attacks and were given their cardiovascular medications for free, average adherence was only 45 percent, only about six percentage points higher than that seen with regular copayments (Choudhry et al., 2011). A recent clinical trial showed that providing patients who had experienced acute myocardial infarction with financial vouchers that reduced their out-of-pocket costs to zero increased adherence by 2.3 percent but had no effect on the rate of adverse cardiovascular event outcomes one year later (Wang et al., 2019). One reason for these disappointing results is the “dog that didn’t bark” problem (Loewenstein, Asch, & Volpp, 2013). People who are nonadherent do not notice that their copays have been reduced because they are not using (and thus are not paying for) the service. In addition, the small effects of reducing copayments reflect that framing matters and that losses (in the form of higher copayments) generally have greater effects than gains (lowered copayments) do (Volpp & Pauly, 2022).

Other sorts of financial incentives regarding medication adherence have been studied. A review of 15 randomized and six nonrandomized studies showed that behaviorally informed financial incentive interventions (in this case, ones that have been framed to address behavioral reasons people fail to adhere) significantly improved medication adherence. This review also found that larger effects were seen in interventions that offered larger incentives, provided at least weekly reinforcement, and were longer in duration (Petry et al., 2012). Researchers hoping to build on these findings by providing more frequent feedback have tested a daily lottery-based reward for adherence. This work is based on the notion that lotteries might have outsize effectiveness in two ways: by leveraging insights from prospect theory that people overestimate very low probabilities, and by

incorporating anticipated regret in the design of the systems (participants are both told whether they won or whether they would have won had they been adherent; e.g., Kimmel et al., 2012; Barankay et al., 2020; Volpp & Pauly, 2022). While some results have been promising, there have been a number of published studies with negative results (and likely many more unpublished ones; see Chapter 12). Researchers have noted the importance of confirming positive results with testing of expected physiological effects of medication adherence (such as reductions in low-density lipoprotein cholesterol), of focusing these efforts on patients who are demonstrably nonadherent at baseline, and of tying these approaches into other evidence-based approaches.

While various intervention strategies have shown promise, there is no single strategy that has been shown to “solve” medication nonadherence at scale (Zullig et al., 2018).¹ The fact that earlier work showing the effectiveness of incentives in improving outcomes has not been confirmed in more recent studies—even when the incentives offered were worth up to $1,024 per year—highlights several points. First, rigorous randomized testing is critical, and it is necessary to measure both adherence and related medical outcomes. Second, interventions involving patients who are discharged from hospitals highlight the importance of enrolling patients in interventions right at the time of discharge, when they are likely to be most vulnerable. Structural approaches that address barriers to adherence (such as forgetfulness, inattention, inaccurate or incomplete information, and logistical challenges with obtaining refills) appear to be the most valuable. These approaches include synchronizing refills and providing 90-day prescriptions and automatic refills. Removing financial barriers is useful but not a panacea and might best be coupled with low-cost approaches to using technology to increase social accountability. All of these are promising areas for future research.

Physical Activity

The many health benefits of physical activity include reductions in risk for cardiovascular disease and related mortality rates—a relationship that has been well established in the literature (Nystoriak & Bhatnagar, 2018). Researchers have explored the effectiveness of behavioral economic strategies at increasing physical activity in a number of contexts. Studies have suggested that sustained behavior change is achievable through a variety of behavioral mechanisms, including classic habit formation, learning by doing, information acquisition, addressing status quo bias, discovery of new

___________________

¹ That is, no approach has achieved effects in large populations, in contrast with laboratory settings or other controlled experimental circumstances.

tastes, changes in norms or network effects, and changes in choice environments (Volpp & Loewenstein, 2020). However, the most studied approach has been the use of behaviorally informed financial incentives.

A frequently cited study examined the use of behaviorally informed incentives provided to college students to go to the gym frequently for 1 month and then compared the effect on the longer-term behavior of students who received the incentives and those who did not. Those who received the incentives for multiple visits attended roughly 25 percent more often than those who did not receive the incentives, both during the intervention and after it was over. As in other contexts, there is evidence that incentives provided relatively immediately following a desired action are more effective than incentives provided after a time delay. In addition, incentives delivered using loss framing are more effective than those using gain framing (Finkelstein et al., 2016; Patel et al., 2016; Adams et al., 2017; Chokshi et al., 2018).

Studies focused on lottery incentives suggest that an approach that combines significant odds of winning a small award with low odds of winning a large award are more effective than either alone and that incentives based on a combination of group and individual performance are more effective than incentives offered on the basis of only individual or group performance (Patel et al., 2016, 2018).² This body of work highlights that while the magnitude of incentives is important, behaviorally informed design often plays a big role in determining effectiveness.

Another approach that has been studied is gamification, which has been used to encourage physical activity through the application of such behavioral principles as variable reinforcement, loss aversion, the endowment effect, social accountability, and anticipated regret, using contests, progress charts, scoring, and other game-like activities. Examples include an intervention with families to encourage them to meet daily step goals using a system of points and levels, in which individuals start with a moderate level of status that they can lose if they do not meet sufficient daily goals (team members are chosen to represent their team daily at random) and in which there is a built-in social accountability mechanism through the ongoing notification of other team members (Patel et al., 2019). Another promising approach encourages people to commit to a target activity, applying the concept of temptation bundling (or combining wants with shoulds), such as by encouraging people to listen to tempting audio novels while exercising, using iPods available only at the gym (Milkman, Minson, & Volpp, 2014).

___________________

² Evidence for the effectiveness of lotteries is mixed. For example, recent efforts to use them to increase rates of COVID-19 vaccination did not show effectiveness or were mixed (Acharya & Dhakal, 2021; Law et al., 2022).

Taken together, these studies suggest that many approaches hold promise, but they have not yet shown the optimal incentive design for time-limited interventions that will result in sustained behavior change. Three important questions to be considered are what later effects can be expected after the intervention is terminated; whether it is cost-effective for the intervention to be continued indefinitely; and, if so, whether its effectiveness wanes over time.

PROMOTING PREVENTIVE CARE

Behavioral economics findings have been applied in interventions to encourage people to engage in a wide variety of other preventive activities to promote health. We identified three examples with a substantial evidence base: colorectal cancer screening, HIV prevention and treatment, and vaccination. The findings from this work can yield helpful insights about the effects of behavioral economics approaches on health behavior change.

Colorectal Cancer Screening

Cancer is a high-burden condition worldwide. A key approach to secondary prevention is screening, particularly for cancers for which a cost-effective and acceptable screening method exists and for which early detection makes a difference for treatment options and prognosis. Evidence-based screening programs for colorectal and other cancers are in place in many countries and have reduced mortality from these conditions (Smith et al., 2019; Aoki et al., 2020). The general population benefits of screening can only be achieved if individuals comply and complete recommended screening, yet many behavioral barriers prevent patients from seeking screening services or from following up on a clinician’s recommendations, referrals, or prescriptions for screening.

Present bias, inaccurate information, and social norms appear to be especially relevant behavioral factors in this context. For example, people may underestimate their risk from cancer because of optimism bias. The screening process itself may be unpleasant, embarrassing, or simply a hassle, and these factors may be perceived as outweighing the possible long-term benefits. Present bias may lead people to overweight those immediate costs and underweight the benefits of screening, which are intangible, probabilistic, and in the future. Finally, people may not be aware of how accepted screening is in their own social network and may be unlikely to get feedback about this from their network.

Interventions have been designed to address such barriers and increase colorectal cancer screening rates. Behaviorally informed approaches to referral and scheduling procedures, incentives, and reminders have been tested

in diverse settings and with diverse populations. A recent systematic review of studies testing such interventions found that default or opt-out interventions were the most successful, with incentives studies showing mixed results and salience interventions being the least successful (Taylor et al., 2022). In one low-cost intervention that led to higher rates of response, providers mailed fecal immunochemical test kits to patients if they did not explicitly say “no” (opt out) rather than sending the kits only to those who explicitly requested them. A separate systematic review and meta-analysis of studies found that financial incentives added onto a mail outreach program may also have modest benefit (Facciorusso et al., 2021). Individual studies point to potential benefits from many low-cost interventions, such as including an endorsement from a primary care physician on invitations for screening and addressing women’s frequent preference for a same-sex screener by including a picture of a female provider on an appointment scheduling website, thus decreasing the perceived cognitive effort of scheduling.

The research to date points to the promise of low-touch,³ low-cost behavioral interventions, as well as the challenge of realizing that potential. Replication studies, systematic reviews and meta-analyses, and the preregistration and publication of all relevant studies are needed to confirm effects and inform implementation and scale efforts.

HIV Prevention and Treatment

The global HIV epidemic has been transformed in the last decade by the availability of highly effective antiretroviral treatment and by the widespread dissemination and adoption of primary and secondary prevention approaches, including self-testing, medical male circumcision, and preexposure prophylaxis. Theories and principles from behavioral economics can help explain why those at highest risk for HIV might be reluctant to test, or why adherence to antiretroviral therapy may be challenging (Linnemayr, 2017; Linnemayr, Stecher, & Mukasa, 2017; George, Maughan-Brown, & Thirumurthy, 2021). In a setting with substantial HIV prevalence, HIV prevention requires sustained attention and motivation; if the salience of HIV risk is low, limited attentional resources may be allocated elsewhere. Present bias, relevant for all prevention behaviors, may lead to procrastination or avoidance of short-term hassles or time costs if people underweight the future benefits of prevention. Researchers have also investigated the role that other behavioral constructs may play in HIV prevention and treatment decisions, including risk assessment heuristics, social norms, status quo bias, salience, scarcity, information avoidance, optimism bias, and mental models

___________________

³ Low-touch interventions are those that involve comparatively little interaction or interference with the recipient, requiring minimal effort to implement.

(Hutchinson, Mahlalela, & Yukich, 2007; Carey et al., 2011; Dalal et al., 2011; Thirumurthy et al., 2015; Montoy, Dow, & Kaplan, 2018; Pettifor et al., 2020; Buttenheim et al., 2022a). A growing body of evidence has demonstrated the success of interventions that address these decision-making factors using a behavioral lens, including behaviorally informed financial incentives (Thirumurthy et al., 2014; Linnemayr, Stecher, & Mukasa, 2017), promotion of testing and treatment services (Chamie et al., 2018), behaviorally framed HIV risk and treatment effectiveness information (Dupas, 2011; Smith et al., 2021), and interventions that leverage social norms to counter HIV-related stigma (Hutchinson, Mahlalela, & Yukich, 2007; Thornton, 2008).

A meta-analysis of 22 studies of financial incentives to support HIV care goals showed that incentives significantly improved HIV testing take-up, antiretroviral therapy adherence, and continuity of care (Krishnamoorthy, Rehman, & Sakthivel, 2021). A second review, focused on demand-side incentives, showed moderate effects for treatment adherence and substantial effects for voluntary male medical circumcision (Choko et al., 2018). A review of behaviorally informed financial incentives to promote HIV testing found that incentives were effective, whether the incentives included a guaranteed monetary reward; a lottery; or nonmonetary rewards, such as a food voucher (Lee et al., 2014). It is important to note that these systematic reviews did not exclude studies that offered purely financial or transactional incentives: that is, only some of the interventions included in these reviews incorporated behavioral factors into the design and delivery of the incentives. However, several studies of incentives for HIV-related behaviors demonstrated that larger incentives do not always produce larger effect sizes than smaller incentives, supporting an underlying behavioral mechanism (Thornton, 2008; Thirumurthy et al., 2014).

Vaccination

Resistance to vaccines has been a reality since long before federal or other government agencies began routinely requiring programs of regular vaccination for a variety of diseases; that resistance reflects complex social factors (Conis, 2015). In more than 20 years of study, researchers have identified aspects of decisions about vaccination (made either for oneself or on behalf of a child) that can be explained by behavioral concepts (Buttenheim & Asch, 2016; Brewer et al., 2017a).

Vaccination decisions involve time-inconsistent preferences: the benefits are in the future and are both probabilistic and intangible. Yet the costs are immediate, certain, and salient. Vaccination decisions are strongly influenced by social norms and are susceptible to predictable errors or biases in risk assessment. Specifically, for example, researchers have examined

omission bias, status quo bias, and availability heuristic as drivers of negative vaccination decisions (Meszaros et al., 1996; Connolly & Reb, 2003; Opel & Omer, 2015; Chen & Stevens, 2017).⁴

Numerous interventions have been studied to encourage people who do not intend to be vaccinated to do so (i.e., to increase motivation or intentions), to address the gap between intention and follow-through, or both. A recent systematic review (Reñosa et al., 2021) found that the most promising behavioral evidence for increasing rates of vaccination (where it is not required) points to three strategies: making information about vaccines more salient (e.g., Milkman et al., 2011; Chen et al., 2020; Szilagyi et al., 2020), offering incentives (e.g., Banerjee et al., 2010; Chetty-Makkan et al., 2022: Schneider et al., 2023), and changing defaults (e.g., Brewer et al., 2017b). Individual studies have also found support for social proof interventions—ways of communicating expert or social group endorsement of vaccination (e.g., Bartoš et al., 2022).

A detailed review of the determinants of vaccine take-up and proposed interventions to address concluded that interventions to establish the intention to get vaccinated—including providing information about benefits, boosting trust in the vaccine itself and in sources of information about it, and combating misinformation—are both necessary and effective (Brody, Saccardo, & Dai, 2022). This finding adds valuable insights to the current state of thinking about behavioral approaches to this persistent health care challenge. For people who already intend to get vaccinated, reminders, prompts that assist with planning, and addressing specific barriers (e.g., simplifying access to vaccination) are most effective.

A second key takeaway from the research to date on vaccine acceptance, which is relevant for many other health topics and diverse policy domains, is that the effectiveness of specific behavioral interventions will vary depending on underlying participant characteristics (including a person’s motivation to get vaccinated and beliefs and attitudes about vaccination) and contextual factors (e.g., the geographic location of the intervention or the epidemiologic characteristics of the vaccine-preventable disease). For example, an experiment conducted in eight countries tested a message about the harm that could come to others who might be infected if the population is not thoroughly vaccinated against the flu—in order to tap prosocial motivations (Li et al., 2016). This intervention was more effective among people who had not received a vaccine the previous year than among those

___________________

⁴ Omission bias is the tendency to consider negative outcomes caused by actions as worse than those caused by a failure to act. Status quo bias is the tendency to prefer things as they are over making a change. The availability heuristic is a term for the tendency to rely on information that is readily available in making a decision rather than seeking information that might be more relevant or helpful.

who had, suggesting that this approach was more likely to show results with people who had a lower motivation to be vaccinated in the first place.

One study provides particularly concrete evidence of this phenomenon. In a head-to-head comparison of multiple nudge-type text messaging interventions designed to increase take-up of the flu vaccine (Milkman et al., 2021), the strategy with the largest estimated effect size was a “reserved for you” or ownership message. These results were supported by an even larger-scale study at a retail pharmacy chain, in which the message was changed slightly to refer to the vaccine “waiting for you” (Milkman et al., 2022). Although this ownership strategy notably did not significantly outperform many of the other tested nudges, many local public health agencies applied the “reserved for you” message in COVID-19 vaccine promotion campaigns based on this promising signal, desperate for strategies that could motivate vaccination and bolster demand. Further analyses of the original study (Buttenheim et al., 2022b; Patel et al., 2022), as well as multiple attempts to apply ownership messages in the COVID-19 vaccine context (Dai et al., 2021; Rabb et al., 2022), have demonstrated that the results are more nuanced, succeeding in some settings, for some populations, and at some moments in a vaccination campaign more than in others. This points to the importance of selecting and designing interventions that are carefully tailored to the barriers to take-up identified in a target population, testing the same intervention in diverse settings and contexts, and evaluating mechanisms of action and subgroup responses wherever possible.

PROVIDER BEHAVIOR

It has been well documented in many countries that patients often either receive too little or too much health care relative to what is clinically indicated. For example, more than 20 percent of care provided may actually be unnecessary (IOM, 2013; Carrol, 2017; Lyu et al., 2017). In some cases, clinicians make errors, despite training and treatment protocols. More generally, clinicians clearly respond to traditional economic incentives: they provide more services when they are paid per service than when they are paid through such arrangements as salary or full-risk capitation, which raises concerns about potential conflicts with their patients’ best interests (Larkin & Loewenstein, 2017). In addition, there is growing recognition that clinicians, like other people, are affected by behavioral factors when they make decisions. Clinicians often work in busy settings that place multiple demands on their attention and must make frequent, often high-stakes decisions under time pressure. Such conditions mean that factors including the structure of choices and defaults, cognitive biases, and decision fatigue are likely to play a significant role in clinicians’ decision making.

A number of behavioral interventions to address these challenges have been explored. A 2020 review of 17 behavioral interventions in the United States found that the interventions most studied were changing default settings (such as encouraging prescribing of generic medications) and using information on relative performance to influence behavior using social norms (prescribing behavior was the most frequently targeted behavior; Wang & Groene, 2020). Both of these interventions were found to be consistently effective at modifying clinicians’ behavior to be more consistent with existing guidelines; the effects were often quite large with minimal costs. However, the authors caution that more research is needed to clarify the mechanisms and determine the optimal approaches to support wider implementation.

Checklists can be seen as a way of both creating a social norm and changing defaults. In essence, a requirement to complete a checklist shifts the default from an opt-in (need to remember each order to write when admitting a patient) to an opt-out (the most relevant orders are listed and the clinician can opt out or modify): the opt-out approach makes it much more difficult to forget something. The many documented benefits of checklists in improving quality and safety in health care include nearly eliminating catheter infections that used to kill about 30,000 Americans per year (Pronovost et al., 2006; Gawande, 2010).

Other work has pointed to the potential for active-choice interventions. In such interventions, decisions about specific treatments for specific patients are made more salient. For example, one study examined the effects of identifying for providers the patients for whom statins may be appropriate and asking them to select whether to prescribe or not; it showed an increase in prescribing rates (Patel et al., 2018). This study also showed that providing feedback on prescribing rates relative to their clinical peers contributed to higher prescribing rates. Similar approaches have also been effective in increasing rates of cancer screenings (Patel et al., 2016).

Some social incentive interventions have had large effects, but the results have been highly variable. For example, rates of inappropriate prescribing of antibiotics were brought close to zero when clinicians were asked to provide a justification in a free-text response for treatment decisions or provided with peer comparisons (Meeker et al., 2016). These effects were much larger than those observed in some other efforts to use social norms and social comparisons to motivate behavior change. One possible reason for the large effect is that physicians are highly motivated not to appear below average or to be seen as practicing inappropriately by their peers. In contrast, an example of a smaller effect comes from the use of Opower, a software platform that uses artificial intelligence to apply behavioral principles that was developed to support utilities in helping consumers reduce energy usage by making more energy-efficient choices. In this case, the Opower social comparisons only showed consumers how their

own household compares with others and therefore invokes significantly less social accountability than the physician’s peer comparisons.

Other studies have shown more mixed results. For example, sending clinicians letters comparing their prescribing rates with those of other local practices had less effect than the study described above (Hallsworth et al., 2016). A recent study by Gauri and colleagues (2021) tested a performance-contingent social recognition intervention to increase performance in the tracking of income and expenditures at primary care clinics in Nigeria. Weekly performance was posted by use of the number of stars on a prominently displayed certificate of excellence, with the best-performing facility promised a public award ceremony with the secretary of health. In one state there were consistent and large effects on performance (roughly 18 percent), but in the other state there was no detectable effect. Differences in observable characteristics between the facilities did not explain the cross-state differences in impacts. In short, it is not clear why some of the interventions that use social incentives to motivate clinician behavior have larger effects than others. That variability may relate to such factors as the timing of interventions, the comparison groups used, the salience of the information provided, the source of information, the fidelity of implementation, or the degree to which comparative information provided new information.

Finally, a number of studies have examined the use of financial incentives, such as bonuses for improved quality, as an enhancement to the existing fee-for-service payment system. These types of incentives are structured as separate incentives or bonuses (rather than simply folded into salaries) to more specifically target behaviors, but the effects have typically been modest. Despite efforts to increase the use of such value-based payment arrangements, standard fee-for-service arrangements still account for roughly 70 percent of compensation for both primary and specialist physicians (Reid et al., 2022). Open questions include how to optimally simplify incentive systems that adequately reflect the quality of care but avoid choice overload; how to distribute incentives among individual physicians in practice groups; the magnitude of incentives needed to change different types of behaviors; and how to optimally combine financial and nonfinancial strategies. All of these questions highlight the need for different approaches to be carefully tested and evaluated (Emanuel et al., 2016).

FINDINGS

The committee’s review of behaviorally based interventions in the three selected areas of health care demonstrate strong effects for many of them. Looking across the work in the areas we examined, we saw considerable support for the effectiveness of interventions that address the five core principles discussed in Chapter 3. Although it is difficult to generalize across

this large volume of work, we note that present bias, limited attention, and social norms were all repeatedly identified as intervention targets. Specific findings from this body of work include the following:

• Structural approaches to lowering barriers to medication adherence (such as forgetfulness, inattention, and inaccurate or incomplete information) appear to be the most promising interventions.
• Behaviorally informed incentives using loss framing show promise for increasing physical activity.
• Default or opt-out interventions were the most successful among interventions tried for increasing rates of colorectal cancer screening.
• Behaviorally informed financial incentives, promotion of testing and treatment services, and provision of information about HIV risk and treatment effectiveness were all promising interventions for increasing HIV prevention and adherence to treatment.
• The most promising behavioral evidence for increasing vaccination has been found for three strategies: making information about vaccines more salient, offering incentives, and changing defaults.
• To modify clinicians’ behavior to be more in line with guidelines, the most frequently studied interventions have been changing default settings and providing comparative information on the performance of peers to influence behavior.

Although health is one of the policy domains with the largest evidence base for policy interventions, many important questions and challenges remain. Below we list priorities for follow-up work, but we stress that, while rigorous testing of conceptually based interventions remains critically important to map out the comparative effectiveness of different approaches in improving health or the value of health care delivery, assessment of how to tie together approaches that demonstrate efficacy into effective population health solutions might benefit from collaboration with experts in innovation and implementation science.

Four topics present important opportunities for further research: making behavior changes stick, preventing nudge fatigue, precision nudging, and designing for scaling and planned replication.

Making Behavior Changes Stick

There are hundreds of published studies demonstrating significant positive (and sometimes large) effects of time-limited behaviorally informed interventions. Evidence of sustained and sustainable behavior change is rarer, unfortunately, although there are good examples from the areas of smoking cessation and financial incentives, among others. Evidence-based strategies for changes that are sustained are badly needed and might benefit

from drawing on some of the different mechanistic approaches for achieving sustained behavior change, such as classic habit formation, learning by doing, information acquisition, addressing status quo bias, discovery of new tastes, changes in norms or network effects, and changes in choice environments (Volpp & Loewenstein, 2020).

Preventing Nudge Fatigue

In some health care settings, many clinicians may be finding there are too many nudges, which runs the risk of diminishing the incremental effectiveness of each such approach (see, e.g., Ancker et al., 2017; Blecker et al., 2019; Kwan et al., 2020; Chen et al., 2022). There are only so many electronic health record pop-ups, best practice alerts, or peer-comparison dashboards that clinicians can take in during a busy day. Attention is needed to the problem of making such nudges less intrusive and for prioritization of the most effective strategies for use at an organizational level.

Precision Nudging

Interventions informed by behavioral economics rarely work equally well for all. Uncovering heterogeneous treatment effects—especially those that widen rather than close health disparities—is an important but sometimes neglected aspect of intervention trials in health and health care. Machine learning and related approaches to identifying variations in treatment effects can support better targeting and tailoring of interventions for maximum impact. While this identification typically requires large samples for adequately powered subgroup analyses, this approach could make possible the tailoring of interventions to different populations, increasing the amount of benefit for the resources expended.

Designing for Scaling and Planned Replication

Recent popular books, such as The Voltage Effect: How to Make Good Ideas Great and Great Ideas Scale by John List, have highlighted that promising ideas often do not scale because the initial demonstration of efficacy may represent a best-case scenario. Replication testing is typically not adequately rewarded in academia, but it is critically important in determining whether a given intervention actually is effective before any decisions are made about further scaling. Careful consideration should be given to the assessment of context and execution, with researchers working together with implementation scientists to design a path for advancing promising ideas that show efficacy to increasingly pragmatic tests of effectiveness in different populations to determine the effectiveness and cost-effectiveness of any given intervention before the decision is made to use it more widely.

REFERENCES