VALUE IN HEALTH 17 (2014) 814–822

Available online at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/jval

Patient-Reported Outcomes

Conceptual Model and Economic Experiments to Explain Nonpersistence and Enable Mechanism Designs Fostering Behavioral Change Behnud Mir Djawadi, PhD1,*, René Fahr, PhD1, Florian Turk, PhD2 1

Department of Management, University of Paderborn, Paderborn, Germany; 2GlaxoSmithKline plc, Brentford, Middlesex, UK

AB STR A CT

Background: Medical nonpersistence is a worldwide problem of striking magnitude. Although many fields of studies including epidemiology, sociology, and psychology try to identify determinants for medical nonpersistence, comprehensive research to explain medical nonpersistence from an economics perspective is rather scarce. Objectives: The aim of the study was to develop a conceptual framework that augments standard economic choice theory with psychological concepts of behavioral economics to understand how patients’ preferences for discontinuing with therapy arise over the course of the medical treatment. The availability of such a framework allows the targeted design of mechanisms for intervention strategies. Methods: Our conceptual framework models the patient as an active economic agent who evaluates the benefits and costs for continuing with therapy. We argue that a combination of loss aversion and mental accounting operations explains why patients discontinue with therapy at a specific point in time. We designed a randomized laboratory economic experiment with a student subject

pool to investigate the behavioral predictions. Results: Subjects continue with therapy as long as experienced utility losses have to be compensated. As soon as previous losses are evened out, subjects perceive the marginal benefit of persistence lower than in the beginning of the treatment. Consequently, subjects start to discontinue with therapy. Conclusions: Our results highlight that concepts of behavioral economics capture the dynamic structure of medical nonpersistence better than does standard economic choice theory. We recommend that behavioral economics should be a mandatory part of the development of possible intervention strategies aimed at improving patients’ compliance and persistence behavior. Keywords: behavioral economics, financial incentives, health behavior, laboratory experiment, medication persistence, medication adherence, mental accounting, patient behavior, patient compliance, prospect theory.

Introduction

interventions aimed at enhancing medication persistence, such as patient education and training, feedback loops and reinforcement, or drug presentation and functionality, were shown to be at best moderately effective and rather limited in promoting sustained behavior change [16–19]. A second approach to realign behaviors is to provide the patient with financial incentives. Once considered as a promising approach toward achieving healthy behavior, the effectiveness of their use remains insufficient and inconclusive, leaving many questions about potential modifiers including form, size, and duration of different financial incentive programs unanswered [20,21]. These empirical findings strongly suggest that the problem of nonpersistence is rooted in numerous causes and requires interventions that use insights from different fields of studies including epidemiology, psychology, sociology, and economics [11,22]. Researchers have started bridging behavioral economics and health to understand how potential modifiers rooted in the psychology of patients such as reasoning fallacies, cognitive

Poor treatment persistence with prescribed medicines has long been recognized as a major obstacle for effective treatment and health care efficiency [1–3]. More recent studies for a wide range of chronic diseases such as asthma, diabetes, and hypertension reveal that only 50% to 65% of the patients adhere to the recommended medication usage [4–7]. The deviation from the prescribed medication-taking behavior exposes the patient to higher risks and leads to poorer health outcomes and increased morbidity and mortality [2,8,9]. The associated costs for additional health care services and forgone investments for the development and utilization of new and efficacious but ultimately ineffective drugs and therapies have escalated into billions of dollars annually [10–14]. Employers report poor health habits as the main challenge to maintaining affordable benefits [15]. Traditionally, one approach to shortcomings in patient behavior has been to view the problem as information gap. Various

Copyright & 2014, International Society for Pharmacoeconomics and Outcomes Research (ISPOR). Published by Elsevier Inc.

* Address correspondence to: Behnud Mir Djawadi, Department of Management, University of Paderborn, Warburger Street 100, D-33098 Paderborn, Germany. E-mail: [email protected]. 1098-3015$36.00 – see front matter Copyright & 2014, International Society for Pharmacoeconomics and Outcomes Research (ISPOR). Published by Elsevier Inc. http://dx.doi.org/10.1016/j.jval.2014.08.2669

VALUE IN HEALTH 17 (2014) 814–822

815

Fig. 1 – Three phases in a stylized medical treatment for chronic diseases. biases, and errors in belief formation affect the patient’s medical decision-making process and prevent behavioral changes. Understanding human biases allows constructing levers through technologies, social networks, gamification, contracts, and incentives. Employers, insurers, pharmacy benefit managers, and companies are starting to experiment with approaches using behavioral insights [23–28]. Comprehensive research about the reasons of nonpersistence from the behavioral economics perspective, however, is scarce. Although patients are viewed as economic agents who perform a cost-benefit analysis when deciding about medication intake [29], the empirical application of economic models to derive individual determinants of nonpersistence and to predict behavior is limited [22]. The reasons why the existing research in health care is slow on that matter is manifold and apparent: 1. The few attempts to economically explain medical nonpersistence only consider rational choice behavior and ignore promising approaches from behavioral economics that augment standard economic theory along with greater psychological realism [30]. 2. There is no theoretical framework that captures the dynamic nature of medication-taking behavior and links the individual decision to the observed outcome over time. 3. The empirical research based on observational data or clinical studies lacks the ability to identify the general drivers of the patient’s decision-making process to discontinue with therapy. Instead, existing clinical meta-analyses and retrospective studies identify determinants that imply correlations rather than causal relationships with observed outcomes [31]. 4. With a focus on clinical trials (randomized and controlled as well as pragmatic) to assess the impact of behavioral incentives, there is a lack of a scalable approach to design, test, and calibrate tailored incentive schemes. The aim of the study is to explain medical nonpersistence from an economics perspective. We develop a conceptual framework that augments standard economic choice behavior with psychological concepts of behavioral economics to understand how patients’ preferences for discontinuing with therapy arise over the course of the treatment. This integrated model incorporates the key features of a medical treatment and generates numerous behavioral predictions. Using the method of experimental economics, the predictions are validated within an economic setting, allowing for general conclusions about the individual’s decision-making process and economic behavior in medical treatments. We proceed as follows. In a first step, we describe the different concepts of our framework and illustrate how we abstract from the medical context and create an

equivalent economic environment to clearly isolate the potential economic drivers of medical nonpersistence. In a second step, we explain the method of experimental economics and present how we designed a randomized economic experiment to validate the applicability of the conceptual framework by testing the underlying behavioral hypotheses under controlled conditions.

Methods Conceptual Framework We build upon a discrete choice framework that models the patient as an active agent who evaluates the benefits and costs for continuing with therapy and decides upon the assessment of this trade-off. Referring to the typical decreasing shape of persistence behavior observed in clinical studies [4,7,12,13,32,33], we argue that this assessment on the part of the patient varies along the treatment and identify three phases: 1) the phase of invasion, 2) the phase of high persistence, and 3) the phase in which discontinuation with therapy is expected to occur (Fig. 1). The phase of invasion represents the beginning of the medical treatment. The patient is managing access to the medicine and takes the medicine without experiencing any improvements because a certain time and a certain threshold level are needed for the medicine to become efficacious. After this threshold level is met, patients are observed to comply extremely well (“phase of high persistence”), yet fail to do so at a specific point in time and then start discontinuing with therapy (“phase of expected variation in persistence behavior”). Assuming that the costs and benefits are constant throughout the treatment, standard rational choice theory is unable to explain the behavioral deviation in the third phase. Once agreeing to the treatment and complying well at the beginning, the patient is always better off by continuing with therapy. A broad range of empirical studies conclude that individuals fall prey to reasoning fallacies and do not make perfectly rational decisions because rationality may be limited by time, risk and uncertainty, incomplete information on alternatives, and complexity [23,34–37]. We therefore build our conceptual framework on the principles of behavioral economics and include limited human rationality into the economic decision-making process. The patients’ decision-making process is assumed to follow the concept of mental accounting [38,39]. This concept has been used to explain a wide range of consumption and spending behavior [40,41]. The economic evaluation of alternatives follows the prospect theory [42] and is modeled by a special value function whose shape exhibits three essential characteristics (Fig. 2): 1) the value function is defined over gains and losses relative to some

816

VALUE IN HEALTH 17 (2014) 814–822

develop preferences to discontinue with therapy over the course of the medical treatment. According to our conceptual framework, the patient incurs costs during the phase of invasion. The feeling of loss is due to the efforts required to manage access to the medicine and the fact that the patient experiences the inconvenience of the unfamiliar medication intake procedures or therapy peculiarities without receiving any benefits in terms of health state improvements. After the medicine becomes efficacious, the patient is highly compliant as a means to compensate for the endured losses from the phase of invasion. Once the losses are compensated, however, patients value any further gains of being compliant lower and consequently tend to discontinue with therapy.

Setting an Economic Environment

Fig. 2 – The value function defined over gains and losses under the prospect theory. reference point rather than final wealth levels. 2) The value of a gain, and likewise, the disvalue of a loss both increase at a diminishing rate. Hence, the value function is concave in the domain of gains and convex in the domain of losses. 3) Last, the value function incorporates the notion of loss aversion, which means that losses are valued more than gains. The dynamics of this value function indicates that benefits and costs can be evaluated differently depending on the timing and the current wealth status. Indeed, Thaler and Johnson [43] find evidence that future gains are valued less if previous losses are compensated and the possibility to break even or enter the sphere of gaining is at hand. This causes individuals to deviate from a formerly riskaverse strategy to more risk-seeking behavior. A combination of loss aversion as described by prospect theory and mental accounting operations can thus systematically influence the decision for a specific option and might explain why patients

Table 1

To examine whether this hypothesized behavioral pattern in fact explains medical nonpersistence, we implement—in accordance to the standard view of health as an investment good and part of the human capital [44]—a simple investment setting over several time periods. Let us assume that a financial investor has an initial wealth status and agrees to a new investment plan that is scheduled over several time periods. In each time period, the investor can choose between two investment options: a risky option that not only generates high revenue but also entails a higher risk of losing money or a less risky option that not only generates lower revenue but also has a far lower risk of losing money. Once money is lost, there is no opportunity to invest anymore and the investment plan is aborted. The same trade-off between benefits and costs associated with choices under risk and uncertainty can be observed in medical treatments (see Table 1). Starting the medical treatment with an initial health status, the patient makes a choice between two options in each time period: investing in health by continuing with therapy, which maintains the current health status but incurs costs of persistence (e.g., adverse effects and medication costs in form of copayments), or not investing in health by discontinuing with therapy, which maintains the current health status at no cost but entails a higher risk of suffering a noncompensable relapse.

– Similarities between a medical treatment and an economic investment setting.

Setting Medical treatment

Objective Maintain initial health state

Duration Treatment is scheduled for several time periods

Decision

Costs/benefits

Outcomes

Patient decides between two options: continuing or discontinuing with therapy

Continuing with therapy:
adverse effects/ decreased risk of suffering a relapse

Health outcome is linked to certain probabilities that are determined by the decision to continue with therapy or not. If relapse occurs, the treatment is cancelled.

Discontinuing with therapy:
no adverse effects/

higher risk of suffering a relapse Economic investment setting

Maintain initial wealth status

Investment plan is scheduled for several time periods

Investor decides between a less or more risky option

Less risky option:
higher investment costs/decreased risk of losing money More risky option:
no investment costs/

higher risk of losing money

The subsequent wealth status is linked to certain probabilities that are determined by the option chosen. Once money is lost, the investment plan is aborted.

VALUE IN HEALTH 17 (2014) 814–822

The decision to invest in the risky or less risky option is the economic equivalent of the medical decision of continuing with therapy or not. Choosing the less risky option represents continuing with therapy because it involves costs but decreases the risk of obtaining the negative outcome such as suffering a medical relapse. Choosing the risky option is equivalent to discontinuing with therapy. In this case, no costs emerge, but the risk of suffering a medical relapse is far higher. Both the patient and the financial investor face probabilistic risks and corresponding consequences with their choices. If a noncompensable relapse occurs, the medical treatment is cancelled. Congruently, once money from the investment is lost, the investment plan is aborted. The transformation of the medical treatment to an equivalent investment setting enables us to simulate the course of events inherent in medical treatments from an economics perspective and isolates the economic drivers of medical nonpersistence. For this reason, we validate the applicability of our conceptual framework in the economic environment by investigating whether investment behavior dynamically changes from choosing the less risky option to choosing the risky option over the course of an investment plan.

Medical Nonpersistence in an Economic Experiment A common approach would be to analyze observational field data on behavior in investment decisions that exhibit characteristics described in our conceptual framework. Those data, however, are not available. Even if we had data on decisions comparable to our investment setting, too many important factors such as the market environment, the experience of the decision makers, and particularities of the investment would be beyond the control of the researcher and could therefore not be considered in the statistical analysis. Economic experiments have emerged as a vital component of economic research and are regarded as a complement to other more traditional forms of empirical economic analysis, such as observations and surveys [45,46]. Unlike survey studies, which create a hypothetical environment with hypothetical responses, experiments offer the possibility to observe actual behavior based on real decisions under monetary incentive conditions [47]. To explain patients’ persistence behavior, we therefore embed the economic investment setting into an experimental environment. As usually done, we run our economic laboratory experiment with a student subject pool. The reasons are that students have the ability to grasp abstract situations more easily and quickly understand the rules of the experiment [48]. Furthermore, as Amiel and Cowell [49] argue, students serve as a better subject sample for testing the validity of theoretical principles than do members of the general public because their motivation is often tied to intuition or specific situations, which, in many cases, have been proven a poor guide to criteria of general applicability. Moreover, the predictions derived from economic models, such as the behavioral economic model used to explain the dynamics of medical nonpersistence in the present article, are independent of assumptions concerning participant pools [50]. Bleichrodt et al. [51] investigate the tradeoff between equity and efficiency in the allocation of health by using the nonlinear rank-dependent quality-adjusted life-year model. In an economic laboratory experiment, preferences are

817

elicited both in a sample of Dutch students and in a sample from the general population. No significant differences are observed between the views of the students and of the representative sample of the Dutch population. The experiment is modeled as a simple investment game and consists of two stages (Fig. 3). The first stage, called the working stage, is intended to induce feelings of losses for the participating subjects. Subjects invest effort and time in the laboratory to work on a real task, which is known in the experimental literature as the slider task [52]. The task is to position tiny sliders on the computer screen. Subjects have no time limit for working on the slider task and only when all 48 sliders are positioned correctly the task is completed and subjects get paid for their work. Subjects, however, get only a fraction of their proper income. Instead of receiving 8 units of the experimental monetary currency taler, they receive only 2 talers, and so subjects endure a loss of 6 talers when they enter the second stage (income o 0). The second stage is called the lottery stage and mimics the investment plan scheduled for 12 time periods. In each period, subjects choose between two investment options: a risky lottery A or a less risky lottery B. Both lottery A and lottery B generate the same monetary outcomes but differ in the risk of receiving either the high or the low outcome of 2 or 0 talers, respectively. In lottery A, subjects have the chance to win and receive 2 talers with a probability of 70% or lose the lottery and receive 0 talers with a probability of 30%. The investment in lottery A comes with no additional costs. Lottery B yields the outcome of 2 talers with a far higher probability of 95% but subjects have to pay 1 taler to play this less risky option. Once the subjects have chosen between the two lotteries, a randomization device determines whether the lottery is won or lost according to the probabilities that are associated with each outcome of the chosen lottery. A crucial feature of the experiment lies in the “sudden death” mechanism. Once the subject loses the lottery, the experiment is over. It is noteworthy that both lotteries may lead to a dropout, but the probability for a dropout is far more when opting for lottery A repeatedly. In the case of winning the chosen lottery, the subject is able to again decide upon the lotteries and the obtained money is cumulated over the single periods. In the case of losing the lottery, no further decisions about the lotteries in the next periods are possible, and so the current account balance remains unchanged until the end of the experiment. The experimental investment setting incorporates the key features of a medical treatment and allows investigating the behavioral predictions of our conceptual framework. The working stage of the experiment mimics the phase of invasion, whereas the lottery stage imitates patient’s decisions to discontinue with therapy by choosing lottery A or to continue with therapy by choosing lottery B. The dynamics of persistence behavior is incorporated by loss aversion in the working stage and repeated decisions on lotteries A and B in the lottery stage, which precisely state at what period discontinuing with therapy should be observed. Under standard economic choice theory, a risk-neutral decision maker decides in favor of the lottery with the highest expected value. Given the probabilities of each lottery and the number of decisions to be made, the optimal decision sequence would be to choose lottery B until period 10 and then switch to lottery A for the two remaining periods (see Appendix A

Fig. 3 – The two stages of the experiment and expected persistence with therapy.

818

VALUE IN HEALTH 17 (2014) 814–822

in Supplemental Materials found at http://dx.doi.org/10.1016/j. jval.2014.08.2669). In line with the evidence of Thaler and Johnson [43], our conceptual framework suggests that the perception and the evaluation of the risk component associated with each lottery depend on the timing and the subject’s current wealth status. Starting the lottery stage with a loss of 6 talers, it is predicted that subjects choose lottery B constantly at the beginning of the investment plan because this is the least risky strategy to compensate for the losses endured in the working stage. Once the losses are compensated, the reference point of the value function is shifted upwards so that a nonnegative income (income Z 0) leads the subject to value future gains lower than before. This means that from period 7 onwards, when all losses are compensated, subjects start to avoid the investment costs and prefer the more risky option lottery A to lottery B, exhibiting an investment behavior that dynamically changes over the course of the investment plan.

Experimental Procedure The experiment was conducted at the Business and Economic Research Laboratory between September 2011 and December 2011 at the University of Paderborn. Subjects were randomly recruited from a pool of approximately 1800 voluntary students from different fields of study who are enrolled as prospective participants in economic experiments. In total, 107 subjects participated in the experiment of whom 56 (52.34%) were female and 51 (47.67%) were male. The experiment was computerized and programmed with the software zTree [53]. As soon as the subjects arrived at the Business and Economic Research Laboratory, they were asked to randomly draw a number from a box and were told to sit down at the assigned computer workplace in a cubicle detached from each other, ensuring complete anonymity. Subjects received the same introductory talk and were told not to communicate during the complete time in the laboratory. Then, the written instructions (see Appendix B in Supplemental Materials found at http://dx.doi. org/10.1016/j.jval.2014.08.2669) were handed out and the subjects had 10 minutes time to read and ask questions in private to clarify any misunderstandings. The instructions were written in a context-free manner so that no associations with medical treatments, pharmacies, and so forth could emerge that might influence subjects’ decisions in an uncontrolled way. In the instructions, subjects were assured that all decisions were made anonymously, so that neither the experimenter nor other participants got to know the identity of the subject who made a specific decision. Each subject had to decide only for himself or herself and was not informed about the decisions of others. All decisions were typed into the computer program. Subjects were informed that their monetary earnings would be a function of their choices during the experiment and would not depend on the decisions of the other subjects. The earnings of the experiment were exchanged from the experimental monetary currency taler to euro and handed out in cash to the subjects right after the end of experiment. The experiment lasted for about 1 hour and subjects earned €16.83 on average. (The amount €16.83 was worth roughly $22.55 at the time the experiment was conducted.)

Results Distribution of Lottery Choices A and B in the Investment Plan Table 2 presents how the participating subjects decided between the more risky lottery A and the less risky lottery B throughout the lottery stage of the experiment. The vast majority of roughly 80% started the decision sequence with lottery B in the first period. Because the risk of dropping out of the experiment

Table 2 – Distribution of lottery choices A and B in lottery stage. Starting lottery in period 1

A (%)

B (%)

Subjects (n ¼ 107) Dropout rate (overall) Survival rate until period Survival rate until period Survival rate until period Survival rate until period

19.70 95.2 57.14 33.33 14.29 4.76

80.03 74.4 89.53 72.09 52.33 25.58

3 6 9 12

Note. Survival rate is the proportion of subjects who won all chosen lotteries A and B, respectively, in the experiment until the designated period.

accumulates with each lottery decision, the risk-dominant strategy favors the constant choice of lottery B. In fact, when lottery B was chosen, the average rate of dropping out of the experiment at any point along the investment plan of 12 periods was 74.42% compared with a dropout rate of 95.24% when subjects started with lottery A. Looking at this result from the other perspective, we see that subjects who constantly chose lottery B had a higher chance of winning the lotteries and thus survive the single periods. For example, only 57.14% of the subjects who constantly chose lottery A survived the first three periods, while roughly the same proportion of subjects survived six periods more until period 9, when lottery B was chosen.

Persistence in Experimental Investment Setting Our persistence measure is based on the lottery choices A and B in the investment plan. In analogy to a medical treatment, the participating subject is said to continue with therapy when he or she chooses lottery B. Once lottery A is chosen, he or she is said to discontinue with therapy. We define the persistence rate as the ratio of lottery B over lottery A choices made by subjects per period. We restrict our sample population to only those subjects who started the investment plan with lottery B and either always continued with therapy or discontinued with therapy at a certain point in time by switching once from lottery B to A and then keeping the constant choice of lottery A until the end of the scheduled time periods or until they lost a lottery and dropped out of the experiment. The reason for truncating the observational data is that only those subjects serve as the population of interest who agree on medical treatment in the beginning, stay with therapy, or exhibit clear preferences for discontinuation at a constant rate (Only 8 of 86 subjects who started with lottery B in the beginning of the investment plan switch multiple times between the two choices. The remaining 78 subjects either stay with B, or stay with A, once they switched from B to A). Figure 4 shows the persistence rate over the course of the 12 periods in the investment plan. Similar to clinical findings, we observe that persistence decreases over time with different intensity. In the beginning of the investment plan, the persistence rate is well above 90% and decreases only slightly to roughly 85% in period 7. From period 7 onwards, the likelihood of observing a subject discontinuing with therapy rises continuously, with many subjects starting to switch from lottery B to lottery A, despite still being the risk-dominant choice of staying with lottery B until period 10. Indeed, if we look at the decisions of each subject over the course of action, we find strong support for the predictions of our conceptual framework. Table 3 illustrates the behavioral pattern of each individual from period t to the next consecutive period t þ 1. Apart from the first period, there is no significant change in individual choice behavior over two consecutive periods until period 6. Consistent with the

819

VALUE IN HEALTH 17 (2014) 814–822

Fig. 4 – Persistence rate in the experimental investment setting.

Fig. 5 – Kaplan-Meier estimates of persistence curves between the baseline experiment (BASE) and rational choice behavior.

characteristics of the assumed value function and the current wealth level, subjects prefer the less risky investment option and choose lottery B to compensate for the endured losses from the working stage. Recall from the parameterization of the experiment that subjects start the lottery stage with a loss of 6 talers and subjects’ income increases with each choice for lottery B by 1 taler. In line with the empirical findings of Thaler and Johnson [43], subjects in period 6 prefer the more risky lottery A because they have the opportunity to reach the domain of gains with the next choice and are therefore extremely risk-seeking in this situation. From period 7 onwards, subjects are in the domain of gains. The underlying value function has a concave shape in that domain, so that the marginal increase of one additional monetary unit of any further gain becomes less valuable, the closer subjects come to the end of the investment plan. The consequences for the assessment of the cost-benefit trade-off of each lottery is that subjects value the revenues of lottery B with each additional period lower and therefore subsequently switch to choose the more risky lottery A. This behavior is apparent in the steep decrease in persistence in the interval between the periods 7 and 12. Further econometric analysis supports these nonparametric results. For this purpose, we estimate a univariate logistic regression with fixed effects and use the choice of the lotteries as the dependent variable (coding the choice of A ¼ 0 and B ¼ 1). If we further include a dummy variable as independent variable, which takes the value of 0 for the periods 1 to 6 and the value of 1 for the periods 7 to 12, we find that the probability of choosing B in the periods 7 to 12 is significantly 45 percentage points lower than in periods 1 to 6 (P ¼ 0.00; robust standard errors [bootstrapped] ¼ 0.01982).

Comparison with Rational Choice Behavior The experimental findings so far suggest that subjects decide according to the concept of mental accounting and evaluate the costs and benefits of continuing with therapy consistent with the predictions of our conceptual framework. It can be argued, however, that the same observed behavioral pattern could have emerged from rational choice theory. To address this question, we compare the persistence rate curve of the experiment (which we call BASE for the sake of clarity) with a hypothesized persistence curve derived from rational choice theory. Given the same dropout rates, we simply assume that the same subjects in our analysis behaved completely rational and chose lottery B until period 10 and then chose lottery A in the periods 11 and 12. Figure 5 shows the persistence curves of the baseline experiment [BASE] and of the “as-if” rational choice subjects as Kaplan-Meier estimates. We investigate whether the potential difference in persistence rates is statistically significant by using the log-rank test, which is commonly used in survival analysis when censored data are present [54]. The test clearly rejects the null hypothesis that the two persistence curves are statistically the same (logrank χ2 ¼ 5.91, 2-sided, P ¼ 0.015). Hence, our observed data cannot be explained by rational choice behavior.

Discussion Our conceptual framework provides a behavioral economic model to increase the understanding of medication persistence and allows the design of mechanisms for effective intervention strategies. In this framework, we argue that patients follow the

Table 3 – Nonparametric McNemar test to assess the difference in choice behavior from one period to the next period. Consecutive periods McNemar χ2 (df ¼ 1)

Period 1 vs. 2

Period 2 vs. 3

Period 3 vs. 4

Period 4 vs. 5

8.00 (0.00)

1.29 (0.26)

0.33 (0.56)

0.00 (1.00)

6.00 (0.11)

0.33 (0.56)

Period 7 vs. 8

Period 8 vs. 9

Period 9 vs. 10

Period 10 vs. 11

Period 11 vs. 12

3.57 (0.06)

7.00 (0.00)

3.57 (0.06)

11.00 (0.00)

5.00 (0.03)

McNemar χ2 (df ¼ 1) Asymptotic P values in parentheses.

Period 5 vs. 6

Period 6 vs. 7

820

VALUE IN HEALTH 17 (2014) 814–822

principles of mental accounting and change the evaluation of expected benefits and costs of medication intake over time. To isolate the economic drivers of the underlying decision-making process, we abstract from any medical context and transfer the key components of a conventional medical treatment into an equivalent economic investment setting. The decision environment in the investment setting differs from the medication intake only in the fact that individuals are confronted with economic goods rather than medication products. Framed as a simple investment game over discrete time, we design a laboratory experiment in which participating student subjects make repeated decisions on two investment options with different risk profiles and outcome probabilities. The two investment options represent the economic equivalent of either continuing or discontinuing with therapy. Consistent with the predictions of our conceptual framework, subjects continue with therapy as long as experienced utility losses have to be compensated. As soon as previous losses are evened out, subjects perceive the marginal benefit of compliance lower than in the beginning of the treatment. Consequently, subjects start to discontinue with therapy. It is evident from observed behavior and revealed preferences in the laboratory that subjects’ choices are not aligned with rational choice theory. Instead, loss aversion and the diminishing sensitivity of future gains account for the increasing tendency of subjects during the course of a treatment to discontinue with therapy at a specific point in time. Incorporated in our conceptual framework, concepts of bounded rationality and behavioral economics seem to capture the dynamic nature of medical nonpersistence much better than do standard economic models and generate more accurate predictions about how patients behave under circumstances characterized by risk and uncertainty. In a similar vein, Volpp et al. [23,25,55,56] provide evidence throughout a series of different medical treatments that behavioral economics explain certain behavioral patterns of patients. Those experiments are using among others tied payments, lottery-based incentives, lotteries and deposit contracts, social incentives, regret contest incentives, hovering, or decision errors [23–28,56–58]. Transferred to the medical context, our findings might explain persistence curves observed in observational studies. High persistence rates typically observed in the beginning of a medical treatment might be the result of patients’ reaction to the efforts and costs incurred in the beginning of a new medication and therapy. For example, new intake procedures have to be learnt and adverse effects evolve but no benefits in the form of health state improvements cover these costs in this phase of the treatment. When the medicine starts being efficacious, patients want their previous investments in their health to be compensated and continue with medication intake because this is the least risky action to compensate the endured losses. This behavior goes on until patients have the impression that their previous efforts have been completely compensated. The economic evaluation of medication intake then starts from another status quo and benefits of future persistence are valued lower than in the past. Consequently, patients can maintain the same utility only by reducing the costs and therefore discontinue with therapy. With the use of methods from experimental economics, we provide a link between medical decision making and health economics. Despite tremendous improvements to set a clear taxonomy for the measurement of persistence and to establish rigor standards for clinical and retrospective studies accordingly, methodological issues and heterogeneity across existing studies make it almost impossible to derive determinants of patients’ preferences and allow for testing hypotheses on patients’ behaviors over the course of the medical treatment. By controlling the rules of the game in a laboratory experiment that includes defining choice sets, information conditions, and payoff structures, we can establish an environment that captures the

assumptions of the economic model or the real-world situation for a direct empirical test of the inherent predictions and systematically investigate the factors that might affect the individual’s decision-making process. This advantage is especially noteworthy when applying the prospect theory to health-related behavior. Previous attempts in that respect report mixed results because of hypothetical situations and biased perception of patients who are not always able to imagine themselves in health states different from their own [59,60]. The fact that this study was conducted in the laboratory sets some limitations to our work and warrants discussion. First, university students who were recruited because of convenience and cost reasons may not be the most appropriate subject pool for studying medication persistence. They are on average younger than medical patients and likely have not experienced severe illnesses or chronic diseases. However, because the experiment is not framed in a medical context and rather economic reasoning such as evaluating the benefits, costs and the risk of investment options are investigated, this limitation may not be that severe. Second, we created a very critical test for validating the predictions of the conceptual framework and therefore chose a special set of parameters with extreme risk values and high dropout rates to maximize the saliency of the difference in risk of the two investment options. Inducing loss aversion through different working tasks, extending the number of periods in the investment plan, and variation in the cost-benefit structure of the investment options, along with different risk profiles to account for different dropout rates, are only some suggestions for a proper sensitivity analysis that should be used as further robustness of the current findings. Although these steps aim to increase the internal validity of the conceptual framework, questions about its external validity must be addressed. Subjects made their decisions in an artificial environment for a limited time frame. Although several experimental studies provide evidence that students’ decisions over abstract goods are not necessarily that much different from those of the target population and their commodities under interest [61– 65], we can only speculate to what extent the observed effects prevail outside laboratory walls. Future research in running experiments with a medical subject pool is therefore needed to increase the external validity of our findings. The need for improved efficiency in health care delivery and thus the need for an evidence-based, outcomes-focused, and behavior-driven model is asking for more emphasis being put on gathering evidence to identify the interventions that are most effective at improving health outcomes, and then realigning the behaviors of all stakeholders—patients, providers, manufacturers, and others—around these interventions. The conceptual model and economic experiments explaining nonpersistence and enabling mechanism designs to deliver behavioral change is a necessary basis for the realignment. For implementation, the conceptual model has to be specified for specific patient populations of interest to identify the influence of potential covariates such as the type of disease, personality traits, and personal experience on the individual hazard of medical nonpersistence. An improved understanding of under what circumstances patients start to discontinue with therapy is pivotal for scalable application in real-world practice: The result of such research would provide the components of a screening device to associate patient- and therapy-related factors to expected behavior. It allows identifying the behavioral segments (described by patient characteristics, disease and disease state, and potential interventions) in which behavioral incentives can be expected to be most effective in getting patients to adopt behaviors enabling therapeutic interventions to be effective. Furthermore, it allows predicting the behavior and thus effective health outcomes of combinations of medical intervention and combinations of medical and behavioral interventions in different behavioral segments.

VALUE IN HEALTH 17 (2014) 814–822

This is inevitable in designing managed access agreements and in assessing the potential benefit and risks of alternative behavioral interventions and their ongoing performance management. Finally, behavioral incentives can be developed to be more targetorientated on observed behavior and effectively calibrated to support patients in optimal medication-taking behavior. Selecting the most effective behavioral levers for individual patient segments and optimal intensity of the incentives is the basis of efficient and effective behavioral interventions. For broad application in realworld practice, those mechanism designs have to be translated into contractual design components in the context of the possible patient-insurance contracts of a health plan or health insurance. The utilization of a behavioral laboratory and the selection of a student sample for this study are giving further direction, insights, and a proof point for the development of a scalable process to design and calibrate targeted behavioral incentives while minimizing patient research to the maximum extent possible.

Conclusions Medical nonpersistence is a worldwide problem of striking magnitude. We augment the research to its causes by explaining medical nonpersistence from an economics perspective. To understand why patients acting as economic agents tend to discontinue the therapy at a certain point in time, we propose a conceptual framework that is built on the principles of behavioral economics. After having transferred the key features of a medical treatment to an equivalent economic investment setting, we design and conduct a randomized economic laboratory experiment with a student subject pool to empirically validate the rich behavioral predictions of our conceptual framework. In line with the predictions, mental accounting explains the observed behavioral patterns and emphasizes that a combination of loss aversion and different reference points changes behavior from formerly being compliant toward discontinuing with therapy. Incorporating these insights from behavioral economics into possible intervention strategies promises to be as beneficial as medical innovations and should be a mandatory part of the development of any intervention aimed at improving patients’ compliance and persistence behavior.

Acknowledgments We thank Helena Müller and Marina Bulgarowski for valuable research assistance. Their assisting help in organizing and conducting the experiments at the University of Paderborn is gratefully acknowledged. Source of financial support: Funding for this research was provided by Novartis Pharma AG, Basel, Switzerland.

Supplemental Materials Supplementary material accompanying this article can be found in the online version as a hyperlink at http://dx.doi.org/10.1016/j. jval.2014.08.2669 or, if a hard copy of article, at www.valuein healthjournal.com/issues (select volume, issue, and article).

R EF E R EN CE S

[1] Cleemput I, Kesteloot K, DeGeest S. A review of the literature on the economics of noncompliance: room for methodological improvement. Health Policy 2002;59:65–94.

821

[2] Cramer JA, Roy A, Burrell A, et al. Medication compliance and persistence: terminology and definitions. Value Health 2008;11:44–7. [3] World Health Organization. Adherence to long-term therapies— evidence for action. Available from: http://www.who.int/chp/ knowledge/publications/adherence_full_report.pdf. [Accessed July 11, 2013]. [4] Bourgault C, Sénécal M, Brisson M, et al. Persistence and discontinuation patterns of antihypertensive therapy among newly treated patients: a population-based study. J Hum Hypertens 2005;19:607–13. [5] Cramer JA, Benedit A, Muszbek N, et al. The significance of compliance and persistence in the treatment of diabetes, hypertension and dyslipidaemia: a review. Int J Clin Pract Suppl 2008;62:76–87. [6] Dunbar-Jacob J, Erlen JA, Schlenk EA, et al. Adherence in chronic disease. Annu Rev Nurs Res 2000;18:49–90. [7] van Wijk BLG, Klungel OH, Heerdink ER, de Boer A. Rate and determinants of 10-year persistence with antihypertensive drugs. J Hypertens 2005;23:2101–7. [8] DiMatteo RM, Giordani PJ, Lepper HS, Croghan TW. Patient adherence and medical treatment outcomes – a meta-analysis. Med Care 2002;40:794–811. [9] Vermeire E, Hearnshaw H, van Royen P, Denekens J. Patient adherence to treatment: three decades of research. A comprehensive review. J Clin Pharm Ther 2001;26:331–42. [10] Bishop MM, Rodrigue JR, Wingard JR. Mismanaging the gift of life: noncompliance in the context of adult stem cell transplantation. Bone Marrow Transplant 2002;29:875–80. [11] Dunbar-Jacob J, Mortimer-Stephens MK. Treatment adherence in chronic disease. J Clin Epidemiol 2001;54:57–60. [12] Hughes DA, Bagust A, Haycox A, Walley T. The impact of noncompliance on the cost-effectiveness of pharmaceuticals: a review of the literature. Health Econ 2001;10:601–15. [13] Siiskonen SJ, Breekveldt-Postma NS, Vincze G, et al. Higher persistence with valsartan compared with enalapril in daily practice. Vasc Health Risk Manag 2007;3:1039–44. [14] Sokol MC, McGuigan KA, Verbrugge RR, Epstein RS. Impact of medication adherence on hospitalization risk and healthcare costs. Med Care 2005;43:521–30. [15] Towers Watson/National Business Group on Health: Performance in an era of uncertainty. United States 2012 17th Annual Towers Watson/ National Business Group on Health Employer Survey on Purchasing Value in Health Care. Arlington, VA: Towers Watson/National Business Group on Health, 2012. [16] McDonald HP, Garg AX, Haynes RB. Interventions to enhance patient adherence to medication prescriptions – scientific review. JAMA 2012;288:2868–79. [17] Nosé M, Barbui C, Gray R, Tansella M. Clinical interventions for treatment non-adherence in psychosis: meta-analysis. Br J Psychiatry 2003;183:197–206. [18] Osterberg L, Blaschke T. Adherence to medication. N Engl J Med 2005;353:487–97. [19] Kripalani S, Yao X, Haynes B. Interventions to enhance medication adherence in chronic medical conditions. Arch Intern Med 2007;167:540–50. [20] Marteau T, Ashcroft R, Oliver A. Using financial incentives to achieve healthy behaviour. BMJ 2009;338:983–5. [21] DeFulio A, Silverman K. The use of incentives to reinforce medication adherence. Prev Med 2012;55:86–94. [22] Elliot RA, Shinogle JA, Peele P, et al. Understanding medication compliance and persistence from an economics perspective. Value Health 2008;11:600–10. [23] Loewenstein G, Brennan T, Volpp KG. Asymmetric paternalism to improve health behaviors. JAMA 2007;298:2415–7. [24] Long JA, Larsen-Helweg M, Volpp KG. Patient opinions regarding pay for performance for patients. J Gen Intern Med 2007;23:1647–52. [25] Volpp KG, Loewenstein G, Troxel AB, et al. A test of financial incentives to improve warfarin adherence. BMC. Health Serv Res 2008;272:1–6. [26] Volpp KG, Troxel AB, Pauly MV, et al. A randomized, controlled trial of financial incentives for smoking cessation. N Engl J Med 2009;360:699–709. [27] Long JA, Jahnle EC, Richardson DM, et al. Peer mentoring and financial incentive to improve glucose control in African American veterans: a randomized, controlled trial. Ann Intern Med 2012;156:416–24. [28] Kullgren JT, Troxel AB, Loewenstein G, et al. Individual-versus groupbased financial incentives for weight loss: a randomized, controlled trial. Ann Intern Med 2013;158:505–14. [29] Ellickson P, Stern S, Trajtenberg M. Patient welfare and patient compliance—an empirical framework for measuring the benefits from pharmaceutical innovation. In: Cutler DM, Berndt ER, eds., Medical Care Output and Productivity. Chicago: University of Chicago Press, 2001. [30] Camerer CF, Loewenstein G. Behavioral economics: past, present, future. In: Camerer CF, Loewenstein G, Rabin M, eds., Advances in Behavioral Economics. Princeton: Princeton University Press, 2003.

822

VALUE IN HEALTH 17 (2014) 814–822

[31] Mihalko SL, Brenes GA, Farmer DF, et al. Challenges and innovations in enhancing adherence. Control Clin Trials 2004;25:447–57. [32] Dezii CM. Persistence with drug therapy: a practical approach using administrative claims data. Manag Care 2001;10:42–5. [33] Yu YF, Nichol MB, Yu AP, Ahn J. Persistence and adherence of medications for chronic bladder/urinary incontinence in the California Medicaid Program. Value Health 2005;8:495–505. [34] Camerer CF, Malmendier U. Behavioral economics of organizations. In: Diamond P, Vartiainen H, eds., Behavioral Economics and Its Applications. Princeton: Princeton University Press, 2007. [35] Kahneman D. Maps of bounded rationality: psychology for behavioral economics. Am Econ Rev 2003;93:1449–75. [36] Munro A. Bounded Rationality and Public Policy—A Perspective from Behavioural Economics. Dordrecht, The Netherlands: Springer Science þ Business Media, 2009. [37] Simon HA. Theories of bounded rationality. In: McGuire CB, Radner R, eds., Decision and Organization. Amsterdam, The Netherlands: NorthHolland Publishing Company, 1972. [38] Thaler RH. Mental accounting and consumer choice. Market Sci 1985;4:199–214. [39] Thaler RH. Mental accounting matters. J Behav Decis Mak 1999;12:183–206. [40] Levav J, McGraw AP. Emotional accounting: how feelings about money influence consumer choice. J Mark Res 2009;46:66–80. [41] Read D, Loewenstein G, Rabin M. Choice bracketing. J Risk Uncertainty 1999;19:171–97. [42] Kahneman D, Tversky A. Prospect theory: an analysis of decision under risk. Econometrica 1979;47:263–92. [43] Thaler RH, Johnson EJ. Gambling with the house money and trying to break even: the effects of prior outcomes on risky choice. Manage Sci 1990;36:643–60. [44] Grossman M. The Human Capital Model of the Demand for Health. Cambridge: National Bureau of Economic Research, 1999. [45] Friedman D, Sunder S. Experimental Methods—A Primer for Economists. Cambridge: Cambridge University Press, 1994. [46] Guala F. The Methodology of Experimental Economics. Cambridge: Cambridge University Press, 2005. [47] Croson R. Why and how to experiment: methodologies from experimental economics. U Illinois Law Rev 2002;4:921–45.

[48] Falk A, Fehr E. Why labour market experiments? Labour Econ 2003;10:399–406. [49] Amiel Y, Cowell F. Measurement of income inequality: experimental tests by questionnaire. J Public Econ 1992;47:3–26. [50] Falk A, Heckman JJ. Lab experiments are a major source of knowledge in social sciences. Science 2009;326:530–8. [51] Bleichrodt H, Doctor J, Stolk E. A nonparametric elicitation of the equity-efficiency trade-off in cost-utility analysis. J Health Econ 2005;24:655–78. [52] Gill D, Prowse V. A structural analysis of disappointment aversion in a real effort competition. Am Econ Rev 2012;102:469–503. [53] Fischbacher U. z-Tree: zurich toolbox for ready-made economic experiment. Exp Econ 2007;10:171–8. [54] Lee ET. Statistical Methods for Survival Data Analysis. New York: John Wiley, 1992. [55] Volpp KG, Gurmankin Levy A, Asch DA, et al. A randomized, controlled trial of financial incentives for smoking cessation. Cancer Epidem Biomar 2006;15:12–8. [56] Volpp KG, John LK, Troxel AB, et al. Financial incentive–based approaches for weight loss—a randomized trial. JAMA 2008;300:2631–7. [57] Asch DA, Muller R, Volpp KG. Automated hovering: watching over the other 5,000 hours. N Engl J Med 2012;367:1–3. [58] Volpp KG, Pauly MV, Loewenstein G, Bangsberg D. P4P4P: an agenda for research on pay for performance for patients. Health Aff 2009;28:206–14. [59] Feeny D, Eng K. A test of prospect theory. Int J Technol Assess Health Care 2005;21:511–6. [60] Happich M, Moock J, von Lengerke T. Health state valuation methods and reference points: the case of tinnitus. Value Health 2009;12:88–95. [61] Cooper J, Kagel JH, Wie L, Qing LG. Gaming against managers in incentive systems: experimental results with Chinese students and Chinese managers. Am Econ Rev 1999;89:781–804. [62] Falk A. Gift exchange in the field. Econometrica 2007;75:1501–11. [63] Carpenter J, Seki E. Do social preferences increase productivity? Field experimental evidence from fisherman in Toyama Bay. Econ Inq 2011;49:612–30. [64] Bleichrodt H, Rohde KIM, Van Ourti T. An experimental test of the concentration index. J Health Econ 2012;31:86–98. [65] Fehr E, List JA. The hidden costs and returns of incentives—trust and trustworthiness among CEOs. J Eur Econ Assoc 2004;2:743–71.