Meng Li1 and Gretchen B. Chapman2
1Department of Health and Behavioral Sciences, University of Colorado Denver, Denver, CO, USA
2Department of Psychology, Rutgers University–New Brunswick, New Brunswick, NJ, USA
Medical decision making encompasses decisions made in all health and medicine contexts by patients, healthcare workers, health policy makers, and the general public. The study of medical decision making follows research on judgment and decision making more generally: It compares actual or descriptive decision making to normative or rational models of decision making, where the latter provide a benchmark for how decisions should be made. It also explores interventions that can improve decision making or guide choice behavior in a particular direction (i.e., toward selection of a particular healthy option).
Medical decisions are often difficult because they entail choosing among options with uncertain outcomes; they require trade‐offs between competing goals, often with overwhelming amounts of information; they involve taking on immediate costs for long‐term benefits; they may also require strategic interactions among multiple agents. Below we characterize medical decision making by discussing some selective theoretical and empirical issues related to each of these topics.
Many medical decisions involve uncertain outcomes. For example, a patient must decide whether to undergo surgery, not knowing if the surgery will be successful. A physician must decide whether to recommend a preventive measure, such as cholesterol‐lowering medication, not knowing whether the patient would be fine even without the medication. According to normative expected utility theory, the utility of each outcome (e.g., stroke) should be weighted by the probability that the outcome will occur. Consequently, it is critical that decision makers be able to assess, at least implicitly, the probability of key outcomes and to weight the utility of outcomes in proportion to their probability. In a number of cases, however, outcomes are systematically over‐ or underweighted, and low likelihood events tend to be overweighted and thus have more influence on decisions than they should. This phenomenon is illustrated in the American response to the 2014 Ebola outbreak in West Africa. Although contracting Ebola in the United States was an extremely low likelihood event (only four cases of Ebola were diagnosed in the United States, only two of which were contracted in the United States), the response of some US public officials was indicative of a much higher level of risk. For example, in October 2014, the governors of New York and New Jersey announced a mandatory quarantine for any healthcare workers returning from treating Ebola patients in West Africa to the United States via a New York or New Jersey airport. It appears that decision makers draw a large distinction between an outcome with a 0% likelihood and one with a small positive likelihood. For example, people will pay more to reduce the risk of an adverse event from a toxin from 5 in 10,000 to 0 in 10,000 than they will pay for a risk reduction from 15 in 10,000 to 10 in 10,000, even though both are the same net reduction in risk (Viscusi, Magat, & Huber, 1987). Similarly, decision makers draw a large distinction between 100% and a proportion that is slightly smaller than 100%. Consequently, 100% of a small category can seem larger than 50% of a large category, whether the percentages refer to probability, effectiveness, disease coverage, or something else (Li & Chapman, 2009, 2013).
A medical decision can be especially difficult if it encapsulates a trade‐off between two important but opposing goals or attributes. Trade‐offs in medical decision making are prevalent, such as in the choice between quality and length of life in end‐of‐life decisions or in the design of screening programs that entail either greater false alarms or potential misses of true pathologies. According to normative multi‐attribute utility theory, each attribute should be weighted according to its importance, and an overall utility (goodness) for each option could be computed by summing the weighted utility of all attributes. However, real‐life decisions often violate this normative prescription, and the weight that people assign to attributes can shift depending on irrelevant factors, such as how the problem is described.
One often‐debated issue in health policy settings involves the trade‐off between efficiency and equality, such as in the allocation of scarce medical resources, including transplant organs, scarce vaccines, etc. For example, quality‐adjusted life year (QALY) is a widely used metric to assess benefits in medical options. The rationale behind the use of QALYs is that more life years saved represents more net benefit, regardless of whose QALYs are saved (QALYs entail younger people being prioritized, as saving them produces more QALYs).
People also hold another important goal in the allocation of medical resources: equality in receiving such resources. People place great value on equality, to the point that they are willing to save fewer lives to achieve equal allocation of lifesaving resources (Ubel, Baron, & Asch, 2001). However, the preference for equity at the expense of efficiency is also volatile and subject to change depending on how the question is posed. For instance, in the choice between younger and older lives when not all lives can be saved, people usually prefer to save younger lives, demonstrating a preference for efficiency (Li, Vietri, Galvani, & Chapman, 2010), but such preferences fluctuate systematically. When people read descriptions on allocation outcomes (e.g., 500 20‐year old people will be saved; 500 60‐year old people will be saved), they prefer to save younger individuals; however, when they read general principles on how different lives should be prioritized under scarcity (e.g., “younger people should be valued more,” “older people should be valued more,” or “all lives should be valued equally regardless of age”), people demonstrate greater preference for equal allocation across young and old recipients (Li, 2012).
Another example in which preference for efficiency versus equality can be shifted with a fairly subtle change in description comes from a study by Colby, DeWitt, and Chapman (2015). Participants allocated 6 transplant kidneys to 12 potential recipients, 6 of whom had a high chance of transplant success. When the 12 recipients were presented in one group, participants tended to allocate the kidneys efficiently, giving them to the 6 recipients most likely to benefit from them. However, when the 12 recipients were presented in 2 groups of 6 each, participants tended to spread the kidneys across the 2 groups, even though this meant sacrificing efficiency.
All decisions entail a comparison and selection among multiple options, even if some options are not salient (e.g., choosing nothing, doing nothing). But what happens when there are a large number of options available, accompanied by a large amount of information about them? This is the situation many Americans face after the recent ratification of the Patient Protection and Affordable Care Act (ACA), which allows them to select from among dozens of healthcare plans. Providing too many healthcare plans can decrease decision quality. For example, Johnson, Hassin, Baker, Bajger, and Treuer (2013) presented participants with premium, co‐pay, and deductible information for multiple health plans and found that participants frequently showed near‐chance levels of performance in selecting the lowest cost plan because they gave too much weight to out‐of‐pocket expenses and deductibles. Although Johnson et al.'s participants made hypothetical choices, other researchers who analyzed data from actual health plan choices also found that consumers frequently chose dominated health plans—plans that were more expensive but provided no better benefits than alternative plans—because consumers focused too heavily on deductibles (Bhargava, Loewenstein, & Sydnor, 2015). Thus, choosing from a large choice set makes for a difficult psychological task, sometimes resulting in the decision maker violating normative principles such as dominance. Therefore, medical decisions can be improved by simplifying choice sets or providing smart defaults to decision makers.
Frequently, medical decisions involve delayed outcomes. For example, a calorie‐dense diet and low levels of physical activity lead to increased risk of obesity and associated comorbidities such as diabetes and heart disease. However, these consequences are delayed, often by years or decades. Decision makers value delayed outcomes less than equivalent immediate outcomes; consequently, it is difficult for decision makers to forgo the immediate temptations of a piece of cheesecake or the TV episode instead of a trip to the gym. According to normative discounted utility theory, delayed outcomes should be discounted in a consistent manner, and thus people should not change their preferences between a larger later reward (e.g., the long‐term benefit of weight loss) and a smaller sooner reward (e.g., the enjoyment of a piece of cheesecake) at different time points, that is, if you decided that the long‐term benefit of weight loss is more important than the enjoyment of cheesecake that you can have soon, you should be able to stick to this decision at any given time. However, preference between these two types of rewards frequently changes with time: People often choose the larger later reward initially (such as on New Year's Day), when they are temporally distant from both rewards, but later fall for the temptation when it becomes available immediately (as the dessert tray rolls by). This inconsistent choice marks the typical pattern of self‐control failure.
As a solution to this problem, decision makers will sometimes take advantage of the opportunity to precommit or bind themselves to the earlier preference for the larger later reward. For example, Schwartz et al. (2014) invited grocery store shoppers to lay their money on the line to precommit to buying more healthy foods such as fresh produce. The shoppers agreed to forfeit the discount they already received for buying healthy foods if they did not meet their healthy foods purchase goal. Indeed, these shoppers increased their healthy food purchases relative to a control group who merely indicated hypothetically whether they would precommit. Similarly, dieters who do not keep desserts in the house are taking advantage of a precommitment device, as are people who prepay for a gym membership in the hopes of motivating themselves to work out frequently to “get their money's worth.” Another solution to the problem of discounted future outcomes is bundling temptation with a desired behavior, so one is allowed to indulge in a pleasure only when one engages in a necessary but unpleasant health behavior. For example, Milkman, Minson, and Volpp (2014) found that people visited the gym more often if they could listen to addictive audio books only when they were at the gym.
Some medical decisions are made in a strategic setting where the choices made by an individual affect the outcomes of others while at the same time the decisions of others affect the outcomes experienced by the individual. Take vaccination as an example. Because of herd immunity, an individual's vaccination protects not only herself from infection but also her social contacts, because she is now less likely to spread the virus to others. Consequently, if enough individuals in a population are vaccinated, the unvaccinated members are protected from infection. Thus, vulnerable individuals who cannot receive vaccination themselves (e.g., infants) can be cocooned in this fashion. Because of this positive externality from vaccination, an individual's decision about whether to vaccinate may be affected by how many others in the population are vaccinating (Galvani, Reluga, & Chapman, 2007). This strategic interaction can be modeled using game theory (Chapman et al., 2012): individuals can “defect” by getting a free ride, that is, forgoing vaccination because they receive protection from others who vaccinated, or they can “cooperate,” that is, vaccinating in part to protect those around them.
Free riding may contribute to the anti‐vaccination movement because those opposed to vaccination can forgo immunization without bearing much risk of contracting infectious disease, as long as the majority of the surrounding population is immunized. When critical numbers of those opposed to vaccination are clustered in one geographical region, however, herd immunity can dip below the threshold that is necessary to prevent the spread of disease, making the larger population vulnerable to an outbreak, such as the measles outbreak in a theme park in California in 2015. Other medical decisions that entail externalities include overuse of antibiotics, which does not affect the user negatively but contributes to the development of resistant strains of bacteria, and the over‐ordering of expensive medical tests for which the physician does not pay but which will drive up healthcare costs and insurance premiums.
Normative models of decision making assume that decision makers will make informed decisions, provided that they are in possession of the relevant facts. However, decision makers vary in the capacity to comprehend and use information to approach choices. For example, numeracy, or the capacity to process numerical information, is essential in medical decision making, as such decisions frequently involve numbers (e.g., risk probabilities, efficacy information, cost, time). Numeracy levels among Americans are worrisome, with many people, even those who are highly educated, are particularly lacking in the ability to process percentages and fractions. These basic numerical competencies are critical in the medical context, such as when one needs to calculate medication dosage based on a child's body weight. A recent review (Reyna, Nelson, Han, & Dieckmann, 2009) concluded that patients who have low numeracy are more likely to misinterpret graphs containing medical information, more easily influenced by factors irrelevant to the decision at hand (such as how the information is presented), and more likely to overestimate small risks and have poorer health outcomes when the health condition requires a high degree of self‐management.
Perhaps more disturbingly, medical professionals do not necessarily possess the level of numeracy that is required to correctly understand risk information. And the lack of capacity to comprehend risk information can incur serious consequences. For example, despite the commonly held belief that cancer screening saves lives, some cancer screening tests can produce a large number of false positives, encouraging overtreatment for nonprogressive cancer while yielding little or no life‐saving benefit, and inflating the 5‐year survival rate. Therefore, to evaluate the effectiveness of a screening test in saving lives, a reduced mortality rate should be used as the gold standard, instead of improved survival rates. Unfortunately, a study by Wegwarth, Schwartz, Woloshin, Gaissmaier, and Gigerenzer (2012) showed that about half of physicians do not share this knowledge and instead incorrectly believe that increased early detection constitutes evidence that such screening saves lives. This lack of numerical competency among physicians may have contributed to the ill‐deserved enthusiasm toward cancer screening tests and the resulting financial and health toll of their overuse.
Given the myriad ways in which actual medical decisions deviate from normative models of decision making, researchers and clinicians alike consider what can be done to improve medical decisions. Decision psychologists and behavioral economists have recently developed nudges (Thaler & Sunstein, 2008), or changes to the decision environment that facilitate selection of the healthy option, while maintaining freedom of choice. One well‐known nudge is the default effect, or the tendency for people to stick with the option they will get automatically if they take no explicit action. For example, people prescheduled for a flu shot appointment (which they can cancel if they do not want it) are more likely to get vaccinated than are those who are not prescheduled (Chapman, Li, Colby, & Yoon, 2010). Although both groups have a choice to have a flu shot appointment or not, for the prescheduled group, the default is having a flu shot appointment, and for the comparison group the default is not having an appointment. Because most people tend to stick with their default status, and because having a flu shot appointment is a strong predictor of actually getting a flu shot, vaccination rates increased by 36% (from 33 to 45%) in the group with a default vaccination appointment relative to the group where an appointment has to be actively scheduled.
There are other types of nudges toward healthy behavior. For example, because people tend to conform their behavior to that of the group, giving people information about what others do will nudge them toward that same behavior. Social norms have been used to reduce caloric intake (by printing recommended calorie intake on menus, combined with menu calorie labeling; Roberto, Larsen, Agnew, Baik, & Brownell, 2010), to increase children's vegetable intake at school lunch (by placing photographs of vegetables on wells of the lunch tray; Reicks, Redden, Mann, Mykerezi, & Vickers, 2012), and to encourage walking (by informing participants who wear pedometers how much others are walking; Chapman, Colby, Convery, & Coups, 2015). Another example of a healthy nudge is the position effect, in which food options displayed at prominent, easy‐to‐reach positions are chosen more frequently. Researchers have used the position effect to encourage healthy eating in cafeterias and restaurants (e.g., Rozin, Scott, & Dingley, 2011).
In summary, medical decision making includes not only decisions that are made in strictly medical settings but also decisions made in everyday life that impact health. Research on medical decision making can inform health professionals as well as the general public about the pitfalls in medical decisions, how to overcome them, and how to use decision biases to the advantage of people's health.
Meng Li is an assistant professor in the Department of Health and Behavioral Sciences at the University of Colorado Denver. Her work, which has been featured on JAMA, Lance, Psychological Science, and other top journals, utilizes decision biases to “nudge” people toward healthy behavior, such as vaccination, healthy diet, and hand sanitizer use. Her more recent work explores policy issues, including resource allocation on money and health, price transparency in healthcare, and work–life balance policies.
Gretchen B. Chapman is a professor of psychology and a member of the Institute for Health and the Center for Cognitive Science at Rutgers University. Her research combines judgment and decision making with health psychology to examine preventive health behaviors such as vaccination. She conducts field studies to evaluate decision theoretic interventions designed to encourage health behavior and laboratory research to examine basic processes in decisions under uncertainty, strategic interactions, and allocation of scarce resources.