Are More Options Always Better? The Attraction Effect in Physicians’ Decisions about Medications JANET A. SCHWARTZ, GRETCHEN B. CHAPMAN, PhD Consumer choice research has shown that, contrary to normative theory, the introduction of an inferior alternative to an existing choice set can increase the likelihood that one of the original alternatives will be chosen. This phenomenon, the attraction effect, is relevant to physician decision making, particularly when the physician is in the role of a consumer who must make decisions about prescribing medications when a number of alternatives are available. To investigate the attraction effect in physician decision making, 40 internal medicine residents reviewed three patient cases (concerning depression, sinusitis, and vaginitis) and then chose the most appropriate medication for each patient. In some versions of the cases, two medication options were available. Other versions included a third medication (the decoy) that was inferior in every way to one of the original options (the target) but not to the other (the competitor). The results showed that addition of the ‘‘decoy’’ medication increased the likelihood of choosing the target medication. That is, the attraction effect does occur in physicians’ decisions about medications. Physicians should be aware of this bias when evaluating or suggesting several similarly attractive medications or treatment options for the same medical condition. Key words: attraction effect; decision making; physicians’ decisions; consumer choice. (Med Decis Making 1999;19:315–323)
When making treatment decisions, physicians often face choices among several medications, all of which are safe and effective. In this sense, physicians are similar to many other consumers who often make decisions among several comparable alternatives. One aspect of consumer behavior that is relevant to physician decision making is that the number of options available is ever-increasing. In fact, research investigating consumer choice behavior has identified a number of decision-making biases that occur as a result of increasing the number of alternatives in a choice set.1 Given the occurrence of those biases, it seems wise to investigate them in consumer-oriented areas of medical decision making, such as choosing among several medications or treatment options.
Normatively, preference for one alternative in an existing choice set should not be increased by the addition of a new alternative2; to do so would be a violation of the regularity principle. However, numerous studies1,3-5 in the consumer-choice literature point to examples in which the addition of a third alternative to an existing two-option choice set can lead to violations of normative principles such as regularity. For example, in one study,1 participants in one condition were given a choice between buying a popular CD player on sale for that day only, or deferring the choice to learn more about the product. The participants in a second condition were given those same two alternatives plus a new alternative: another attractive CD player (with slightly better sound quality and a slightly higher price), also on sale for that day only. The participants were more likely to choose the ‘‘deferred’’ choice option, or maintain the status quo, in the latter condition than in the former. The authors suggested that adding another similarly attractive CD player to the choice set increased the amount of conflict felt by the participants. That is, when only one CD player was being considered it seemed like a good value and the option of delaying the purchase did not seem justified. However, when two similarly attractive CD players were being considered there was no justifiable reason to choose
Received August 24, 1998, from the Psychology Department, Rutgers University, Piscataway, New Jersey. Revision accepted for publication February 26, 1999. Presented at the annual meeting of the Society for Medical Decision Making, October 1998, Boston, Massachusetts. Supported in part by grants from the Agency for Health Care Policy and Research and the National Science Foundation to the second author. Address correspondence and reprint requests to Ms. Schwartz: Psychology Department, Rutgers University, 152 Frelinghuysen Road, Piscataway, NJ 08854-8020; phone: (732) 4451289; fax: (732) 445-2263; e-mail: ^
[email protected]&. 315
316
●
Schwartz, Chapman
one over the other. Since one CD player was superior in terms of price while the other CD player was superior in terms of brand-name reputation, the decision maker was put in the position of making a difficult tradeoff between reputation and price. Making that tradeoff might have been even more difficult if the decision maker had had the same utility for reputation and price. The option of deferring the choice became attractive because it allowed the participants to avoid the tradeoff. In a related study, Redelmeier and Shafir6 found a tendency to maintain the status quo in physicians’ decision making. Like the subjects of the abovementioned study, physician participants read a patient history, after which they were asked to decide what medication, if any, to prescribe for a patient who had been referred to a specialist. In one condition the physicians’ choices were ibuprofen or no medication. In the other condition, another medication (piroxicam) was added to the choice set. Significantly more physicians chose to write no prescription when two medications were available than when only one medication was available. These results were consistent with the findings of the earlier consumer choice study,1 and highlight the fact that cognitive biases resulting from increasing the number of alternatives in a choice set can affect patient care. Given that physicians’ behaviors may be influenced by the numbers of treatment alternatives available, it is of interest to know whether other biases occur in physicians’ decision making. Specifically, the purpose of the present study was to determine whether the attraction effect,7 a bias commonly found in consumer-choice studies, would also occur in physicians’ decision making. Like the phenomenon examined by Redelmeier and Shafir,6 in the attraction effect the addition of a third alternative to a choice set influences preferences for the two original options. However, in this case the additional (third) option does not strengthen preference for inaction or the status quo (e.g., the decision not to write a prescription), and the third option is not an alternative that would itself be chosen. For example, imagine that there are two medications available to treat a certain condition. Medication X is highly effective, but reportedly has many side effects. Medication Y is moderately effective but has few side effects. Now imagine that a third medication, Z, is added to the list of potential medications. Medication Z (the decoy) is moderately effective and it also has a moderate amount of side effects. Obviously, no one would ever choose medication Z over medication Y, because Z has more side effects than Y but is no more effective. The addition of Z to the choice set, however, increases the likelihood that medication Y will be chosen.
MEDICAL DECISION MAKING
The attraction effect is another example of a regularity principle violation, because the addition of an inferior alternative to a choice set increases the likelihood that an existing option will be chosen. It is also referred to as the asymmetric dominance effect because the decoy (e.g., medication Z) is dominated by the target (e.g., medication Y), but not by the competitor (e.g., medication X). One explanation for the occurrence of the effect is similar to that suggested by Tversky and Shafir.1 The fact that one alternative clearly has an advantage over another might help people make a justifiable decision.8 In other words, when considering only the target and the competitor, the decision is difficult because both options are attractive and the decision involves tradeoffs. However, when the decoy is introduced, the decision maker can identify a clear instance in which one alternative in the choice set is better than another. Thus, choosing the target might offer an ‘‘easy out.’’ Although a number of studies have demonstrated the attraction effect in consumer decision making, it has never been investigated in physician decision making. The attraction effect might be particularly relevant to physicians’ decisions about medications because of the way that medications are listed in reference guides. In the Physician’s Desk Reference,9 for example, the effectiveness of medications and the symptoms associated with them are often reported as percentages, similar to the format used in the present experiment. In a multiattribute comparison of several effective medications described similarly, it is possible that physicians might avoid making difficult tradeoffs by choosing a medication because it dominates an alternative. This point becomes particularly salient when we consider that there are numerous medications available to treat a single condition, and that number is ever increasing. The attraction effect may represent a heuristic used to simplify decision making that involves large choice sets. To determine whether the attraction effect occurred in physicians’ decision making, we presented internal medicine residents with three patient cases. The scenarios described the symptoms and histories of patients with three common diagnoses: depression, sinusitis, and vaginitis. The participants were asked to rate and choose medications based on each drug’s performance with respect to the following attributes: efficacy and potential side effects in the depression and sinusitis scenarios, and length of treatment and potential side effects in the vaginitis scenario.* *Efficacy and side effects are common attributes on which medications differ and were therefore used in two of the scenarios. Length of treatment was used as one of the attributes in the vaginitis scenario because it represented a salient difference among the medications in this class.
VOL 19 / NO 3, JUL–SEP 1999
Subjects and Procedure The participants in this questionnaire study were internal medicine residents (PGY1–3). Questionnaires were placed in the participants’ hospital mailboxes. Of the 72 participants originally contacted, 40 (28 male, 11 female, 1 not indicating gender; average age 28.2 years) completed and returned the questionnaires (56% response rate). The following reminder system was used: two weeks following the initial disbursement of questionnaires, postcard reminders were put in the mailboxes of respondents who had not returned the questionnaire. Two weeks after the postcard reminders were sent, full-copy reminders were placed in the mailboxes of those who still had not responded. This reminder system was repeated up to four times. Participation in this study was voluntary and the subject’s participation status was not revealed to medical school staff (i.e., chief resident, instructors, etc.). Responses were separated from identifying information prior to analysis. All of the participants were paid for their time. Each questionnaire contained three patient case scenarios: depression, sinusitis, and vaginitis. The scenarios had been reviewed by a professor of internal medicine and the chief resident to ensure clinical realism. The order in which the cases were presented and the position of the medication options (e.g., target, competitor, decoy) were counterbalanced using a Latin square design. Each scenario could appear in one of three conditions: a two-option control condition (no decoy), an A target condition, and a B target condition. That is, each of the medications in the two-option choice set appeared in both experimental decoy conditions, once as the target and once as the competitor. This design allowed us to control for potential confounders such as brand-name allegiance or drug familiarity. If familiarity were the driving force behind the choice of a particular medication, then one would expect to find that drug preferred to the others in all three conditions, including the decoy condition, where it was the competitor. Each questionnaire contained one scenario in each of the three conditions, and across all subjects, each scenario appeared in all three conditions. The participants were instructed to read each patient case and then review the list of medications that followed. The medication descriptions were abstracted from the Physician’s Desk Reference,9 with only minor adjustments made to fit the needs of the experimental design. In some cases, we manipulated the dominance relationships among the medications using the description of the patient’s presentation and medical history in order to maintain the accuracy of the medication descriptions. The same physicians who reviewed the case scenarios
Decisions about Medications
●
317
also reviewed the medication descriptions to ensure their accuracy and plausibility. The participants were told to imagine that the medications shown were the only options available. Presenting a limited-choice set of only two or three medications involved more constraints than are usually encountered in real prescribing decisions, but did not seem unreasonable given the limitations often imposed by insurance formularies, patient histories, and potential drug interactions. After reviewing the table of medications, each participant rated the appropriateness of each drug on a scale of 1 to 5, where 1 was least appropriate and 5 most appropriate, and then chose the medication that he or she would prescribe. Finally, the participant was told to imagine that the attending physician had questioned his or her choice and to offer a defense of that choice.
DEPRESSION SCENARIO
In the depression case, the residents read the following: The patient is a 56-year-old female. She was widowed six weeks ago, after watching her husband suffer from an illness for several months. During a routine physical examination, the patient reports that for the past three weeks she has had difficulty sleeping, diminished appetite, increased anxiety, and a lack of interest in previously enjoyed activities. She also reports that weekly appointments with her therapist, which were of great help during her husband’s illness, have not been helping lately. Other than these symptoms the patient has no other medical conditions, nor does she have a psychiatric history. Your diagnosis is new-onset depression. The patient is an appropriate candidate for antidepressant medication. Please read the medication options listed below and assume that they are the only options available to you. Her insurance will cover the cost of any prescriptions after she satisfies the usual $3.00 co-pay.
Table 1 presents the medication options that were available and their descriptions for this scenario. In the two-option condition, paroxetine HCl and nefazodone HCl were shown alone. In the three-option A condition, paroxetine HCl (superior with respect to side effects) was the target as defined by fluoxetine (the decoy). In the three-option B condition, nefazodone HCl (superior with respect to efficacy) was the target as defined by imipramine (the decoy). Each participant was presented with only one version of the three choice sets shown in table 1. In all three depression scenarios, paroxetine HCl and nefazodone HCl had the same description, and all of the participants read the same patient history. Table 2 shows the participants’ responses on the
318
●
Schwartz, Chapman
MEDICAL DECISION MAKING
Table 1 ● Medications in the Depression Scenario Medication
Effectiveness of Treatment†
Possible Side Effects
Nefazodone HCl (e.g., Serzone)
In clinical trials, nefazodone HCl has been shown to be superior to a placebo on 3 of 4 valid depression measures
25% of patients report dry mouth, nausea, and somnolence
Paroxetine HCl (e.g., Paxil)
In clinical trials, paroxetine HCl has been shown to be superior to a placebo on 2 of 4 valid depression measures
5% of patients report dry mouth, nausea, and somnolence
Imipramine HCl (e.g., Tofranil) *Decoy for nefazodone HCl
In clinical trials, imipramine HCl has been shown to be superior to a placebo on 3 of 4 valid depression measures
35% of patients report dry mouth, nausea, and somnolence; also, reports of abnormal ECG, hallucinations, and delusions
Fluoxetine HCl (e.g., Prozac) *Decoy for paroxetine HCl
In clinical trials, fluoxetine HCl has been shown to be superior to a placebo on 2 of 4 valid depression measures
15–20% of patients report dry mouth, nausea, and somnolence; 10% of patients report dream disturbances
† The depression measures used in clinical trials with these medications are: the Hamilton Depression Rating Scale, the Hamilton Depressed Mood Item, the Clinical Global Impression Severity of Illness Rating, and the Clinical Global Impression Improvement Scale.
two dependent measures (medication choice and appropriateness rating) for each medication in each of the three conditions. In the three-option A condition where paroxetine HCl was the target, the frequency with which people chose the target was not significantly greater than in the two-option condition (86% vs 75%), x2 (1, n = 30) = 0.536, power = 0.11. In other words, there was not a significant attraction effect in this condition, possibly because of a ceiling effect (a disproportionate number of subjects chose paroxetine HCl in the two-option condition) and low power due to the small sample size. The attraction effect was seen in the three-option B condition; significantly more participants chose nefazodone HCl when it was the target than when it was presented in the two-option condition (70% vs
25%), x2 (1, n = 23) = 6.512, p # 0.05, power = 0.72). We conducted a 2 3 3 (drug by condition) factorial analysis of variance (ANOVA) on the appropriateness ratings of paroxetine and nefazodone and found a significant drug-by-condition interaction F (1, 37) = 5.5, p # 0.05, power = 0.51, indicating the attraction effect. That is, the ratings of both paroxetine HCl and nefazodone HCl, relative to their mean ratings in the two-option condition, increased when each was the target. Only paroxetine HCl’s rating decreased when it was the competitor; nefazodone HCl’s rating increased slightly. However, on average, each drug was rated as most appropriate when it was the target and least appropriate when it was the competitor, indicating the attraction effect. SINUSITIS SCENARIO
In the sinusitis case the residents read the following:
Table 2 ● Choice and Preference Ratings for the Depression Scenario Choice and Percentage Condition
Drug†
Rating
No.
%
Mean
SD
Three-Option A Paroxetine HCl *Target superior (T) on efficacy Nefazodone HCl (C) Fluoxetine HCl (D)
12
86
4.57
0.65
2
14
3.71
0.73
2.64
1.08
Two-option
Paroxetine HCl Nefazodone HCl
12 4
75 25
4.31 3.38
0.79 1.02
Three-option B Paroxetine HCl *Target superior (C) on side effects Nefazodone HCl (T) Imipramine HCl (D)
2
20
3.40
0.97
7
70
4.10
1.10
1
10
2.80
1.13
† T = target; C = competitor; D = decoy.
0
The patient is a twenty-five-year-old male complaining of the following symptoms: pain and pressure around and behind the eyes, sensitivity to touch above and below the eyes, severe headache, congestion, postnasal drip, productive cough, and an oral temperature of 100.1 degrees (F). After examining the patient your diagnosis is acute sinusitis. The patient is a recovering substance abuser, [which eliminates a narcotic antitussive as a treatment option]. He is otherwise healthy and active. In addition to prescribing Biaxin 500 mg (Q 12 h) 3 10 days to treat the infection, please read the medication options listed below and assume that they are the only options available to you for treating the secondary symptoms of sinusitis. His insurance will cover the cost of any prescriptions after he satisfies the usual $3.00 co-pay.
The statement about eliminating a narcotic antitussive was not included in the scenario where Nu-
VOL 19 / NO 3, JUL–SEP 1999
cofed was the decoy. Appendix A describes the medications in the sinusitis scenario. Table 3 shows the participants’ responses on the two dependent measures. An analysis of the choice data reveals that the attraction effect was not significant in either of the three-option conditions. In the three-option A condition, choice for the target (Cardec) did not significantly increase relative to choice in the two-option condition (57% vs 30%), x2 (1, n = 24) = 1.73, ns, power = 0.26. In the three-option B condition, choice for the target (Tylenol) did not significantly increase relative to choice in the two-option condition (87% vs 70%), x2 (1, n = 26) = 1.21, ns, power = 0.20. However, in both cases the results are in the predicted direction and the nonsignificant test might reflect the small sample size. For both medications, compared with the two-option condition, the percentage of participants choosing a particular medication increased when it was the target and decreased when it was the competitor. A 2 3 3 factorial (drug by condition) ANOVA similar to the analysis for the depression scenario was performed on the ratings data. There was a significant drug-by-condition interaction, F (2, 36) = 5.92, p # 0.01, power = 0.35, indicating the attraction effect. That is, the mean rating for each drug presented in the two-option choice set increased when that drug was the target and decreased when it was the competitor (see Table 3). VAGINITIS SCENARIO
In the vaginitis scenario the residents read the following: The patient is a twenty-seven-year-old female complaining, for the first time, of the following symptoms: vaginal itching, burning, and discharge. After examining the patient, your diagnosis is vaginal candidiasis. In addition to being a first-time sufferer of a yeast infection, the patient is also nursing a twomonth-old infant [eliminating the oral medication Diflucan (Fluconazole) as a treatment option]. Please read the medication options listed below and assume that they are the only options available to you. Her insurance will cover the cost of any medications after she satisfies the usual $3.00 co-pay.
The statement about fluconazole was removed in the condition where it was the decoy. Appendix B describes the medications used in the vaginitis scenario. Table 4 shows the participants’ responses on the two dependent measures. In the three-option condition A the number of participants choosing the target (butoconazole) did not significantly differ relative to choice for the same medication in the two-option condition (90% vs 57%), x2 (1, n = 24) = 3.05, ns, power = 0.42; however,
●
Decisions about Medications
319
Table 3 ● Choice and Preference Ratings for the Sinusitis Scenario Choice and Percentage
Rating
Condition
Drug†
No.
%
Mean
SD
Three-option A *Target superior on side effects
Cardec (T) Tylenol (C) Rhinosyn (D)
8 6 0
57 43
4.21 3.71 2.79
0.90 1.13 0.80
Two-option
Cardec Tylenol
3 7
30 70
3.40 4.20
0.79 1.17
Three-option B *Target superior on efficacy
Cardec (C) Tylenol (T) Nucofed (D)
2 14 0
13 87
3.37 4.43 2.19
0.72 0.63 1.33
† T = target; C = competitor; D = decoy.
the results were in the predicted direction. In contrast, for the three-option condition B where tioconazole was the target, the results were in the opposite direction to what we predicted (12% vs 43%). That is, choice for the competitor increased and choice for the target decreased relative to the two-option condition, although this was not statistically significant, x2 (1, n = 30) = 3.52, ns, power = 0.47. A 2 3 3 (drug by condition) factorial ANOVA similar to the analyses performed for the other two scenarios was performed on the ratings data. There was no significant interaction in this scenario F (2, 37) = 2.66, ns, power = 0.26. That is, there was no evidence of the attraction effect in the vaginitis scenario. This result most likely reflects the fact that butoconazole was preferred over all of the other medications in all three conditions. A subsequent 2 3 2 factorial ANOVA was performed without the fluconazole condition, the results of which were not significant, but were in the predicted direction.
Table 4 ● Choice and Preference Ratings for the Vaginitis Scenario Choice and Percentage Condition
Drug†
Rating
No.
%
Mean
SD
Three-option A Butoconazole (T) *Target superior Tioconazole (C) on side efClotrimazole (D) fects
9 1 0
90 10
4.60 2.50 3.50
0.70 1.18 1.18
Two-option
8 6
57 43
3.86 3.29
0.95 1.33
14
88
4.81
0.40
2 0
12
3.38 1.06
0.89 0.25
Butoconazole Tioconazole
Three-option B Butoconazole (C) *Target superior Tioconazole (T) on length of Fluconazole (D) treatment † T = target; C = competitor; D = decoy.
320
●
Schwartz, Chapman
REPEATED-MEASURES ANALYSIS
In addition to investigating the attraction effect in each individual scenario, we investigated the attraction effect within subjects, collapsed across scenarios. That is, we assessed, for both dependent measures (choice and appropriateness rating), whether individual participants chose or rated a medication highest when it was the target and lowest when it was the competitor. The proportion of participants choosing a target in both decoy conditions (10/39, or 26%) was, in fact, significantly higher than the proportion of participants choosing the competitor in both decoy conditions (2/39 or 5%), McNemar’s x2 (1, n = 39) = 7.143, p < 0.01, power = 0.76). The remaining 69% chose a target in one case and a competitor in another, with the exception of one participant who chose a decoy and was excluded from this analysis. We also analyzed the ratings to determine whether there was a within-subject effect. In each scenario, for each participant we subtracted the rating from the B medication from that of the A medication, where the A medication was the one serving as the target in the A condition, and the B medication was the one serving as the target in the B condition. That is, in the A condition we subtracted the rating of the competitor from that of the target, and in the B condition we subtracted the rating of the target from that of the competitor. We expected that there would be a high positive difference in the ratings of the medications in the A target condition, a near-zero difference in the two-option condition, and a negative difference in the B target condition. The mean differences were 1.250 in the A target condition, 0.775 in the two-option condition, and 0.225 in the B target condition. A 3 3 3 factorial ANOVA (decoy condition by counterbalance condition) showed that there was a main effect for decoy condition, F (2, 37) = 6.97, p # 0.02, power 0.42). That is, the rating differences were ordered as predicted (although the difference in the B target condition was not negative, indicating an overall preference for the A medication.)† The participants rated target †We suspected that the reason the difference in the B target condition of the within-subjects analysis was positive rather than negative and that the difference in the two-option condition was not zero was that the competitor in the vaginitis B target condition was rated higher than the target. Further analysis of the data show that when the vaginitis scenario was removed, the mean difference in the B target condition was indeed negative (20.583). However, these analyses also revealed that the vaginitis scenario was not responsible for the non-zero difference in the two-option condition. The mean differences in the two-option condition and the three-option A condition, without the vaginitis scenario, were 0.884 and 0.967, respectively, suggesting again that subjects showed a slight overall preference for the A medication.
MEDICAL DECISION MAKING
medications higher and competitor medications lower across the three conditions. Thus, although choice and rating comparisons were not significant for each individual scenario, the within-subject analyses indicated an overall attraction effect.
Discussion Given the prevalence of decision-making biases in consumer-choice studies, particularly violations of the regularity principle, examining the presence of similar biases in consumer-oriented areas of physician decision making seemed warranted. In this study we found that a violation of regularity called the attraction effect7 occurs in physicians’ decisions about medications. That is, adding an inferior alternative (a decoy) to a choice set of existing medication options increased the likelihood that the target medication would be chosen. In both the depression and the sinusitis scenarios the results were rather straightforward: the respondents chose a particular medication more frequently and rated it higher when it was the target than when it was presented in the two-option condition. The results of the vaginitis scenario were not as straightforward. On the one hand, when butoconazole was the target, the number of respondents choosing it was greater than when butoconazole was in the two-option condition. On the other hand, the percentage of respondents choosing butoconazole also increased when it was the competitor. There are a number of possible explanations for this finding. First, the attributes used in this scenario were side effects and length of treatment, whereas the other two scenarios used effectiveness and side effects. It is possible that potential side effects were considered much more important than length of treatment, thus causing participants to respond differently to the vaginitis scenario. If this were the case, however, a disproportionate number of participants should have selected butoconazole in all three versions, including the two-option condition. An alternative explanation for the results in the vaginitis scenario concerns the use of a nursing mother as the patient. The physicians were not only deciding which was the most appropriate medication for the patient, but also for her infant. One reason the three-day option seemed so attractive when it was the competitor might have been that the respondents felt that any medication in a concentrated one-day dose was too risky for an infant. Although none of the intravaginal medications described in this study is contraindicated for use in nursing mothers, fluconazole is. Moreover, the fluconazole decoy condition was the only scenario where the medication chart had an explicit warning against pre-
VOL 19 / NO 3, JUL–SEP 1999
scribing the medication for nursing mothers. That warning might have led the respondents to believe that the other single-dose medication was potentially harmful. This might also explain why the choices of the two intravaginal medications were relatively equal in the two-option condition, because in that condition physicians were not exposed to an explicit warning about the risks of another single-dose treatment. The fact that the competitor was chosen with greater frequency when another alternative was added to the choice set represents a violation of regularity, just not the one we were predicting. Apart from the fluconazole condition in the vaginitis scenario, the present study provided support for an attraction effect in physicians’ decisions about prescribing medications in hypothetical cases. The study had some limitations, however, resulting from the sample used. First, generalizability may be limited by the use of residents as subjects. Future research should examine this bias with more senior physicians. Second, the study had a small sample size, so there was low power to detect differences in choice percentages for the individual scenarios. The within-subject analyses provided more support, however, and clearly indicated the attraction effect. Collapsed over the three scenarios, the physicians were more likely to choose target medications and rate them as more appropriate than competitor medications. The current study used paper-and-pencil cases and thus did not demonstrate that the attraction effect influences physicians’ actual decisions or that it impacts patient outcomes. In addition, the scenarios used in this study did not represent all of the medications available in their respective classes. The results nevertheless suggest that the attraction effect could have an impact on real decisions because the descriptions of medications used in this study are quite similar to the way that medications are listed in reference guides and formularies. The layouts used in such guides make it easy to notice dominated alternatives and thus open the possibility of biasing effects. A savvy pharmaceutical company, for example, could position its medication near a dominated alternative as a means of increasing the market share of its medication. Even without explicit intent, such choice sets could occur. A constant flow of new medications, many of which make older medications obsolete, inundates the market each year. Older medications that remain on the market may act as decoys. Although physicians are unlikely to prescribe an obsolete medication, its (strategically located) presence in the choice set might increase prescriptions for a very similar medication. It is important to note that a dominance relationship depends not only on the characteristics of the medication but also on the characteristics of the pa-
Decisions about Medications
●
321
tient. A certain co-morbidity, medical history, or symptom could make a medication dominated for one patient but not another. Some of the decoy medications used in this study would be fine choices for patients with different symptoms or histories. Nucofed, for example, was dominated because the patient was described as a recovering substance abuser, not because it is, in general, an inferior medication. It is not difficult to imagine another situation where that medication might have even been the preferred choice. Thus, identification of a decoy is not always straightforward. While physicians should be aware of advertising and marketing techniques that might make one medication appear more attractive than another, they should also be aware of individual patient characteristics that make one medication dominate another. One question that might be raised about the present results concerns the influence of the brand names of medications. We presented both brand and generic names to ensure that all the medications were recognizable. That is, we did not want physicians to choose a particular medication because they recognized one generic name but not the others. Brand-name allegiance or medication familiarity should not have influenced the likelihood of detecting the attraction effect. If a particular medication had been preferred because of its brandname reputation, that allegiance should have had the same influence in the two-option control condition and the ‘‘competitor’’ experimental condition as it did in the ‘‘target’’ experimental condition. In other words, if subjects chose Tylenol when it was the target simply because it was the medication with which they were most familiar or experienced, one would expect that Tylenol would receive the majority preference when it was in all the conditions, including when it was the competitor, which was not the case in this study. Thus, including the brand names did not confound the results. Like the results reported by Redelmeier and Shafir,6 the findings of this study highlight the fact that increasing the number of available alternatives in a choice set changes the decision outcome. The attraction effect and maintaining the status quo are only two examples of consumer decision-making biases that might be relevant to physician decision making. Since many aspects of health care, for both the physician and the patient, are becoming increasingly more consumer-oriented, it seems beneficial to identify such biases and develop strategies that will help avoid them. How might the attraction effect be reduced or debiased? A debiasing strategy would need to ensure that dominance relationships were noticed, so that the dominated alternative was not chosen, but also that the dominance relationships did not influence
322
●
Schwartz, Chapman
choice among the remaining options. Thus, dominated alternatives should be identified and then eliminated from the choice set in such a way that they do not influence decisions about the remaining medications. Debiasing is complex because the attraction effect may occur, in part, because decision makers wish to avoid making difficult tradeoffs or are searching for a choice that is justifiable. Past research has shown that decision makers are more likely to show the attraction effect when they are asked to defend their decisions.8 In the current study we asked respondents to defend their choices but did not manipulate accountability. An effect of accountability is relevant to medical decision making because of the current emphasis on quality control. Many health care organizations conduct quality reviews or issue report cards on clinical performances. Although these practices have many benefits for the quality of patient care, asking physicians and other hospital staff to defend their choices explicitly may lead to a search for decisions that are easily justified and result in biases such as the attraction effect. Future research will examine the effects of accountability on biases such as the attraction effect in physician decision making. Knowing that there is a tendency to avoid difficult decisions may help physicians make the most appropriate choices.
MEDICAL DECISION MAKING
Thanks are due to Elaine Leventhal, MD, for assistance with designing the cases, to Jack Cappitelli, MD, and Laura Niedermayer for help with administering the study, and to Alan Schwartz, Patrice Tremoulet, and an anonymous reviewer for helpful comments on an early version of the manuscript.
References 1. Tversky A, Shafir E. Choice under conflict: the dynamics of deferred choice. Psychological Science. 1992;3:358–61. 2. Shafir E, Tversky A. Decision making. In: Smith EE, Osherson DN (eds). Thinking: An Invitation to Cognitive Science. 2nd ed. Cambridge, MA: MIT Press, 1995. 3. Slovic P. Choice between equally valued alternatives. J Exp Psychol Hum Percept Perform. 1975;104:280–7. 4. Shafir E, Simonson I, Tversky A. Reason-based choice. Cognition. 1993;49:11–36. 5. Tversky A, Shafir E. The distinction effect in choice under uncertainty. Psychological Science. 1992;3:305–9. 6. Redelmeier DA, Shafir E. Medical decision making in situations that offer multiple alternatives. JAMA. 1995;273:302–5. 7. Huber J, Payne JW, Puto C. Adding asymmetrically dominated alternatives: violations of regularity and the similarity hypothesis. Journal of Consumer Research. 1982;9:90–8. 8. Simonson I. Choice based on reasons: the case of attraction and compromise effects. Journal of Consumer Research. 1989;16:158–74. 9. Physician’s Desk Reference. Oradell, NJ: Medical Economics, 1997.
VOL 19 / NO 3, JUL–SEP 1999
Decisions about Medications
●
323
Appendix A Descriptions of Medications in the Sinusitis Scenario Medication
Effectiveness of Treatment
Possible Side Effects
Pseudoephedrine HCl, 60 mg, dextromethorphan HBr, 15mg, carbinoxamine maleate, 4 mg (e.g., Cardec)
This medication (5 ml qid) has been shown to be 70% effective in relieving the secondary symptoms of sinusitis
Less than 10% of patients report side effects as a result of pseudoephedrine; no report as a result of DM Hbr
Dextromethorphan HBr, 30 mg, acetaminophen, 650 mg, pseudoephedrine HCl, 60 mg (e.g., Tylenol cough/decongestant)
This medication (5 ml qid) has been shown to be 90% effective in relieving the secondary symptoms of sinusitis
Less than 10% of patients reports side effects as a result of pseudoephedrine; 15% report side effects of DM HBr; 5% report stomach upset from acetaminophen
Pseudoephedrine HCl, 30 mg, chlorpheniramine maleate, 2 mg, dextromethorphan, 15 mg (e.g., Rhinosyn) *Decoy for Cardec
This medication (5 ml qid) has been shown to be 60% effective in relieving the secondary symptoms of sinusitis
Less than 10% of patients reports side effects as a result of pseudoephedrine; 5% of patients report somnolence as a result of chlorpheniramine
Pseudoephedrine HCl, 60 mg, codeine phosphate, 20 mg (e.g., Nucofed) *Decoy for Tylenol
This medication (5 ml qid) has been shown to be 90% effective in relieving the secondary symptoms of sinusitis
Less than 10% of patients reports side effects as a result of pseudoephedrine; 25% of patients report constipation, stomach upset, and somnolence as a result of codeine
Appendix B Description of Medications in the Vaginitis Scenario Medication
Length of Treatment
Possible Side Effects
Tioconazole 6.5% (e.g., Vagistat 1)
Single dose cream inserted intravaginally.
Exacerbation of symptoms: e.g., itching, swelling
Butoconazole nitrate 2% (e.g., Femstat 3)
Cream medication inserted intravaginally for three consecutive days
No known side effect
Clotrimazole 1% (e.g., Mycelex 7) *Target for butoconazole
Cream medication inserted intravaginally for seven consecutive days
No known side effect
Fluconazole (e.g., Diflucan) *Target for Tioconazole
150-mg single dose oral medication
Fluconazole is secreted at high concentrations in human breast milk, thus, not recommended for nursing mothers
