* Candidate for the Lee B. Lusted Student Prize Competition
Purpose: We aimed to explore the effect of the way numerical information is framed on participants’ judgments. Specifically, based on a real story about a man who died after hernia surgery, we investigated the extent to which participants revised their liability judgment about the medical staff for not having used heparin, based first on ambiguous numerical information (relative risk reduction: 50%) and then on unambiguous numerical information (absolute risk reduction), that was framed either in terms of survival (e.g., 999 vs. 998 out of 1,000 patients survived) or mortality (1 vs. 2 out of 1,000 patients died), and presented as either resulting from small (1,000) or large (10,000) sample size groups.
Method: Participants (N =160) were students (aged 19 to 45, M = 22.07, SD = 3.60) who volunteered to take part in the experiment. Perceived liability and perceived risk reduction were measured on 7-point scales.
Result: When presented with relative risk information, participants judged the medical staff as more liable (M = 4.84, SE = .10) than when presented with absolute risk information (M = 4.22, SE = .11, F(1, 156) = 48.47, p = .0001, hp2= .24). The change in judged liability depended both on frame and on sample size, which did not interact: Participants’ rated the liability higher in the mortality condition (M = 4.76, SE = .13) than in the survival frame condition (M = 4.31, SE = .13) F(1, 156) = 5.61, p = .02, hp2 = .04 (and post-hoc analyses showed that the judgment was revised only in the survival condition), and participants’ ratings were lower in the small-1,000 size condition (M = 4.34, SE = .13) than in the large-10,000 size condition (M = 4.72, SE = .13) F(1, 156) = 3.89, p = .05, hp2 = .02. Participants’ estimates of perceived risk reduction were also affected both by the frame and the sample size as for liability. The results of mediational analyses were consistent with the hypothesis that perceived risk reduction plays a mediational role between the frame of information and the judgment of liability.
Conclusion: Even when provided with unambiguous numerical information, people seem to be affected by the way in which it is framed, and the frame affects liability judgment through the perceived risk reduction.
Purpose: Paper-based patient decision aids generally present risk information using numbers and/or static images. However, limited psychological research has suggested that when people use computers to interactively graph risk information, they process the statistics more actively, making the information more available for decision making. Such interactive tools could potentially be incorporated in a new generation of web-based decision aids. But, little is known about whether interactivity improves communication of health risk information.
Method: 3,377 members of a demographically-diverse Internet panel viewed a hypothetical scenario about two hypothetical treatments for thyroid cancer. Each treatment had a chance of causing one of two side effects, but we randomly varied whether one treatment was better on both dimensions (strong dominance condition), slightly better on only one dimension (mild dominance condition), or better on one dimension but worse on the other (tradeoff condition) than the other treatment. We also varied whether respondents passively viewed the risk information in static pictograph (icon array) images or actively manipulated the information by using interactive Flash-based animations of “fill-in-the-blank” pictographs. Our primary hypothesis was that active manipulation would increase respondents’ ability to recognize dominance (when available) and choose the better treatment.
Result: The interactive risk graphic conditions had significantly worse survey completion rates (65.7% vs. 79.5%, p<0.001) than the static image conditions. In addition,respondents using interactive graphs were less likely to recognize and select the dominant treatment option (e.g.: 61.1% vs. 73.1%, p<0.001, in the strong dominance condition).
Conclusion: Interactivity, however visually appealing, can both add to respondent burden and distract people from understanding relevant statistical information. Decision aid developers need to be aware that interactive risk presentations may create worse outcomes than presentations using static risk graphic formats such as pictographs.
Purpose: There is limited understanding of how physicians assess the effectiveness of prescription drug options and use that knowledge in their treatment decisions.
Method: We used a qualitative mental models research method to explore physicians’ perceptions of drug efficacy and effectiveness; how they use and interpret clinical efficacy information when making decisions; and how they communicate to their patients about effectiveness. We conducted 44 in-depth telephone interviews, averaging 55 minutes in length, with office-based physicians (23 rheumatologists and 21 general practitioners) across the United States. Thirty-two percent of respondents were female and 20% identified their race or ethnicity as being other than non-Hispanic white. The interviews focused respondents’ thinking on assessing and communicating about drugs to treat rheumatoid arthritis or psoriasis. Interviews were coded and analyzed to develop a graphical model of the key influences on respondents’ judgments and decisions regarding treatment effectiveness.
Result: The interviews revealed respondents' complex thinking on drug effectiveness. (a) Respondents believe that finding an effective treatment for a chronic condition is a process of “trial and error” and rely heavily on follow-up “global assessments” to evaluate the effectiveness of a prescribed treatment. (b) Respondents said that they balance the drug’s expected benefits, safety profile and the likelihood of patient compliance when assessing drug effectiveness before prescribing. (c) Respondents generally find expert assessments and treatment guidelines more relevant to their decisions than clinical efficacy and other statistical data. (d) When communicating to patients about effectiveness, a primary goal is to manage patients' expectations and comfort with the treatment decision, and thus respondents said they tailor communications according to their assessments of their patients’ prior beliefs about the condition or treatment, desire for information, ability to understand and comfort with the information.
Conclusion: Physicians need and desire clinically-relevant (“real-world”) and accessible information on drug effectiveness that enables tradeoffs among benefits, risks and costs. The research findings and resulting models can inform the development of guidance on how to provide physicians with the balanced and accessible information they need to make judgments and communicate with patients about drug effectiveness.
Purpose: To develop a test of a measure of health numeracy that is based on an empirically derived framework, cross-culturally equivalent for Hispanic and Non-Hispanic populations, and uses Item Response Theory scaling methods.
Method: A cross-cultural approach was used in the development of the Numeracy Understanding in Medicine Instrument (NUMi). Qualitative methods used to generate the item bank (n=110) included focus groups, convening of an expert panel, and cognitive interviews. Participants (n=1000) were recruited from community and clinical populations to test the items. A 2-Parameter IRT model was used for analysis of the first 500 respondents. Participants also responded to the Test of Functional Health Literacy in Adults-Short Form (S-TOFHLA) and a cognitive reasoning and aptitude test (Wonderlic). A sample (n=200) of participants were also administered the Lipkus numeracy scale and a mathematic achievement test (WRAT-M).
Result: Of the first 500 participants 50% were White and 40% were Black. Over 30% were Hispanic and 40% had a high school level education. Based upon the IRT parameters of difficulty (range of -3.0 t0 3.0) and discrimination (range of 0 to 3.0), 20 items were chosen to form the NUMi. The most difficult items in each domain assessed the following skills: 1) Determining how many 500 mg pills add up to 3 grams (number sense), 2) interpreting a risk of 2/1000 in a pictograph (tables and graphs), 3) interpreting a relative risk reduction (probability), and 4) interpreting the meaning of a p-value that is < 0.05 (statistics). The Test Information Function peaked at a difficulty level of -1.0 indicted that the test is most discriminating for people with lower than average health numeracy. Test performance was positively associated with cognitive reasoning (0.79) and print health literacy (0.54).
Conclusion: The NUMi is a 20 item paper and pencil test that measures an empirically based construct of health numeracy. In future work, a computerized adaptive test will be developed in which items are selected based upon the ability of the respondent leading to a shorter test that will reduce respondent burden and allow for oral or print administration. The NUMi will provide a feasible and valid assessment of health numeracy skills for use in both research and clinical settings.
Purpose: Research in behavioral economics suggests that individuals facing complex decisions benefit from being given a “nudge” towards one option, especially in situations where making any choice, as opposed to none, is preferred. Decisions about colorectal cancer (CRC) screening are of this type, since several tests are recommended by guidelines, including colonoscopy, sigmoidoscopy, and stool testing. No studies have examined the use of a nudge in the context of CRC screening. In this study, we compared the effects of two different approaches to providing quantitative information about CRC risk and benefits of screening, one with and one without a nudge towards fecal immunochemical testing (FIT) (a stool test).
Method: 186 adults aged 40-70 years visiting a state fair viewed a general video about CRC screening and then were randomized to view a computer-based presentation of natural frequency data regarding the risk of CRC and the risk reduction provided by the three most common tests. Half (n=93) were randomized to view a version that included a nudge towards FIT, while the other half viewed an identical version without the nudge. Subjects completed a survey before and after the presentation that assessed interest in CRC screening (10-point scale, with 1 = not interested at all to 10 = very interested) and preferred test (stool testing, sigmoidoscopy, or colonoscopy). Paired t-tests were used to assess differences in interest before and after viewing the presentation (differences by group compared with two-sample t-tests). Test preferences were compared using chi-square tests.
Result: Subjects reported significantly greater interest in CRC screening after viewing the natural frequency data, compared to baseline (8.9 vs. 8.3, p < .001), whether they received the nudge (8.8 vs. 8.2) or not (9.0 vs. 8.3) (p=.59). A greater proportion of subjects who received the nudge chose FIT as their preferred test compared to those who viewed the presentation without the nudge (50% vs. 30%, p=.007), and were more likely to change their preference from colonoscopy at baseline to FIT after the presentation (27% vs. 7%, p = .005).
Conclusion: Providing quantitative information increases interest in CRC screening and providing a nudge towards one test can have a significant impact on individuals’ test preferences. Further research in a clinical setting is needed to confirm and more precisely characterize these effects.
Purpose: Public ambivalence towards H1N1vaccination provides an opportunity to explore decision making facing incomplete information and beliefs regarding risks and vaccine effectiveness. We hypothesized that individual behaviour might be influenced by mimicry of others.
Method: 95 medical students were asked about perceived risks and utilities regarding H1N1 influenza and whether they would take the vaccine, under varying putative rates of immunization among their classmates. Expected utilities of vaccination and non-vaccination were generated from individualized decision trees and Reed-Frost models.
Result: Expected utilities of vaccination and non-vaccination were positively correlated with rate of class immunization (r=0.95, p<0.001 for both). At low class immunization rates, vaccination was found to dominate. The herd immunity threshold was 78%, beyond which the dominant strategy would be not to take the vaccine. However, among 59% of the students, decision was not affected by rate of class immunization: of these, nearly half would take the vaccine. In 41% of students, decision was correlated with rate of immunization (r=0.27, p=0.001).
Conclusion: Some individuals gain utility by mimicking others, forfeiting seemingly rational strategies. Others opt for rational strategies, while distancing themselves from the majority they sometimes depend on. Maximising populations’ utilities might require seemingly paradoxical global strategies. It would be in the public’s interest to allocate some members to lesser utilities than their comrades, as occurs with vaccination and herd immunity. This balance would be difficult to maintain without external intervention. These findings can be generalized to other scenarios where personal utility might not correspond to the public’s preferences.