Risk communication formats for low probability events: an exploratory study of patient preferences

Clear communication about the possible risks and benefits of proposed medical interventions has been an integral part of medical care since the doctrine of informed consent was adopted. In recent years, it has become increasingly more important due to changes in the accepted model of the doctor-patient relationship that promote more active involvement of patients in decisions about their care, the rise of evidence-based medical practice, and the increasing emphasis placed on preventive measures that are intended to reduce future health risks [1, 2]. It is likely to become more important in the future as new information about personal health risks becomes available through work in the "new sciences" such as medical genomics, metabolic profiling, and proteomics [3]. Richard Smith, former editor of the British Medical Journal, recently called risk communication the "main work of doctors" [4].

Current recommendations for communicating information about uncertain future events emphasize the importance of presenting data in a balanced manner that avoids framing effects, provides baseline risk information, and uses graphic risk displays whenever possible. Recommended risk displays include bar charts, grouped icon displays (also known as dot or face diagrams), and flow diagrams [5–7]. The use of multiple risk presentation formats has also been suggested [7, 8].

Despite the importance of risk communication, there have been few empirical studies comparing different graphic formats for presenting probabilistic information to patients for use in shared decision-making [8–12]. In comparative studies, both bar charts and icon displays have been more effective than simple numeric statements and several other graphic formats for comparing the outcomes of alternative decision options [13, 14]. Ordered icon displays have also been shown to minimize decision biases due to vivid anecdotal information and to be well understood by patients over 75 years old [16, 17]. Flow diagrams, developed by Gigerenzer and colleagues, have been shown to help people better understand risk forecasts as well as more complex probabilistic data, such as the change in risk of disease after a positive or negative test result [18, 19].

A limitation of the currently available data is that most studies to date have compared risk displays using relatively common events with likelihoods of 20% or more. Outcomes with expected likelihoods of less than 5% occur frequently in medical decision making situations and in many cases have an important impact on medical decisions. Probably the most common of these low likelihood outcomes are risks of serious adverse events. For example, in 2004 the drug rofecoxib, a COX-2 inhibitor, was removed from the market in the United States due to concerns about an increase in the annual rate of myocardial infarctions from one per thousand to four per thousand, an absolute risk increase of three per thousand [15].

Small risks are difficult to depict in bar charts and icon displays because it is hard to include all of the necessary information in a single display. Studies with both bar charts and icon displays have shown that graphs that do not include information about the chances of both experiencing and not experiencing the outcome can lead to misinterpretation of the data presented [12, 16]. The need to properly illustrate this part-to-whole relationship is what makes it difficult to use these formats to convey information about low likelihood events

A possible solution to this problem is to use graphs that include a special section designed to highlight information about low likelihood events. This approach has been used successfully in two studies that elicited probability estimates from patients using graphs that included a "magnified" zone for probabilities less than 1% [22, 23]. Whether this approach can also be used to create effective graphic displays for conveying low likelihood information to clinical decision makers is currently unknown.

This study was designed as the first step in a series of investigations designed to address the question of how to most effectively present information about low likelihood events to patients and other clinical decision makers. The purpose of this study was to explore the potential usefulness of several new small-risk graphic communication formats.

The mean age of the study population was 56.4 years. There were 22 men and 7 women. The preference scores for the six presentation formats are summarized in the box plot shown in Figure 6. The most preferred format was the combined augmented bar chart + flow diagram with a mean preference score 0.43. This format is shown in Figure 4. The combined augmented icon display + bar chart was second most preferred with a mean preference score of 0.22. A small group of patients strongly preferred this format as indicated by the outlier circles on the box plot. The combined augmented icon display + flow chart had the third highest mean preference score, 0.15. The three single presentation formats were the least preferred with mean preference scores of 0.10 for the augmented bar chart, 0.07 for the flow diagram, and 0.04 for the augmented icon display.

The overall differences in preferences among the six formats were statistically significant: Kruskal-Wallis Chi Square = 141.4, p < 0.0001. The Bonferroni multiple comparison test indicated that the mean preference scores for all three combined formats were significantly different from all other options and the mean preference score for the augmented bar chart was significantly different from the mean score of the augmented icon display, (p < 0.05).

The consistency indices for the comparisons ranged from 0.08 to 0.20 with a mean of 0.135. All subjects met the predefined standard for acceptable consistency of 0.20 or less.

A recently issued set of guidelines for creating patient decision aids recommends the use of multiple risk presentation formats [7]. These results support this recommendation. The most preferred format was a combined format and all three combined formats were more preferred than the three single format options included in the study.

We were unable to find previous studies that have assessed patient preferences for single versus combined graphic risk presentation formats. In a study that compared text formats for presenting treatment effectiveness information, Sheridan and colleagues found that patient understanding was worse for a combined format than several single formats [18]. The discrepancy between this result and ours may reflect differences in the way people process textual versus graphical information [19, 20]. Alternatively, it could be due to the differences in the outcomes studied: preferences versus understanding.

Given the limited scope of this study, we do not know why the patients preferred the combined formats so strongly. The cost-benefit theory of decision strategy choice posits that the way information is displayed affects a person's choice of how they will process the information based on anticipated effort and anticipated accuracy [19]. It is possible that the increased variety of information presentation provided by the combined formats led the patients to perceive that they could be used to make more accurate decisions with less effort than the single format displays. Additional research is warranted to confirm our results and to address additional outcomes including understanding of the information presented, cognitive effort, the rationale underlying risk presentation preferences, and the effect of presentation preferences on the decision making process.

This study also demonstrates the usefulness of using the Analytic Hierarchy Process as a technique for creating measurement scales in situations where no pre-existing scale exists. Although the AHP is most widely known as a multi-criteria decision making method, it is fundamentally a system of measurement that derives a ratio-level scale from a series of pairwise judgments [21–28]. In addition to producing a robust measurement scale, the AHP also yields an assessment of the consistency of the component judgments that can be used as an indicator of the reliability of the measurement process. The AHP therefore is a valuable assessment tool that can enhance our ability to investigate preferences and other important, but difficult to measure, concepts.

This study has several limitations. First, the only outcome studied was patient preference. At least two studies have shown that the most preferred risk formats are not necessarily the same as the most effective formats for risk communication in medical decision making [13, 14]. Assessing preferences as the initial step in evaluating new risk communication formats, however, has the advantage of providing easy to obtain information about an important dimension of a risk display that can be used to help plan additional investigations. Second, only two likelihoods for a single outcome were studied. We were, therefore, not able to determine if there is an interaction between the magnitude of the risk being presented and preferred display format. Third, the study population was a small volunteer convenience sample of people who were not facing a real decision. Fourth, the reproducibility of preferences over time was not tested. Finally, only one version of the newly created graphs was studied. Whether patient preferences are affected by different color schemes, axis formatting, the size of the display, and other design characteristics is unknown.

Despite these limitations, the results of this study suggest that patients may prefer combined, rather than single, graphic risk presentation formats and that augmented bar charts and icon displays may be useful for conveying comparative information about small risks to clinical decision makers. Further research to confirm and extend these findings is warranted.