Testing a Best Practices Risk Result Format to Communicate Genetic Risks

Abstract

Objective

To investigate the effect of a genetic report format using risk communication “best practices” on risk perceptions, in part to reduce risk overestimates.

Methods

Adults (N=470) from the Coriell Personalized Medicine Collaborative (CPMC) were randomized to a 2×2 experimental design to receive a hypothetical “personalized” genetic risk result for leukemia (relative risk=1.5 or 2.5) through either the standard CPMC report (N=232) or an enriched report informed by best practices (N=238). A one-time, online survey assessed numeracy and risk perceptions including “feelings of risk” and a numerical estimate.

Results

Regardless of numeracy, participants who received the enriched report had fewer overestimates of their lifetime risk estimate (LRE; odds ratio=0.19, p<.001) and lower feelings of risk on two of three measures (p<.001). Participants with higher numeracy scores had fewer overestimates of LRE (OR=0.66, p<.001) and lower feelings of risk on two out of three measures (p≤.01); the interaction between numeracy and report format was non-significant.

Conclusion

The enriched report produced more accurate LRE and lower risk perceptions regardless of numeracy level, suggesting the enriched format was helpful to individuals irrespective of numeracy ability.

Practice Implications

Best practice elements in risk reports may help individuals form more accurate risk perceptions.

1. INTRODUCTION

The increasing availability of genetic risk information has the potential to allow people to tailor health behaviors and healthcare to reduce personal risk. However, behavioral changes (or intentions) resulting from genetic risk information have been reported in only a small number of studies and one meta-analysis [1–3], though a different meta-analysis reported no association between receiving genetic risk information and risk reducing behaviors [4]. One potential reason for this lack of effects may be that the risk communications employed in the studies are themselves ambiguous and not systematically informed by best practices in risk communication. We sought to create an optimal personalized risk communication format informed by best practices that appear applicable irrespective of user level of literacy and numeracy.

Genetic risk information can be complex and can be communicated both in prose and numerical formats, often using probabilities, such as relative risk, absolute or lifetime risk, or absolute risks presented within a timeframe (e.g., 5-year cancer recurrence risk) [5,6]. This complexity is likely one barrier to comprehension; only 13% of American adults are considered “proficient” at reading, interpreting, and performing calculations [7], whereas approximately half of Americans have literacy skills at basic or below basic levels, generally considered as below 8^th grade equivalence [8–10]. Unfortunately, clinical genetic reports are often written at or above a college reading level [11]. Illustrating this difficulty interpreting risk information, one qualitative study found laypeople to interpret a 9% lifetime cancer risk in multiple ways: 9-out-of-100 people, 9-out-of-10 people, in the 9^th percentile of risk, and as 9 out of a 10-point scale [12].

Accurately interpreting risk information relies in part on an individual’s numeracy skills. A study of genetic testing comprehension in a highly-educated sample found numeracy to be the strongest independently associated variable [13]. Previous work also found that lower numeracy is associated with overestimating risk, which may influence health decisions, though people have been found to overestimate risks despite numeracy ability [8,14]. Overestimates also occur more often for smaller risks, such as contracting HIV from a single event of a risky behavior [15]. Further, overestimating cancer risk has been found to promote overuse of screening while also being associated with excessive cancer worry [8,16]. Thus, numeracy is an important skill for comprehending test results and simply presenting numerical risk information is not adequate to ensure accurate interpretation and behavior modification.

Many attempts have been made to improve communication of personal risk reports, with previous work testing formats that vary text, numerical, and visual information. Numerical formats can include ratios, relative risk, absolute risk, or other formats, with absolute risk as one of the best options to communicate risk [5,6]. Visual numerical information formats can include pie charts, bar charts, number lines, survival curves, and icon arrays; when risk reports lack a visual representation they may be more confusing or of limited benefit [17–21]. Generally, previous work has found that icon arrays (preferably matrices of people) may be the most ideal way to present numerical information, as they are preferred by laypeople and are usable and trusted by individuals regardless of numeracy ability [6,17,22–25].

Despite the body of risk communication literature, it appears little of this work has been incorporated into current reporting formats in genetic testing contexts. Clinical genetic reports use minimal (if any) visual formatting and are written beyond the college level [11] while direct-to-consumer reports (e.g. 23andMe) have employed only some optimal strategies [26].

These risk communication methods likely influence how people interpret risk from a genetic risk report and how they construct their risk perception. Recent work identified three ways in which people interpret personal risk: deliberative risk (a rule-based judgement about perceived likelihood), affective risk (emotional reactions to risk such as worry), and experiential risk (gut-level beliefs such as vulnerability). Each construct is empirically independent and also uniquely predictive of intention and behavior [27]. However, much of the previous work evaluating risk communication and perception has evaluated only one or two of these constructs. Further, previous research on risk report formats has almost exclusively focused on individual “best practices,” but no work appears to have studied the combined effect(s) of these strategies—particularly in a genetics context—meaning there is no consensus “gold standard” risk report nor an understanding of how such a report might affect risk perceptions.

The objective of this study is to improve the accuracy of risk perceptions, focusing on individuals with lower numeracy. To this end, we developed an enriched risk report format that expands upon a “standard” communication of genetic risk to participants in the Coriell Personalized Medicine Collaborative (CPMC). In this trial, we explored the extent to which the enriched format, informed by eight best practices design elements, effectively calibrated or corrected biased risk perceptions and determined the extent to which numeracy might moderate this effect. We hypothesized that participants with lower numeracy would overestimate their numeric risk more often than those with higher numeracy and that participants who received the enriched rather than the standard risk report would have more accurate risk estimates (e.g., lower risk perceptions and fewer numerical overestimates).

2. METHODS:

2.1. Participants and Recruitment

2.1.1. CPMC Study Description

Study participants were recruited online from the CPMC, a prospective, multi-cohort study evaluating the utility of personalized genetic risk assessments in health outcomes. CPMC participants receive risk reports for complex diseases with information regarding their genetic contribution to disease as well as other factors (e.g., family history). Relative risk values span 0.08 to >6.0. Participants had access to educational tools and free genetic counseling [28].

2.1.2. Study Recruitment

This study received ethics approval from the Coriell Institute for Medical Research IRB (study # R158) and was performed in accordance with the Declaration of Helsinki. We sent email invitations in 2015 to CPMC members who had 1) active email accounts and reviewed at least one CPMC personalized risk report and 2) who had previously elected to receive notices about optional studies. Participants were also recruited via notice of the study on the CPMC web portal’s “optional studies” tab. All consented participants completed the survey at a location other than the CPMC (e.g. home or office) and received a $10 Amazon e-gift card for participating.

2.2. Study Design

2.2.1. Study Administration

We administered a one-time, online experiment to CPMC participants. Recruitment, randomization, surveys, and data collection were performed online. Participants were randomized at enrollment to one of four groups in a 2×2 (relative risk level by risk report type) between-subjects factorial design (Figure 1). Randomization was not blocked or stratified, and both participants and researchers were blinded to participant allotment for each of the four groups. The study was not pre-registered as it was conducted prior to the current NIH guidelines. Prior to receiving a hypothetical result (i.e. the vignette), we collected: numeracy assessments, personal and family history of leukemia, risk perceptions for leukemia, and genetic self-efficacy ratings. After participants received the vignette, risk perceptions were obtained again.

Figure 1:

CONSORT Study Flow Diagram

Our target enrollment was 500 participants and we enrolled 491 individuals from the larger study population of N=2,815 who had access to this study. Recruitment ran from August 2015 – January 2016 and ended after a lack of uptake from successive reminder emails. We excluded 21 participants: 16 due to incomplete surveys, four due to an enrollment error, and one due to incomplete demographic data. The final study was N=470.

2.2.2. Risk Vignette

Leukemia was chosen as the target of risk results because 1) the CPMC did not have a current or future risk report for leukemia; 2), leukemia was thought to be uniquely salient as a cancer that was not obscure but for which people would have few (if any) preconceived notions of risk; 3), preventive measures could theoretically be taken, such as learning to recognize symptoms, smoking cessation, reducing chemical exposure, and discussing the risk report results with a doctor.

After answering pre-vignette survey questions, participants were instructed to imagine that they have received a new personalized risk result from the CPMC. Due to concerns that this study’s risk report could be confused with a real CPMC result, the word “fictional” was used whenever describing the increased leukemia risk. Participants then reviewed their risk report (RR=1.5 or 2.5). We used two relative risk levels to determine if participants were sensitive to different low-risk figures. Half received this information in only the standard CPMC format (Figure 2). The other half received the standard format with the accompanying enriched risk report (Figures 2 and 3). The enriched format included all elements of the standard CPMC format so that any differences in risk perceptions could be attributed to the enriched format.

Figure 2:

The Hypothetical “Standard” Genomic Risk Report

Figure 3:

The Enriched Report for the 2.5 Relative Risk Level

For participants to easily calculate their new “personal” risk of leukemia, we chose an “average” leukemia population risk (i.e. absolute risk) of 1% in all vignettes. The 1% number was chosen to reduce the possibility of miscalculation as a potential explanation for inaccurate risk estimates.

As one limitation of vignette methodology is that participants may have difficulty immersing themselves in the imagined scenario [29], we designed the standard risk report to mimic the typical CPMC report in style, formatting, language, and informational content with the number of risk alleles identified (one or two), a table showing the relative risk result, a graph depicting the elevated risk, and a statement explaining their absolute baseline risk. Though the CPMC report format during the time of this study did not necessarily apply all of the “best practice” elements found to promote interpretation and comprehension, CPMC risk reports did use simplified language, graphs to visualize risks, and applied more best practices than the typical clinical genetics result report. The CPMC report design was thus considered an appropriate comparator.

2.2.3. Design of the Enriched Risk Result Format

We developed the enriched report format after review of the relevant literature prior to 2015, which yielded eight “best practices”:

1. perform necessary calculations to reduce miscalculation errors [10,14];

2. present risks in the same numerical format to eliminate the need to convert between risks (e.g. comparing 1-in-4 to 20%) [30];

3. write text at an ~8^th grade level [6,31];

4. omit qualitative risk descriptors (e.g. “high”), which are not interpreted uniformly [5,15];

5. uniformly frame risk information both in terms of gains versus losses (i.e., risk of getting cancer versus a risk of not getting cancer) as well as the type of risk information presented (e.g., only absolute risks vs. absolute and relative risk formats) [15];

6. include a pictograph, which is easily interpreted and among the most preferred formats for non-scientific risk communication [5,17,22–25].

7. use absolute risks, as a relative risk can appear as a larger risk (e.g. a 100% increase from a 1% lifetime risk is a 2% risk, but the “100%” relative risk can be misinterpreted) [5,6];

8. omit other types of numerical information, such as disease incidence [6,25].

All authors (except LS), as well as multiple individuals not enrolled in the study with varying levels of experience in math and science, reviewed the enriched and standard risk reports for clarity and errors.

2.2.4. Survey Instruments

Participants completed the Lipkus Expanded Numeracy Scale [10] to measure objective numeracy and the Subjective Numeracy Scale [32]. Despite the high average scores of the Lipkus scale in educated populations [10], the scale was chosen due to its wide use [10,14,20,33]. Scores on the subjective scale were used as a control in regression models.

Risk perceptions were measured with four questions to measure the three independent constructs of risk perception [27]. Deliberative risk perception (i.e., cancer likelihood) was measured in two ways, with both a numerical estimate of lifetime leukemia risk and as a Likert-like question adapted from previous work [34]. The numerical estimate was measured before and after the vignette with the same single-item, fill-in-the-blank question eliciting a numerical estimate: “If my current behaviors don’t change, I think I have a _____% chance of getting blood cancer in my life.” Because generating a specific, numerical risk estimate can be a difficult exercise, a risk magnifier scale (i.e. number line) was provided with this question. This scale was adapted from previous work and showed a continuum from 0–100%, with low percentages (0–1%) provided in detail [35]. The second question addressing deliberative risk was measured using a 7-point scale with the question: “If my current behaviors don’t change, I think my likelihood of getting blood cancer in my life is____.” (1 = not at all likely, 7 = almost certain).

Three questions were used to capture the emotional response to the risk result. Affective risk (i.e., worry about developing cancer) was assessed with two items, each on a 7-point scale: “If my current behaviors don’t change, I am ____ worried about getting blood cancer.” (1 = not at all, 7 = extremely) and “If my current behaviors don’t change, I am ____ concerned about getting blood cancer.” (1 = not at all, 7 = extremely) [34]. Experiential risk (i.e., perceived vulnerability or “gut feelings” of being at risk) was measured with one question using a 4-point scale: “I feel vulnerable to blood cancer.” (1 = agree strongly, 4 = disagree strongly) [36].

2.2.5. Statistical Analyses

Our analytical strategy consisted of statistical tests with the key independent variables being receipt of the standard vs. enriched report, numeracy, and risk result level (1.5 or 2.5 relative risk). We used logistic regression to explore whether the enriched result format (received vs. not received), the relative risk level, or a participant’s numeracy (the independent variables) were associated with numerical overestimates on the percentage risk perception scale (lifetime risk estimate or LRE; the dependent variable). To do this, we created two dichotomous variables: 1) categorizing LRE as either an overestimate or not (>1.5 or 2.5 or not) and 2) participants’ numeracy scores were classified as a higher level (numeracy score ≥9) or lower level (numeracy score <9). We dichotomized LRE into over- or underestimates because previous work has found that asking participants to generate a specific number may lead to a “guess” with the common choice of 50% used to indicate a numerical “shrug” and therefore may not be a precise measure [37]. We dichotomized numeracy to provide easier interpretation for logistic regression results as well as to account for the highly skewed results that are typical of the Lipkus numeracy instrument. Our cutoff for high versus low numeracy was used in a previous study of genetic risk and numeracy [38].

We used linear regression to evaluate the effect of the enriched result format (received vs. not received), numeracy (as a continuous variable), and risk level (the independent variables) on risk perceptions (continuous variables for deliberative, affective, and experiential risks; dependent variables). For both models, we assessed for interactions between risk report format and numeracy.

In both regression models, demographic variables, pre-vignette risk perceptions and subjective numeracy were included as covariates. Each demographic variable (age, sex, ethnicity, education level) was tested separately for each outcome. Variables with significant associations (p<0.05) were retained as covariates in the regression model for that outcome. Statistical modeling was performed using R.

3. RESULTS

3.1. Cohort Characteristics

Most of the population was female (67.4%) and Caucasian (93%), with incomes over $50,000 (82.8%), high educational achievement (78.7% with a bachelor’s degree or higher), and no family or personal history of leukemia (92.1% and 98.5%, respectively; see Table 1 for group details). Average age was 51.5 years and about half (49.6%) worked in healthcare, education, or office administration. The median numeracy score was 10 (of 11 total); the mean score was 9.5 (standard deviation = 1.5; range = 3–11). The majority (77.7%) of participants had “high” numeracy scores (≥ 9). Lastly, no specific harms were noted or brought to the researchers’ attention during or after conducting this study, including confusing the hypothetical result with a true risk report.

Table 1:

Group Demographic Details

Report Type; Risk Level		Standard; RR=1.5	Enriched; RR=1.5	Standard; RR=2.5	Enriched; RR=2.5
N		116	117	116	121
Age (Years)		51.6	52.3	50.1	52.2
Female N (%)		74 (63.8)	77 (65.8)	82 (70.7)	84 (69.4)
Race N (%)	Caucasian	106 (91.4)	110 (94.0)	108 (93.1)	113 (93.4)
	Af. American	2 (1.7)	2 (1.7)	4 (3.4)	3 (2.5)
	Asian	4 (3.5)	1 (0.8)	1 (0.9)	2 (1.7)
	Multiple	2 (1.7)	4 (3.4)	3 (2.6)	2 (1.7)
	No Answer	2 (1.7)	N/A	N/A	1 (0.8)
Personal History of Leukemia N (%)	Yes	3 (2.6)	0 (0)	1 (0.9)	3 (2.5)
	No	113 (97.4)	117 (100)	115 (99.1)	118 (97.5)
Family History of Leukemia N (%)	Yes	7 (6.0)	3 (2.6)	8 (6.9)	8 (6.6)
	No	106 (91.4)	110 (94.0)	107 (92.2)	110 (90.9)
	Don’t Know	3 (2.6)	4 (3.4)	1 (0.9)	3 (2.5)
Income N (%)	<$25,000	5 (4.3)	10 (8.5)	8 (6.9)	7 (5.8)
	$25,000 – $99,9999	58 (50.0)	44 (37.6)	60 (51.7)	48 (39.7)
	>$100,000	53 (46.6)	62 (53.0)	48 (41.4)	66 (54.5)
	No Answer	0 (0)	1 (0.8)	0 (0)	0 (0)
Educational Exposure N (%)	High School	4 (3.5)	4 (3.4)	1 (0.9)	4 (3.3)
	College	68 (58.6)	66 (56.4)	64 (55.1)	55 (45.5)
	Graduate School	44 (37.9)	47 (40.2)	51 (44.0)	62 (51.2)

Lifetime Risk Estimates and the Enhanced Risk Report, Risk Level, and Numeracy

We first explored the relationships of risk report format (standard vs. enriched), risk level (1.5 vs. 2.5), and numeracy level dichotomized as “high” (score ≥9) vs. “low” (score <9) with lifetime risk estimate (LRE) of leukemia. We used logistic regression to determine the odds ratio (OR) for participants’ likelihood to overestimate LRE of leukemia (N=470) after receiving genetic risk information (Table 2). Receipt of the enriched format (N=238) produced fewer overestimates of risk (OR=0.19, p<.001). Additionally, participants with higher numeracy were less likely to overestimate their LRE (OR=0.66, p<.001). There was no significant interaction between report format and numeracy level, nor any effect of risk level (p>.05).

Table 2:

Logistic Regression of LRE for Correct Estimates Compared to Overestimates

	Eta (SE)	Z	OR	2.5 – 97.5% CI	p
Intercept	3.07 (0.95)	3.25	21.63	3.38 – 140.36	.001
Sex	−0.93 (0.32)	−2.89	0.40	0.21 – 0.73	.004
Pre-vignette LRE	0.05 (0.01)	3.90	1.05	1.03 – 1.08	<.001
SNS	0.01 (0.02)	−5.70	1.01	0.97 – 1.06	.59
RR level	−0.81 (0.27)	−2.99	0.45	0.26 – 0.75	.003
Enriched report	−1.66 (0.29)	0.55	0.19	0.10 – 0.33	<.001
Numeracy level	−0.42 (0.09)	−4.42	0.66	0.54 – 0.79	<.001

OR: odds ratio; CI: confidence interval; RR: relative risk; SNS: subjective numeracy scale; LRE: lifetime risk estimates

Risk Perceptions and the Enhanced Risk Report, Risk Level, and Numeracy

We then explored the impact of numeracy score (as a continuous variable), risk level, and risk report format on the three Likert risk perception measures. We observed significant main effects of format type on affective and experiential risk perceptions (Table 3), such that receiving the enriched report (N=238) lowered both affective and experiential risk perceptions. We also found significant main effects of relative risk level (1.5 or 2.5) on affective risk (worry about developing cancer) and experiential risk (perceived vulnerability), but not deliberative risk (Table 3), such that both measures of affective and experiential risk were higher for participants in the 2.5 relative risk arm (N=237). Lastly, we found significant associations between numeracy score and affective and deliberative risk perceptions (Table 2), such that higher numeracy was associated with lower risk perceptions on both measures. There were no significant associations between numeracy and experiential risk (perceived vulnerability). The interaction between numeracy and the enriched report was not significant for any outcome (p>.05).

Table 3:

Linear Regression Results for Leukemia Risk Perceptions

	Affective			Deliberative			Experiential
	β (SE)	t	p	β (SE)	t	p	β (SE)	t	p
Intercept	5.85 (0.97)	6.03	<.001	3.27 (0.48)	6.74	<.001	0.50 (0.37)	1.37	.17
Sex	−0.55 (0.26)	−2.15	.03	−0.21 (0.13)	−1.62	.11	0.28 (0.09)	3.20	<.001
Pre-vignette risk perception	0.45 (0.08)	5.78	<.001	0.24 (0.06)	4.27	<.001	0.44 (0.06)	7.45	<.001
SNS	0.01 (0.02)	0.37	.71	0.01 (0.01)	1.26	.21	0.00 (0.01)	−0.03	.98
Enriched report	−1.04 (0.24)	−4.35	<.001	−0.22 (0.12)	−1.84	.07	0.35 (0.08)	4.24	<.001
RR level	1.32 (0.24)	5.49	<.001	0.11 (0.12)	0.92	.36	−0.36 (0.08)	−4.45	<.001
Numeracy	−0.23 (0.09)	−2.58	.01	−0.16 (0.04)	−3.64	<.001	0.06 (0.03)	1.84	.07

RR: relative risk; SNS: subjective numeracy scale

4. DISCUSSION

4.1. Discussion

We assessed the impact of an enriched risk report as well as numeracy on risk perceptions after receiving a hypothetical genetic risk result; we experimentally manipulated the format of the risk information as well as risk magnitude. Our first hypothesis was supported, as higher numeracy was associated with fewer overestimates of LRE after receipt of the genetic risk report with corresponding lower levels of affective and deliberative risk perceptions. This result was obtained despite the relatively high level of numeracy in the sample. Our second hypothesis was also supported, as participants who received the enriched report were less likely to overestimate numerical risk and perceived lower affective and experiential risk, reflecting a more accurate (lower) perception of their leukemia risk. Interestingly, there was no interaction between risk report format and numeracy in either analysis, suggesting that the enriched risk format was effective regardless of participant numeracy. These results demonstrate the benefit of receiving the enriched risk report in reducing overestimation of leukemia lifetime risk.

It is important to note that although the enriched format reduced the proportion of participants who overestimated LRE, some participants still overestimated their risk. These participants may have been affected by factors beyond the vignette such as health habits (e.g., smoking) or extended family history. Some overestimates may also be attributed to the challenge in creating a numerical estimate, which may promote guessing [37]. Also, despite small difference in relative risk levels, main effects were observed in risk perceptions between the 1.5 vs 2.5 risk levels for both affective and experiential risk perception measures. The lack of main effect on the Likert-like deliberative risk measure may be due to the two risk levels being too similar or the risk measure being less sensitive than the other two measures. Lastly, a personal or family history of leukemia may have increased risk estimates and ratings, but we did not perform sub-analyses on these groups due to small sample sizes (N=7 and N=26, respectively).

Fuzzy Trace Theory (FTT), a form of the dual-processing model, may best explain our results for the enriched risk report format performance. FTT posits that people use one of two mental systems to interpret new risk information: deliberating and encoding a verbatim (numerical) risk or a second system that derives the gist (“bottom-line”) and emotional meaning of the message [40,41]. While neither interpretive system is inherently better than the other, people who have difficulty thinking about numerical risk in a deliberative way are likely at a disadvantage when presented with that type of information. Modern healthcare relies heavily on numerical risk information, which likely limits individuals with lower numeracy from adequately using this type of health information. A review of multiple studies of FTT suggests that altering the gist risk message may be more important for affecting downstream behaviors than improving understanding of a verbatim risk [40]. Consistent with these results, our enriched report appears to have helped participants interpret both their verbatim risks (fewer LRE overestimates) and adjust their gist-based and emotional risks (lower worry and vulnerability ratings) by reducing the numeracy burden and providing a complementary icon array.

In the current study, the goal was to reduce risk perceptions due to overestimation, but both lowering and increasing risk perceptions (achieving risk accuracy) is desirable. Lowering risk perceptions may be helpful to reduce overtreatment from overdiagnosis. This may be most promising in conditions like prostate cancer, with screening identifying high rates of indolent tumors, leading patients and providers to decide on potential invasive treatment options that have many complications. Making this decision is likely difficult for many men. Illustrating the difficultly of balancing risk information, a group of researchers published a model estimating the chance the cancer diagnosis was of a result of overdiagnosis (would not have been caught without screening); the authors state the information should inform cancer risk perceptions and intervention choices [39]. On the other hand, clearly communicating risk to increase risk perceptions may be critical to motivate preventive health behaviors that reduce risk, particularly among individuals with lower numeracy. Those who incorrectly use or are unable to use numerical risk information may react differently to risk information [8]. Studies have reported that experiential, deliberative, and affective risk perceptions are associated with intentions for taking protective actions (getting a vaccine) or protective health decisions (discussing genetic risk results with a healthcare provider) [27,42]. Altering these risk perceptions, rather than a numerical estimation, may be more effective in motivating downstream changes in behavior.

As numeracy is difficult to improve, simplifying complex and abstract risk information is an important strategy to improve the utility of risk results. This study provides evidence that an enriched risk report format can influence risk perceptions in desired ways in an educated and highly numerate population. It is possible that these effects might be more substantial in a population with lower literacy and numeracy abilities. If further studies support the generalizability of these findings, this type of risk report format could be adapted and used to present personalized genetic risk information at scale.

4.2. Limitations

Key limitations to this study include that there was a small proportion of participants with “low” numeracy in our sample (scores<9: N=104, 22%). Next our fictional population risk of 1% may have been too close to the natural limit of 0%, and thus more obviously “low” relative to other lower risks. The CPMC population is not representative of the US population, as it is older, mostly female, with higher earnings and education level. Our study does not address how risk perceptions might change in the context of high-risk results (e.g. a pathogenic BRCA1 variant). And lastly, we used a hypothetical vignette, which may have influenced risk perceptions as participants knew that they were not truly at an increased risk of leukemia.

4.3. Conclusions

Our enriched risk format can increase accuracy by reducing numerical overestimates while also calibrating affective and experiential risk perceptions. Gaining perspective on how people interpret genetic risk information, form risk perceptions, and ultimately act on them can help practitioners deliver risk information using effective means so people can take informed actions to manage their risks.

4.4. Practice Implications

Our enriched risk format appears to be effective across numeracy abilities, requires few resources for implementation, and can be adapted to different genetic risk results, leading to scaling for a large volume.

Highlights.

• We developed a novel risk report addressing multiple, common numeracy issues

• Risk accuracy improved with the new risk format regardless of numeracy ability

• “Feelings of risk” were lower with the new risk format regardless of numeracy

• Using numeracy best practices seems to improve numerical genetic risk communication