1.1 DTC genetic testing

DTC genetic testing provides opportunities for consumers to learn about their health risks related to a variety of diseases and conditions. Customers may seek this information because of negative experiences in traditional healthcare settings (Lowes et al., 2023) and a desire to be more informed about their health and make relevant lifestyle changes (Roberts et al., 2017). Over half of DTC genetic test customers say the results will impact how they manage their health (Roberts et al., 2017). However, the test insights are only useful if they are both accurate and effectively communicated. As noted by Leighton et al. (2012) “misunderstanding of the results could lead to negative consequences including unnecessary concern, false reassurance or unwarranted changes in screening behaviors” (p. 11). Likewise, in a letter to the editor of Genetics in Medicine, Schleit et al. (2019) stated that harm in the field of genetics overall could include “(1) failure to diagnose a genetic condition; (2) inappropriate diagnosis of a genetic condition, resulting in unnecessary treatment, or failure to find the true etiology; and (3) emotional distress from a misdiagnosis or misunderstanding of risk” (p. 510). Indeed, a survey of customers who uploaded their genetic data to a third-party website to generate polygenic risk scores found that nearly two-thirds of participants experienced negative feelings such as being upset or anxious (Peck et al., 2022). Another study found that 38% of participants had not considered they might receive unwanted information (Roberts et al., 2017). As personalized medicine and DTC genetic tests have become more available, professionals and others have called for greater regulation (Cernat et al., 2022).

1.2 Communicating absolute and relative risk

DTC or any genetic testing companies that conduct predictive testing must communicate probabilistic information about the risk or likelihood of developing certain diseases or other conditions. More specifically, companies may communicate health risk in terms of relative risk or absolute risk. Absolute risk indicates the level of one's risk susceptibility compared to absolute zero (Covey, 2011). For example, stating there is a 3% chance an individual will have a heart attack in their lifetime. On the other hand, relative risk indicates one's level of risk susceptibility compared to other risks or compared to the risk to other individuals (Covey, 2011). For example, stating someone is at 200% higher risk of having a heart attack than the general population. In both cases, the risk can be determined by genetic, lifestyle, or environmental factors. Although absolute risk and relative risk metrics are provided in efforts to mitigate negative health risks and both are mathematically correct representations of risk information, each type of risk information seems to differentially affect individuals' decision-making (Reimer et al., 2015).

Health communication research suggests that communicating relative risk can be misleading and often exaggerated, but communicating only absolute risk can downplay the importance of risk factors (Citrome, 2010). This is particularly true in low-probability scenarios that are “essentially nil” when presented in absolute terms but may be meaningful when presented in relative terms (Stone et al., 1994). Research has demonstrated that when communicating relative risk, it is important to include the baseline risk to avoid alarm and improve understanding of risk estimates (Berry et al., 2006). Even clinicians have been found to confuse absolute and relative risks. In one study, those who mistook relative risk for absolute risk on a scenario-based question reported higher levels of enthusiasm for medical screenings (Caverly et al., 2014). Still, some companies might opt to provide more information or multiple assessments of the same risk for the sake of transparency and to help consumers make the most informed health-related decisions possible. However, scholars caution that comparing multiple estimates of the same risk reinforces and highlights the uncertainties involved in risk assessments, and perhaps could further confuse or alienate receivers of such information—particularly if the reported results seem contradictory (Lundgren & McMakin, 2018).

In a review of the comparative effectiveness of ways to communicate probabilistic information, data synthesis showed that the use of visual aids (icon arrays and bar charts) improved both understanding and satisfaction, and that the presentation of numbers reduced understanding (Zipkin et al., 2014). Other research found that lay publics believe risk communication is improved with “more explanatory detail, reporting probabilities for individual health impairments, and specifying risks for subgroups” (Ellermann et al., 2022, p. 1).

In addition to visual aids designed to improve understanding, other visual aspects of risk communication have been found to impact risk perceptions. For example, the use of color can draw attention to certain design aspects and signal safety or danger (Sutton & Fischer, 2021). Visuals in the form of charts (Garcia-Retamero & Cokely, 2011) or photos (Reifegerste & Rossmann, 2017) have also been shown to increase the effectiveness of health-related risk messages.

Different DTC genetic tests vary in their prediction of disease occurrence, with some genetic variants conferring relatively low risk and others indicating near certainty (Cameron et al., 2009). DTC genetic testing companies may choose to communicate these risks in different ways, which existing research indicates can have an impact on consumers' risk perceptions (Cameron et al., 2009). A survey conducted in 2010 indicated that many DTC genetic testing participants altered their health-related behaviors, including information-seeking and lifestyle adjustments, based on how they perceived their results of colon cancer risk (Kaufman et al., 2012). Furthermore, their changes in behavior were different depending on whether the consumer received a “lifetime” absolute risk or relative risk compared with the general population. Other research has found that consumers who perceived a greater risk were more likely to seek alternative testing and participate in lifestyle changes (Sherman et al., 2015) and that perceived levels of risk influenced feelings of worry and distress (Cameron et al., 2009). The research on relative and absolute risks leads to our first research question:

RQ1: (a) How do DTC genetic testing companies communicate relative and absolute health risks? (b) Are these risks defined or explained by the companies?

1.3 Communicating limitations

Closely related to the communication of risk, this study also examined the communication of limitations. Limitations, and uncertainty more broadly, are often discussed as a necessary part of science and scientific findings; all research contains limitations, which must be considered for the appropriate interpretation of research findings. Previous research outside the realm of genetic testing has examined portrayals of scientific uncertainty (Ratcliff et al., 2021), and some research suggests the effects of uncertainty portrayals in messaging often appear contingent on individual differences such as one's preferences for uncertainty information and uncertainty tolerance (Ratcliff, 2021; Ratcliff & Wicke, 2022).

Scientific uncertainty, including limitations, is often communicated through hedging language. Hedging is a linguistic concept used to express doubt or uncertainty, often through words such as might or probably (Caffi, 1999). The inclusion of hedging in scientific discourse acknowledges caveats, limitations, or uncertainties related to research findings. This is important in the context of genetic testing because although certain genetic variants may be detected, other genetic and lifestyle factors also determine disease risk. Previous research has examined the consequences of including hedging language in public communication about science (Jensen, 2008), and VanDyke et al. (2023) found that such hedging language can increase trust in DTC genetic testing companies among those who have already taken a test. In that study, hedging was specifically related to tests not being used to make medical decisions and that healthcare professionals may suggest further testing. Taken together, prior research suggests that scientific uncertainty portrayals may influence individuals' reactions to scientific information; accordingly, consideration of the prevalence and nature of such information was warranted. Based on this review, we posed our second research question:

RQ2: How do DTC genetic testing companies communicate product limitations in the context of risk assessment?

2.1 Sample

Top companies were determined by Google searches and consulting online popular press lists and reviews1 because usership data is not available for most companies. No tests that required a physician's referral or recommendation were included. Ten companies that offer both ancestry and health or wellness information consistently showed up in these lists and the same researcher took each of the 10 tests as part of a broader research project. Of the 10, only three of them offered specific health risk information related to monogenic and complex diseases. Thus, those three companies were chosen for this analysis. This differs from many popular genetic testing companies that only offer tests related to nutrition and exercise. Two of the companies provided single nucleotide polymorphism genotyping and one offered whole-genome sequencing.

The design of each portal varied, but each one included summary pages or dashboards along with detailed reports for each disease or trait that was tested for. Results from both the dashboards and individual reports were included in the analysis. It is important to note that the companies also communicate through email and one-time messages that pop up during the first time a page or report is accessed. The current study only examined how results were conveyed on the portals that customers can continually access.

A purposive, maximum variation approach (Creswell & Poth, 2018) to sampling was employed to select reports (i.e., results for individual diseases) for analysis that (a) at least two companies tested for, (b) varied in terms of the seriousness and types of disease, and (c) varied in terms of whether variants detected were associated with increased or decreased risk. For example, of all the monogenic reports, only one (hereditary hemochromatosis) indicated a present variant so it was important to include this report so the sample represented how carrier status was communicated. Thirty-nine reports were included in the analysis. These included 15 different diseases and 13 reports from each company. When converted to PDF, there was a total of 246 pages of reports. They ranged in length from 2 to 17 pages (M = 6.34, SD = 4.03). All three companies had reports for nine of the diseases chosen and at least two companies included reports for the remaining six diseases. For the full list of reports and top-level results, see Table 1.

2.2 Analysis

The analysis of dashboards and individual reports aligned with Roller's (2019) eight-step process for qualitative content analysis and used a hybrid approach of inductive and deductive coding. Specific topics such as absolute and relative risks were derived from literature, but the qualitative content analysis process also allowed coders to identify codes in the text that were not initially planned. First, all web pages that communicated results were read in their entirety to better understand what content was relevant to the research questions. Based on this, individual web pages were chosen as the unit of analysis. Full pages provided enough content to provide important contextual information but were narrow and defined enough to identify differences (Roller, 2019).

The sample consisted of the main results pages and individual reports. Because reports minimally varied within each company in terms of the type of information provided and formatting, and because of the specificity of the research, we determined the initial sample had adequate information power (see Malterud et al., 2016) and further sampling was not needed. After the initial reading and absorption of content, coders developed unique codes related to the research questions. These were often guided by literature but the operationalizations were specific to the content. On a second read, information related to the research questions was recorded and coded. Based on the noted portions, topics were first listed and then organized into categories. Each coder (the first two authors of the study) then identified common patterns across codes and categories. To ensure the confirmability of the findings, the coders met to debrief and arrive at an agreement regarding the final themes. Finally, exemplar quotes were pulled to illustrate the findings. Because visual elements such as position, size, and color were also relevant to the research questions, these were noted along with the themes so findings could be conveyed more holistically.

3.1 Absolute and relative risks

Results reported by all three companies included both relative and absolute risks in some way, but relative risk was consistently most emphasized. For example, for two of the companies (A and B), diseases were labeled with terms including “increased likelihood,” “slightly increased risk,” and “high risk” when detected mutations suggested a higher risk compared to the average in the referent group, their user database. These results were highlighted by using a different color text. For Company A, this text was red for “high risk” vs. green for “reduced risk.” This company organized reports on a table based on relative risk under the headings “High Risk,” “Reduced Risk,” and “Typical Risk.” The columns included “Your Risk” (i.e., absolute risk), “Average Risk” (i.e., population incidence), and “Compared to Average” (i.e., relative risk).

All tests included the “Compared to Average” multiplier but not all included the absolute percentage. To demonstrate how relative risk was used as the organizing variable, the following are the examples of reports included in the table. Coronary heart disease was categorized as “high risk” with 1.0% absolute risk, which was 2.51× the average risk. In contrast, intracranial aneurysm was categorized as normal risk. It had a higher absolute risk at 2.08%, but this was only 1.04× the average. For Company B, it was blue for “increased likelihood” vs. black text for all other tests. For both companies, test results indicating any increased risk were visually located at the top of the list.

Company C presented results in a report library that could be sorted by polygenic score percentile (relative risk), by the date the test was added to the user's portal, or when the research was published that the test was based on. For each report, the most prominent result was the polygenic score percentile, which was displayed in larger text with an accompanying graphic. If users clicked “view full report,” then the polygenic score was explained. This was consistent with how each test was explained by Company C. In short, effect sizes for all variants tested were summed, and the total effect size was compared to a sample of 5000 other customers to calculate the percentile. The most prominent metric, which was the percentile, could be the same or similar for different tests, but the amount of variance explained varied greatly. For instance, the report for schizophrenia showed results in the 93rd percentile, but all the variants tested explained only 3% of schizophrenia risk.

Overall, estimates of absolute risk were less prominently displayed. For example, Company B showed a “slightly increased risk” of celiac disease and positioned it visually toward the top of the reports, but later in the report stated, “However, studies estimate that only about 3% of people with one or more copies of the HLA-DQ2.5 or HLA-DQ8 haplotypes develop celiac disease.” Similarly, for restless leg syndrome, the report stated “An estimated 25% of people with genetics and other factors like yours develop restless legs syndrome by their 70s.” In this case, it was unclear what “other factors” lead to the absolute risk estimate, but because limited information is provided along with the genetic sample, these could refer to age and sex. As previously indicated, Company A included estimates of both absolute and relative risk on its main reports page, but reports were still organized based on relative risk.

3.2 Disclaimers

Disclaimers were consistently included with test results to communicate how test results should be used. These disclaimers included warnings about the tests not being diagnostic, the need to consult a healthcare provider, nongenetic factors that can also play a role in risk susceptibility, unknown or nontested genetic factors that may also play a role in risk susceptibility, and in some cases, the fact that some racial groups have been studied more than others. For example, a typical disclaimer by Company B was, “This report does not diagnose any health conditions or provide medical advice and should not be used to make medical decisions.” For certain diseases, Company B also linked to a tutorial on genetic health risk, which included more about limitations, but after reports were initially accessed, this tutorial was an external link customers would need to click on for the information. Company C used the same language in all reports. The end of each report stated, “However, please note that genetic predispositions do not account for important nongenetic factors like lifestyle. Furthermore, the genetics of most traits has not been fully understood yet, and many associations between traits and genetic variants remain unknown.”

Other examples included “Keep in mind that these reports do not include all possible genetic variants that could affect these conditions. Other factors can also affect your chances of developing these conditions, including lifestyle, environment, and family history” (Company A) and “The variants included in this test are common in many ethnicities, but are best studied in people of European descent” (Company B).

Hedging language was commonly used in reference to health risks because probabilities are inherent to that information. Hedging language was less common in direct relation to test results or their accuracy. One exception was a statement Company B included in several reports: “This estimate is based on currently available data and may be updated over time.”

3.3 Lifestyle changes

Lifestyle was often cited as a contributing risk factor in disclaimers. However, its prevalence in reported results warrants discussion. In addition to being mentioned as a contributing factor in the development of certain diseases, lifestyle was also discussed in terms of how to prevent those diseases. Lifestyle recommendations were often made by Company A and Company B. Company C less often and less specifically made lifestyle recommendations, but at times noted that lifestyle factors matter generally.

Lifestyle recommendations were not included with every test result and appeared to be excluded specifically in cases where it is well established that there are no known lifestyle factors that contribute to risk susceptibility. For example, Company A included with its Parkinson's disease report, “the majority of the times the disease is sporadic (not hereditary) and there is no effective way of preventing it.” For celiac disease, all three companies mentioned avoidance of gluten as a treatment for those diagnosed with celiac. Only Company B mentioned diet as a potential preventative measure by saying, “Gluten (found in wheat, barley, and rye) is the main nongenetic factor that triggers the development of celiac disease in people with increased genetic risk.”

Lifestyle recommendations also ranged in terms of how actionable or impactful the advice was. For instance, a report from Company A indicated the customer has a “high risk” for schizophrenia. Under the section on prevention, the report stated, “There are triggering factors that can be easily avoided, but it seems that hereditary factors play a very important role in developing schizophrenia, which makes it difficult to prevent.” One of the “easily avoided” triggers listed by Company A was stress.

General health advice, such as maintaining healthy sleep, exercise, and nutrition habits, was often included. Although overall health may be a factor for many of the diseases tested, the advice applied was displayed regardless of the genetic test outcome.

3.4 Additional screening

Some reports also indicated when additional screening was available through a physician for particular diseases or when there were certain indicators that customers should look out for. For melanoma, all three companies included advice related to monitoring moles. Company C worded the advice: “The easiest check is called the “Ugly Duckling” test—if a mole looks different from the other moles on your body like the ugly duckling looked different from its siblings, it's worth getting checked out! Melanomas are often asymmetric, have unusual borders, have different colors, are large in diameter, or change over time.”

Only Company A and Company B included monogenic disease tests, which included carrier status reports. Both companies included a warning that carriers can transfer a genetic variant to their children, but only Company B mentioned that the other prospective parent might also want to be tested. They stated, “If you're starting a family, a genetic counselor can help you and your partner understand if additional testing might be appropriate.”

4.1 Limitations and future research

The limitations of this study suggest several avenues for future research. The current study assesses what companies are currently doing and not the effects of those practices beyond what associations we can make to prior studies. Future survey and experimental research will provide additional insight into how customers interpret genetic test findings and potential best or worst practices. Eye-tracking data would be particularly useful in assessing what messages customers are attending to, and how information processing may affect health-related decision-making and behavior. However, the current results cannot make inferences regarding company intentions behind the information design choices observed. Future research should strive to connect the links between production, content, and the consequent effects of the information provided to consumers.

Although the results provide valuable insights, the nature of qualitative research means we cannot generalize findings to all companies, tests, or consumers' results. Companies beyond these three may communicate results differently than these findings suggest. We also focused specifically on complex and monogenic diseases. Companies may communicate differently about general wellness information or the limitations of tests related to ancestral information.