A Pilot Randomized Controlled Trial of the Feasibility, Acceptability, and Impact of Giving Information on Personalized Genomic Risk of Melanoma to the Public

Primary and secondary prevention strategies are essential for reducing melanoma incidence and mortality. It is estimated that more than 80% of melanomas in high-incidence countries could be prevented through reduced sun exposure (1), and regular sunscreen use can halve melanoma incidence, regardless of skin color or age (2). There is also strong observational evidence that skin self-examination and clinical whole-body skin examination are associated with lower risk of thick melanoma and reduced melanoma mortality (3). Reducing melanoma incidence and mortality is also important from an economic perspective, as together with the keratinocytic carcinomas, they place substantial resource burdens on healthcare systems (4, 5).

Communication of personalized genomic risk information has the potential to improve primary and secondary melanoma prevention behaviors. Common genomic variants for melanoma (6) have been demonstrated to have a strong contribution to melanoma risk prediction (7, 8), including for those with a low-risk phenotype (e.g., darker skin; refs. 9–11). Some people who perceive themselves to be at low risk of melanoma on the basis of their phenotype and adapt their sun protection behaviors accordingly could actually have higher-than-average genetic susceptibility to melanoma. Social and behavioral theory suggests that the highly personalized nature of providing disease risk information based on numerous common gene variants may be a more powerful motivator of behavior change than standard approaches (12).

To date, few published randomized controlled trials have examined the population health impact of genomic risk information based on common variation in many genes on health-related behaviors (13), and none have examined skin cancer prevention behaviors. Although some studies have examined the impact of single-gene variants or “genetic risk” (referring to single, usually high-penetrance genetic mutations) on health behaviors (12–14), any one single-gene common variant captures only a fraction of the genomic contribution to risk of a common disease, usually has a small individual effect on personal risk, and its measurement is therefore likely to be translated as a risk message with low motivational potency (12, 15). However, motivation to change behavior is influenced by a range of factors such as personalization, health literacy, personal skills, self-efficacy, social support, and risk perception (12). Risk precision may be a less important element in motivation than these other influences.

Some studies have shown that providing genetic risk information motivates preventive behaviors among people with a strong family history of melanoma (16, 17), but there has been limited research on communicating disease risk based on common genomic variants to the general population. Current evidence suggests minimal impact on behavior, emotions, and knowledge; however, these studies have been impeded by poor methodological quality (12, 13, 18). Social and psychologic outcomes following an offer of genomic risk information to the public also remain underexplored, despite their importance in contextualizing behavioral effects and informing future implementation (19, 20).

We conducted a pilot randomized controlled trial to evaluate the feasibility and acceptability of giving information on personalized genomic risk of melanoma (based on variants in 21 genes) to the public, and its preliminary impact on sun exposure, sun protection, and skin examination behaviors and broader social, psychologic and economic outcomes. We hypothesized that the impact of personalized genomic risk information may differ according to genomic risk category. We chose a pilot design due to the novel nature of the proposed intervention and as a way to assess the study design, acceptability, feasibility, and operational aspects of a protocol being considered for implementation in a larger study (21). Although not sufficiently statistically powered, a pilot study provides preliminary data on efficacy of an intervention (21).

This pilot study used a randomized controlled trial design (see Fig. 1) and is reported according to CONSORT guidelines (22). We identified potentially eligible participants from the Cancer Council NSW “Join a Research Study” database, comprising people with cancer, relatives, friends, and the wider public, who have agreed to be contacted by researchers conducting ethically approved, cancer-related research studies. People aged 18 to 69 years, residing in a range of geographical areas in the state of New South Wales (NSW) Australia, with sufficient English language capability to complete questionnaires, and no personal history of melanoma were eligible to participate. In April 2015, potentially eligible participants were mailed a study invitation pack containing an invitation letter, information sheet, consent form, and participation card. Written, informed consent was obtained from all participants. Telephone genetic counseling was available at the time of consent. Ethics approval was obtained from The University of Sydney (2014/868). The study was registered at the Australian New Zealand Clinical Trials Registry (ACTRN12615000356561).

At baseline, participants provided a saliva sample using a mailed Oragene kit (DNA Genotek). The saliva samples were returned to the research team by mail, then sent in bulk to the NATA-accredited Australian Genome Research Facility, who extracted DNA, and performed genotyping. The Supplementary Online Materials describe genotyping and calculation of genomic risk estimates for melanoma. Supplementary Table S1 shows a list of the 42 successfully genotyped variants (SNP) from 21 genes/regions.

Randomization to the intervention or waitlist control arm (allocation ratio 1:1) was performed by a statistician not involved in recruitment, based at the NHMRC Clinical Trials Centre, The University of Sydney, thus ensuring allocation concealment. Minimization was used to ensure the groups were balanced by risk category (high, average, low), age (18–44, 45–69 years), and sex. It was not possible to blind participants to study arm allocation.

During August to September 2015 (on average, 3.5 months after providing a saliva sample), all participants in the intervention arm received a telephone call from the study's genetic counselor, guided by a detailed telephone-based communication manual for the delivery of genomic risk information. Telephone-based genetic counseling has been found acceptable for the communication of genetic test results, and patients have reported similar satisfaction between in-person and telephone genetic counseling (23, 24). Participants could elect to receive their melanoma genomic risk information either via telephone from the genetic counselor, followed by a mailed personalized booklet or via mailed booklet only. All participants were given a telephone number to contact the genetic counselor if desired. Those who had a high-risk estimate and chose not to receive their risk information via telephone (i.e., mailed booklet only) received a follow-up call from the genetic counselor. Participants were also asked if they wanted a copy of their risk information sent to their primary care physician.

The personalized risk information booklet described their melanoma genomic risk estimates (absolute risk, relative risk, risk category) in simple language, using pictographs and words (see Fig. 2). It also contained brief information on other risk factors, reducing risk, relevance for relatives, and genetic counseling (see Supplementary Online Materials for more details including a copy of the booklet).

Alongside their mailed personalized risk booklet, intervention arm participants received a nonpersonalized educational booklet “Melanoma information: prevention and early detection” developed for the study. This booklet described melanoma and its causes, preventive behaviors to reduce melanoma risk including sun protection and self- and doctor-conducted skin checks, and information on Vitamin D.

Participants in the waitlist control arm (also known as a delayed intervention control arm) received a mailed copy of the non personalized “Melanoma information: prevention and early detection” booklet at the same time as the intervention participants. To avoid confounding by weather, the timing of the mail-out of booklets was matched between intervention and control groups, by age group, sex, and day. Waitlist control participants were offered their personalized genomic risk of melanoma booklet with genetic counseling after they had completed the 3-month follow-up measures.

Participants completed the baseline questionnaire (by mail or online, according to personal preference) at the same time as their saliva sample, and the follow-up questionnaire 3 months after receiving their booklet/s. Behavioral and psychosocial measures were collected at both time-points, but UV dosimeters were worn at follow-up only.

We measured participation rates, completeness of data collection, and follow-up. Satisfaction with different aspects of the study was evaluated using scales on a rating of 0 to 10. Space for qualitative feedback was also provided at follow-up. A priori feasibility objectives were based on other studies (25–27) and our collective experience: >20% consent, >90% with successful genotyping, >90% of intervention participants receive risk feedback, <20% lost to follow-up. Acceptability objectives were: average satisfaction scores ≥7/10, <30% negative qualitative comments from questionnaire.

Self-reported time spent outdoors, sun protection, sunburn, and skin examination behaviors were collected using validated questionnaire items (28, 29). Further details are provided in the Supplementary Online Materials.

An objective measure of sun exposure was collected at 3-month follow-up using polysulfone film ultraviolet (UV) dosimeters. This is a gold standard method for assessing total daily dose of UV radiation exposure (30, 31). The data were analyzed as standard erythemal doses (SED), which is a standard measure of UV dose; a person's daily SED exposure is influenced by time of day (i.e., more SEDs per hour in the middle of the day), time in the sun, use of sun protection, and season (32). Participants were asked to wear a custom-made wristband with one dosimeter badge inserted each day, over 2 weekdays and 2 weekend days. The wrist has been shown to be a practical and reliable location for personal UV dosimetry (33).

Skin cancer–related worry (34) comprised a mean score from 3 questionnaire items: “The possibility of one day developing melanoma worries me,” “Whenever I hear of a friend or relative (or public figure) who has melanoma, it makes me realize that I could get it too,” and “It would be terrible to get a melanoma,” with a 5-point Likert score response. These items have been shown to be associated with the frequency of skin self-examination in people without melanoma (34). Psychologic distress and well-being were measured using the 5-item version of the Mental Health Inventory (MHI-5) designed for primary care settings (35). Participants' views on genetic determinism were measured using items on the degree to which genetic factors and sun habits (e.g., use of sun protection) cause melanoma (36), perceived personal control over developing melanoma (37), and beliefs such as “There's not much you can do to lower the chance of getting melanoma,” and “For me, using sunscreen can reduce my risk of developing melanoma” using 5-point Likert scales.

Hypothesized mediators of skin cancer prevention behavior change were measured at baseline and 3-month follow-up using questions previously developed for skin cancer research (34, 36, 38) and on the basis of established health behavior theories (ref. 39; see Supplementary Online Materials for details of measures).

On the follow-up questionnaire, participants provided self-reported information about health system resource use including visits to primary care and specialist doctors and individual out-of-pocket expenses including purchase of sun protection items.

Intention-to-treat analyses compared outcomes for intervention and control arms. Our a priori hypothesis was that outcomes may differ by genomic risk category; thus, in addition to presenting overall results, we stratified analyses by high-, average-, or low-risk groups. For continuous outcome measures, we used ANCOVA adjusted for baseline values to estimate the mean difference between intervention and control groups. UV dosimeter values were log-transformed because of their right-skew distribution and analyzed using ANCOVA adjusted for baseline self-reported total sun h/wk because UV dosimetry was not collected at baseline. ANCOVA results based on the log-transformed values were interpreted as a percentage change in the geometric mean of SEDs/day (40). Log-transformed UV dosimeter values summarized as geometric means and 95% confidence intervals (CI). For binomial outcome variables, we used log-binomial models to estimate relative risks and 95% CI, adjusted for baseline values. Two-sided tests were used for all analyses. Statistical significance was inferred at P < 0.05, although the pilot study was not powered to measure the effectiveness of the intervention.

The consent rate was 41% overall (Fig. 1) but differed by age and sex; for those aged 18–44 years, it was 21% for men and 32% for women, and for those aged 45–69 years, it was 53% for men and 46% for women. The average age of people who gave consent was 49 years for women and 59 years for men, compared with 43 years (women) and 51 years (men) for those who declined or did not respond to the invitation. People living outside metropolitan areas made up 22% of those who gave consent and 20% of those who did not. Reasons for declining to participate included overseas travel, medical issues, and concerns about the potential impact of their genetic results on their future life insurance. Study enrolment was capped at 120 participants. Of these, 119 completed baseline measures including a questionnaire and DNA sample. Three of the saliva samples were found not to contain measurable DNA for genotyping (consistent with expected failure rate); of these participants, 2 successfully repeated their saliva sample. Thus, 118 participants had complete baseline questionnaire and genotyping data and were randomized to either the intervention (n = 60) or the waitlist control group (n = 58). No potential participants contacted the telephone genetic counselor at the time of consent when considering study participation.

Participant characteristics at baseline are summarized in Table 1. The ages of participants ranged from 22–69 years with an equal proportion of men and women. The mean age was 51 years (SD, 14 years) for the intervention arm and 55 years (SD, 13 years) for the control arm. There were similar numbers of control and intervention participants in each genomic risk category (Table 1). Compared with Australian population data, the sample had a higher proportion with a family history of melanoma (25) but a similar proportion with personal and family history of other types of skin cancer (41).

Most (78%) intervention participants elected to receive their genomic risk information via telephone from the genetic counselor, followed by mailed personalized booklet; 22% chose to receive the mailed booklet only, but of these, 5 were in the high-risk category and so received a follow-up call from the genetic counselor. Most participants (87%) elected to have a copy of their genomic risk information posted to their primary care physician.

At 3-month follow-up, questionnaires were completed by 108 (92%) participants and 102 (86%) wore the UV dosimeters and completed the self-report sun habits diary (100 completed 4 days, 1 completed 3 days, and 1 completed 2 days). One waitlist control participant requested not to receive their personalized genomic risk information after the 3-month follow-up due to concerns about life insurance implications.

Intervention participants reported high satisfaction with the personalized genomic risk booklet (mean = 8.6, SD = 1.6) and the genetic counselor telephone call (mean = 8.1, SD = 2.2) on a 0–10 scale. Satisfaction was similar for the 3 genomic risk categories. In response to the question, “Would you have rather received your risk information differently to the way you received it?” 93% selected ‘No’. More than half (57%) reported reading the personalized genomic risk booklet from “cover to cover,” 15% reported reading “most of it,” 9% reported reading “only the parts I felt were relevant to me,” 13% read the booklet “briefly,” 2% did not read it, and 2% were unrecorded. Some intervention participants (n = 16, 30%) provided qualitative feedback at follow-up: 6 comments were categorized as positive, for example: “I appreciated that you sent the [information] to my general practitioner. That prompted a conversation and a whole body skin check”; 10 as neutral, including: “I did not need to contact a genetic counselor,” “information was very simplified”; no comments were negative.

The nonpersonalized “Melanoma information: prevention and early detection” booklet was also rated highly in terms of satisfaction by control (mean = 8.1, SD = 1.8) and intervention (mean = 8.3, SD = 1.7) participants.

There were no statistically significant differences in the objectively measured SEDs per day, overall, or by subgroup. The 3-month effect estimate was −16% SEDs per day in the intervention group compared with the control group overall (Table 2). Stratified by risk category, the effect estimates were −19% SEDs/day for the high-risk group, −29% for the average-risk group, and 13% for the low-risk group.

A borderline significant reduction in intentional tanning was found overall (−0.23, P = 0.06; Table 2); when stratified by risk level, the effect size was −0.33 for the high-risk group (P = 0.16), −0.38 for the average-risk group (P = 0.009), and 0.12 for the low-risk group (P = 0.67). There was no change in the sun protection index for intervention compared with control participants overall (mean difference, 0.05; Table 2); when stratified by risk group, the effect size for the average-risk group was 0.17 (P = 0.07). When the 6 components of the sun protection index were analyzed separately (summarized in Supplementary Table S2), the effect sizes were stronger for limiting time in the sun during midday hours and staying in the shade, particularly for the average-risk group (P < 0.05).

There was no evidence that skin cancer–related worry or psychologic distress and well-being scores differed between the intervention and control groups at follow-up (Table 3). Intervention participants of all risk groups reported higher confidence in identifying melanoma than controls (overall 0.40-unit increase, P = 0.008). Nonsignificant changes to risk perception were observed: on a scale of 1 to 5, the high-risk group increased by 0.36, the average-risk group had no change (−0.04), and the low-risk group reduced by 0.43 (Table 3). There were no appreciable changes to other hypothesized mediators of behavior change or views on genetic determinism.

At 3-month follow-up, health system and individual costs were generally higher in the intervention group; however, the differences were not statistically significant (Supplementary Table S3).

The findings from this pilot study demonstrate the feasibility and acceptability of giving information on personalized genomic risk of melanoma to the public. Our study population, not defined by personal or family disease history, expressed strong interest in receiving their personalized genomic risk information for melanoma (41% consent), although similar to some other population studies (42), participation was lower for younger people. Nevertheless, participants in this study were registered on a cancer research database and thus may have been more interested in genetic testing for cancer risk compared with unselected general population samples. Other large Australian health research studies that have recruited population-based samples using the Medicare database or electoral rolls achieved 16% to 18% consent rates (25, 43). Obtaining saliva samples, questionnaires, and UV dosimeter measurements by post was feasible and acceptable with minimal loss to follow-up, and genotyping was successfully completed for all but one participant.

There was high satisfaction with the delivery process for the personalized risk information, including telephone-based genetic counseling and the personalized booklet. Telephone-based counseling is commonly used for other population-level health interventions in Australia (44, 45), where vast travel distances are a barrier to accessing health services and contribute to health disparities. Internationally, telephone-based genetic counseling is becoming more common following research that the majority of recipients are satisfied with this approach (46) and evidence of lower costs compared with face-to-face counseling (47). Genetic counseling is currently the standard care for delivering genetic testing results in Australia, but scaling up this intervention to the whole population (assuming this would be acceptable) would require an increased workforce. Implementation research should evaluate alternate service delivery models, such as primary care clinicians or trained health educators.

We emphasize that with our small sample size, this pilot study was not powered to find clinically important effects with precision. This fact likely contributes to why results were mixed and CIs wide for many of the outcomes. Our preliminary findings suggest some potential beneficial impacts of the intervention on some skin cancer prevention behaviors such as reducing sun exposure and intentional tanning, in a subgroup of participants. However, the efficacy of providing personalized melanoma genomic risk to the public as a potential melanoma prevention strategy needs to be investigated in a larger study.

When stratified by genomic risk, the effect estimates appeared stronger for the average-risk group, whereas we might expect them to be stronger for the high-risk group. Because about half of the genomic variants are in pigmentation and nevus genes, there is likely to be some overlap between phenotypic risk (e.g., observable sun-sensitive characteristics and nevus counts) and genomic risk categories. Thus, one possibility for this pattern is that participants with phenotypic risk factors may already have good sun protection behaviors and therefore have less capacity to change these behaviors compared with those without phenotypic risk factors. The larger proportion of participants in the average-risk category may have increased the precision of the effect estimate for this group. The results may also be due to chance. A larger study, together with planned qualitative interviews with participants, may give further context to these results. Other potential health behavior theory–driven mediators of behavior change such as health literacy (12), family communication (48), and risk-taking behaviors (49) are also important aspects for further research.

Our analysis compared the effects of giving personalized genomic risk information versus not giving this information. A strength of our study design is that it enabled stratified analysis according to genomic risk category, as we hypothesized that the impact of personalized genomic risk information may differ according to genomic risk category. However, controls also undertook genetic testing and received their risk information after completion of the 3-month follow-up outcome measures. Thus, it was not possible to assess whether behaviors or psychosocial outcomes were influenced by the process of genetic testing per se or by the differences in waiting periods between intervention (3.5 months) and control groups (7 months) receiving their risk information. Future studies could compare “genetic testing plus risk feedback” versus “no genetic testing,” stratified by high/low sun-sensitive phenotype, as this may inform more closely a real-world scenario where decisions about testing may depend on other risk factors and genetic testing plus receipt of results is a joint intervention.

We found a nonsignificant increase in objectively measured SEDs per/day for intervention participants compared with control participants in the low-risk group; however, overall, there was no consistent pattern to suggest that participants at low-risk adopted less sun protection. One ethical concern about delivering genomic risk information is that those at lower risk may respond by becoming complacent or adopting harmful behaviors, but this has not been borne out in previous studies (13). Our previous focus group research (50) suggested that potential, unintended negative effects of genomic risk information could be minimized through the provision of educational information about melanoma risk and prevention alongside the personalized risk information, as we did.

A recent systematic review of 18 randomized and quasi-randomized controlled trials concluded there was no evidence that genomic risk information motivates preventive behaviors (13). Only one of these studies (17) focused on sun-related behavior outcomes. Glanz and colleagues randomized 73 adults with a family history of melanoma to be offered individualized risk estimates on the basis of genotyping of high-risk mutations in CDKN2A and moderate-risk variants in MC1R, versus no disclosure of genotyping results. Comparing the intervention versus control group, they found an increased frequency of skin self-examinations (P = 0.002) and wearing a long-sleeved shirt (P = 0.047) and a borderline-significant increase in the sun protection index (standardized mean difference, 0.43; 95% CI, −0.03–0.90, P = 0.07; refs. 13, 17). Other studies in the review had important limitations, including poor methodological quality with high or unclear risk of bias, being underpowered, or risk estimates, on the basis of single or few genomic variants. A recent cohort study observed improved sun protection behaviors after receiving melanoma risk based on a single-gene, common variant (rs910873 in the PIGU gene; ref. 14). However, this study lacked a control group and baseline measurement of behaviors and used nonvalidated outcome measures (14).

Melanoma is highly preventable through behavior change, and preventive behaviors for melanoma do not require extensive lifestyle modifications. There is a dearth of evidence about risk feedback and sun-related behavior change (13), and our findings indicate that this potential effect of genomic risk information warrants further investigation.

There was no evidence that the intervention increased skin cancer–related worry or psychologic distress and well-being, consistent with other research on this topic (13, 51). There was no evidence that the intervention encouraged genetic determinism or fatalistic views on the possibility of developing melanoma, similar to findings in other studies examining participants’ responses to genetic testing results (52).

Increased confidence in identifying melanoma was observed in the intervention group at 3-month follow-up. Improved confidence (self-efficacy) in identifying suspicious changes on the skin is associated with greater intention (53) and frequency of self-skin examinations (34). Future analyses will explore mediation pathways through which behavior change occurs (54, 55).

In conclusion, this pilot study demonstrated feasibility and acceptability of giving information on personalized genomic risk of melanoma to the public, with some indication of some improved preventive behaviors and no evidence of adverse psychologic outcomes. A larger trial with longer follow-up is required to evaluate the effectiveness and cost-effectiveness of this intervention as a potential novel melanoma prevention strategy.

No potential conflicts of interest were disclosed.

Conception and design: A.J. Newson, R.L. Morton, L.A. Keogh, S.J. Dobbinson, J. Kirk, G.J. Mann, A.E. Cust

Development of methodology: A.J. Newson, R.L. Morton, K. Dunlop, P.N. Butow, M.H. Law, M.G. Kimlin, L.A. Keogh, S.J. Dobbinson, J. Kirk, G.J. Mann, A.E. Cust

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.K. Smit, G. Fenton, L. Freeman, K. Dunlop, A.E. Cust

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.K. Smit, D. Espinoza, R.L. Morton, M.H. Law, M.G. Kimlin, L.A. Keogh, A.E. Cust

Writing, review, and/or revision of the manuscript: A.K. Smit, A.J. Newson, R.L. Morton, K. Dunlop, P.N. Butow, M.H. Law, M.G. Kimlin, L.A. Keogh, S.J. Dobbinson, J. Kirk, P.A. Kanetsky, G.J. Mann, A.E. Cust

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.K. Smit, G. Fenton, A.E. Cust

Study supervision: G.J. Mann, A.E. Cust

We thank Alison Brodie and Huong Tran Cam Dang for analyzing the UV dosimeter data.

This study received funding from Sydney Catalyst Translational Cancer Research Centre and The University of Sydney Cancer Strategic Priority Area for Research Collaboration (SPARC) Implementation Scheme. A.E. Cust received Career Development Fellowships from the National Health and Medical Research Council of Australia (NHMRC; 1063593) and Cancer Institute NSW (15/CDF/1-14). R.L. Morton was supported by a NHMRC Sidney Sax Fellowship (1054216). M.G. Kimlin is supported through a Cancer Council Queensland Professorial Chair in Cancer Prevention.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.