Design and implementation of an Internet‐Based cancer risk assessment tool: Use over 10 years

Abstract Background Prevention and early intervention can improve survival and quality of life across all cancers. Patient understanding of risk factors and associated actionable lifestyle changes and screening programs is not well understood by clinicians Methods An Internet‐based tool, Reduce My Risk, was created in 2009 and made available on oncolink.org. Users voluntarily completed a survey regarding demographics and cancer risk factors, and received information about their cancer risk Results Twenty eight thousand and one surveys were completed from 2009 to 2019. Median age was 26 years (18–101); 60% were females, 87% lived in North America, and 37% had at least a bachelor's degree. Users reported on behavioral/ modifiable risk factors: 13% were current smokers, 52% were current consumers of alcohol, and 8% of those had ≥14 drinks/week. Body mass index (BMI) was ≥30 in 19%; 74% of all surveys reported dietary risks and 36% reported infrequent exercise. Excess UV exposure was reported by 19%. Among women, 36% reported performing breast self‐examinations monthly, and 50% reported receiving clinician breast examinations at least once every 3 years. Sixty seven percent of men 55–75 years reported screening prostate specific antigen testing, with 50% receiving annual digital rectal examinations. Nonmodifiable risk factors included family cancer history (64%), genetic syndrome (3%), and cancer‐predisposing health conditions (26%) Conclusions Ninety‐seven percent of users reported modifiable risk factors, and 60% reported ≥4 of these risk factors. Understanding detailed characteristics of a large number of respondents has the potential to improve educational interventions to reduce cancer risk through behavioral modification and cancer screening across the general public.

modifiable risk factors: 13% were current smokers, 52% were current consumers of alcohol, and 8% of those had ≥14 drinks/week. Body mass index (BMI) was ≥30 in 19%; 74% of all surveys reported dietary risks and 36% reported infrequent exercise. Excess UV exposure was reported by 19%. Among women, 36% reported performing breast self-examinations monthly, and 50% reported receiving clinician breast examinations at least once every 3 years. Sixty seven percent of men 55-75 years reported screening prostate specific antigen testing, with 50% receiving annual digital rectal examinations. Nonmodifiable risk factors included family cancer history (64%), genetic syndrome (3%), and cancer-predisposing health conditions (26%) Conclusions: Ninety-seven percent of users reported modifiable risk factors, and 60% reported ≥4 of these risk factors. Understanding detailed characteristics of a large number of respondents has the potential to improve educational interventions to reduce cancer risk through behavioral modification and cancer screening across the general public.

| BACKGROUND
In 2003, the Institute of Medicine and National Research Council National Cancer Policy Board published a seminal report outlining the need to improve primary and secondary prevention of cancer. 1 Primary preventionreducing risk factors for the development of cancerand secondary prevention-early detection of treatable cancer-have been identified as critical targets to improve survival and quality of life across a range of cancers. 1 Risk factors for the development of cancer have been described as intrinsic, or random DNA replication errors, and non-intrinsic risk factors, some of which are modifiable. 2 Identifying risk factors for the development of cancer, including those that are both modifiable and non-modifiable, has been a key public health focus over the last half-century, supported by strengthening evidence that non-intrinsic risk factors are responsible for as many as 60%-90% of cancers in adults. 2 With this recognition have come meaningful improvements in screening, early detection, and overall survival. Assessing patients' cancer risk is a key task for public health researchers, primary care physicians, and oncology providers. A large body of research demonstrates associations with a wide swath of variables, including demographic factors, carcinogen exposures, family history, screening practices, and modifiable behaviors.
While public health researchers and clinicians often recommend targeted screening and interventions on the basis of these risk factors, scant literature exists studying individual patients' understanding of their own personal risk factors for cancer, and particularly the impact of copresence of multiple non-modifiable and modifiable risk factors. Assessing patients' understanding is critical to developing effective, targeted behavioral interventions to reduce cancer risk.
In an effort to better inform targeted interventions for those at risk of developing cancer, we produced a publicly available, Internet-based tool for individuals to answer a comprehensive, detailed list of questions related to their individual risk factors for cancer, with a goal of improving individual understanding of applicable modifiable and non-modifiable risk factors. In the present report, we summarize the results of the 28,001 surveys submitted over a 10-year period and propose future analyses and interventions that can be undertaken using these data.

| METHODS
As previously described, OncoLink (https://www.oncol ink.org) is a website housed at the Abramson Cancer Center of the University of Pennsylvania. 3 The OncoLink Reduce My Risk (originally What's My Risk?) tool was created by a group of oncology physicians and nurses in 2009 as part of the OncoLink Risk and Prevention program (https://www.oncol ink.org/risk-and-preve ntion). Beta testing and review was performed by a group of institutional and multidisciplinary specialists prior to public launch, and the program underwent a major user interface overhaul in 2018. The tool has been publicly available since 2009 on oncol ink.org. Although it is only offered in English, the tool is available to users throughout the world. It is designed to provide information about individual cancer risk after users voluntarily respond to a comprehensive survey regarding demographics and cancer risk factors. Its current version can be found at https://risk. oncol ink.org/. The survey was not advertised, but as the OncoLink website sees over 500,000 visitors per month, a portion of these patients discovered the tool through the OncoLink website or through searching for information about cancer risk information using a search engine. Users also learned about the tool through OncoLink cancer risk-related web-log ("blog") and social media postings referencing the tool. While the survey's underlying algorithm remained unchanged over time, improvements were made to the user interface to make the content more accessible with simpler language, as well as to update content as new research became available.
Survey questions used in the tool are shown in Table S1; respondents entered numeric answers or selected from the responses shown in Tables 1-4. Respondent answers are collected anonymously, with no identifiable personal data other than Internet protocol (IP) addresses, which allow identification and deletion of duplicate entries. After the survey is complete, an individualized "Risk Report" is generated to inform the user of specific cancer risks related to modifiable, non-modifiable, and familial/ genetic risk factors identified from the user's survey. Risks are not quantified, but rather the types of cancers associated with certain risk factors (e.g., tobacco use and lung cancer) are reported. At the same time, healthy behaviors are identified and reinforced (e.g., participating in appropriate cancer screening), with education provided regarding cancer screening and risk modification, where applicable. Each survey's responses are maintained in a secure database. Survey data from this convenience sample frame were extracted, and descriptive statistics were used to summarize results. Research related to these data has been approved by the University of Pennsylvania Institutional Review Board, which determined that informed consent was not required.

| RESULTS
At the time of analysis, 28,001 surveys had been submitted. Sixty percent of respondents were female, and most (82%) were between 18 and 45 years old (median age 26 years, range 18-101). Seventy-six percent identified as white/non-Hispanic, 9% multiple/other, 6% Asian/ Pacific Islander, 5% Hispanic/Latino/Latina, 3% African American or Black, and 1% Native American/Aleutian/ Eskimo. Further responses to demographic questions are summarized in Table 1, and responses from all demographic groups are included in this analysis. The majority of respondents (87%) lived or spent the majority of their life in North America, followed by the United Kingdom (9%), Australia (2%), and India (1%) (Figure 1).
Modifiable risk factors with respect to smoking, alcohol, and recreational drugs are summarized in Table 2A.
Thirteen percent of respondents were current smokers and 23% were former smokers. Most current and former smokers separately reported smoking one pack per day or less (96% and 97%); 60% percent of current smokers reported smoking for less than 10 years, and 26% of former smokers quit within the year prior to taking this survey. Only 10% of respondents reported cigar or pipe tobacco use and 4% oral snuff/ chew/quid. Ten percent of respondents reported current significant secondhand smoke exposure, and 19% reported prior exposure. Fifty-two percent of individuals reported consuming alcohol, and an additional 5% reported previous heavy alcohol use. Among alcohol users, 78% reported drinking 7 or fewer drinks per week, and 8% 14 drinks or more. Sixty-eight percent of respondents reported never smoking marijuana. Among the 10% of respondents who reported current marijuana use, the majority (63%) reported marijuana use "more than a few times" per month. A minority of respondents reported using cigar or pipe tobacco (10%), oral snuff/chew/quid (4%), areca nut or betel leaf (1%), or "exotic smoking" (1%).
To assess risk factors related to obesity, exercise, and diet, individuals first reported height and weight, from which body mass index (BMI) was calculated. Respondents then answered questions related to exercise and dietary patterns known to be associated with the development of certain cancers (Table 2B). Five percent had an underweight BMI <18.5 kg/m 2 , 50% normal (18.5-25 kg/m 2 ), 25% overweight (25-30 kg/m 2 ), 11% with class I obesity (30-35 kg/m 2 ), 5% class II obesity (35-40 kg/m 2 ), and 4% with class III obesity (≥40 kg/m 2 ). Only 16% of users reported at moderate-to-vigorous exercise five or more times per week. In terms of diet, 34% of individuals reported eating red meat at least three times per week, and 18% consume charred or well-done meats. Twenty-five percent consumed fermented, smoked, pickled, or salt-preserved foods. Fifty-nine percent often ate white bread, white rice, or processed grains. Ten percent of respondents consumed vegetarian or vegan diets.
Sexual practices associated with certain cancer risks were also surveyed (Table 2C). Forty-three percent of individuals reported age of first sexual intercourse before age 18; 21% reported not having had sexual intercourse. Eighteen percent of respondents reported at least 10 or more partners, 69% had engaged in oral sex, and 16% had engaged in receptive anal intercourse.
Responses regarding risk factors for the development of skin cancers are summarized in Table 2D. Of note, 1% of respondents reported a personal history of diagnosed nonmelanoma skin cancer. Twenty-nine percent of individuals had a fair complexion, 27% reported childhood or adolescent blistering sunburns, 24% reported freckles and sun spots, and 15% had more than 50 moles or birthmarks. Modifiable risk factors included sunbathing and frequenting tanning salons (11% and 20%, respectively). The majority of respondents reported no occupational or environmental exposures (Table 2E), but a small proportion of individuals reported sustained exposure to asbestos (560, 2%), hydrocarbons (759, 3%), heavy metals (582, 2%), mustard gas (100, <1%), industrial dyes (625, 2%), leather, Have you ever been exposed to radiation by: a radiation accident? (e.g., Chernobyl) No 100% 27,878 Yes 0% 120 Have you ever been exposed to radiation by: at least 10 diagnostic medical tests?  Table 3A. Among all women, 560 (3%) had taken hormone replacement therapy for at least 2 years. Among post-menopausal women, this proportion was 21% (419/2007), with an additional 141 women using long-term hormone replacement therapy before menopause. One-hundred forty-one (1%) women reported that their mother took diethylstilbestrol (DES) while pregnant with the respondent, and 23 (<1%) women reported themselves taking DES while pregnant.
Preventive care practices surveyed are summarized in Table 3B. Sixty-two percent of respondents reported having received the hepatitis B virus (HBV) vaccine. Among women of any age, 28% reported receipt of the human papilloma virus (HPV) vaccine; 49% of women between 18 and 25 years received the HPV vaccine. Twelve percent of all men received the HPV vaccine, in contrast to 23% of men between 18 and 25 years. Among women surveyed, 36% reported performing breast selfexaminations monthly, and 50% reported receiving a breast examination by a clinician at least once every 3 years. Among 11,120 male respondents, 51% reported performing at least monthly testicular self-examinations. Sixty seven percent of men between 55 and 75 years reported undergoing screening prostate specific antigen testing, and 50% of men in this age group received annual digital rectal examinations. Respondents answered personal medical history questions related to 34 conditions known to be associated with the development of specific cancers (Table 3C). Gastroesophageal reflux disease (GERD) was the most commonly reported item, present among 10% of respondents, followed by a diagnosis of HPV among 7% of respondents.
Seventy-one percent of respondents reported a family history of cancer (Table 4A-C), the majority of those reporting two or fewer family members with cancer. The most frequently reported cancers were breast cancer (24%) and lung cancer (both non-small-cell and small-cell lung cancers, 19%). Nonmetastatic cancer was reported in 37% of family members, metastatic cancer in 28%, and unknown metastatic state in 28%. Maternal grandmother (19%) and grandfather (16%) were the most common family members affected, followed by mother (16%) and paternal grandmother (15%). Thirteen percent of family members were diagnosed with cancer within the year prior to the individual submitting this survey.  Although this tool was freely available, demographic features of this convenience sample frame cohort differ from the general population as respondents skew younger (median age 26 years, 82% ≤45 years), female (60%), White/ non-Hispanic (76%), and better-educated compared to the larger US and world populations. These characteristics are largely consistent with the demographics of internet surveys, which skew females, White/non-Hispanic, and college-educated. 4 The majority of respondents are from North America, followed by other English-speaking countries, likely because the tool was developed in the United States and published in English. Importantly, the T A B L E 4 (Continued) generalizability of this data is limited by the fact that the majority of responses come from white, North American individuals. Future surveys should specifically seek to understand risks among non-white populations, both within and outside of North America. Substance use is an important modifiable risk factor for the development of multiple cancers. The majority of respondents, 64%, reported never smoking tobacco, and only 13% are current smokers. This proportion is consistent with recent estimates of smoking prevalence and also is a testament to the successful public health programs that have reduced smoking from peaks of over 50% of men and 30% of women. 5 The fact that 26% of former smokers quit within the year prior to submitting this survey raises the question of whether those individuals were referred to the OncoLink Reduce My Risk tool by a clinician, or whether concern about future cancer risk led to both smoking cessation and seeking out cancer risk information. Although the majority of respondents do not use tobacco, 52% consume alcohol, a known carcinogen that is currently the target of multiple public health campaigns. 6 Notably, the American Society of Clinical Oncology (ASCO) released a statement in 2017 underlining the increased cancer risk associated with even light alcohol consumption, additionally calling into question purported cardiovascular benefits. 7 Among the respondents surveyed, though only 8% of those who drink alcohol would be considered heavy drinkers, over half of respondents engage in behavior that can be modified to meaningfully reduce their cancer risk. This represents an area in which patient counseling and education may have significant impact, and should be an area of focus for clinicians. Interestingly, the proportion of survey respondents who drink alcohol (52%) is actually lower than the general population (70% in the past year). 8 This is only partially explained by the fact that 30% of respondents were under 21 years, the legal age of alcohol consumption in the United States, and that 60% of respondents were female, as women have been reported to consume alcohol at lower rates than men 9 ; even analyzing the group of men ≥21 years, only 65% of respondents reported alcohol use. With respect to marijuana, although only 10% respondents reported current marijuana use, 63% of those individuals reported using marijuana at least several times per month. Among those who use marijuana, an estimated 20% consume marijuana daily, 10 underscoring the need to understand the risks associated with heavy marijuana consumption. As the co-presence of other cancerrelated risk factors confounds the ability to determine the cancer risk associated with marijuana specifically-(e.g., daily marijuana use is associated with tobacco use 10 ) longitudinal research among individuals who consume marijuana, without concomitant tobacco use, is needed. 11 F I G U R E 1 Geographic location of respondents outside of North America ("Where have you spent the majority of your life?").
Obesity, diet, and exercise have significant bearing on the risk of developing several cancers. Survey respondents were more likely to report non-obese BMI compared to the general population, 82% vs. 60%.[https://www. cdc.gov/obesi ty/data/adult.html] Among respondents ≤19 years, 90% had non-obese BMI, compared to 79% of 12-19-year-olds in the general population. 12 With the exception of the 59% of individuals who often consume processed carbohydrates, the majority of respondents' diets are low in foods whose preparation is associated with gastric and colorectal cancers. Despite the favorable diet and BMI profiles reported, only 16% of respondents reported moderate exercise at least 5 times per week. Thus, nonobese individuals may represent an important group to target to recommend increased exercise to reduce the risk of certain cancers. 13 A unique aspect of this survey is that captures both common and uncommon environmental and occupational exposures. For instance, voluntary sun and tanning salon exposures were reported among 3122 (11%) and 5690 (20%) of respondents, respectively. With respect to ionizing radiation exposure, only 37% of individuals reported that their homes have been tested for and do not contain excess radon; 381 surveys reported known radon exposure, leaving 62% of respondents with unknown potential risk with regard to radon exposure. Similarly, 1843 (7%) individuals reported receiving at least 10 diagnostic medical tests that rely on ionizing radiation. By contrast, only a very small number of surveys reported exposures to carcinogenic occupational agents, asbestos, and DES. In a similar vein, this survey captures individuals with uncommon personal medical histories, such as Fanconi anemia (54 surveys), xeroderma pigmentosum (45 surveys), primary sclerosing cholangitis (43 surveys), and others. These respondents warrant further investigation to learn more about their additional modifiable and nonmodifiable risk factors.
An additional strength of this survey is the depth of individuals' family history of cancer. In line with national trends, the most commonly reported cancers include breast (24%), lung (19%), colorectal (10%), prostate cancer (9%), and melanoma (7%). These reports appear to approximate cancer trends across recent decades (reference). Although other cancer types are overrepresented, such as brain (6%), leukemia, ovarian (5%), and cervical (4%) cancers, these in particular may be subject to inaccurate recall: Gynecologic cancers may be misclassified and brain metastases may be incorrectly considered primary brain tumors. Twenty-eight percent of respondents, for instance, answered "I don't know" if the family member had metastatic disease.
Preventive care questions represent an interesting aspect of this survey, potentially guiding future interventions, but raise an important limitation of this survey. Namely, breast clinical and self-examinations, breast cancer chemoprevention, prostate and testicular cancer screening, and HPV vaccination, are subject to discordant national guidelines, as well as guidelines that have changed since this survey was first made available. Additional limitations include potential self-selection bias due to the fact that use of the survey is voluntary, and users of this survey may have a specific reason for seeking education about cancer risk reduction. For instance, 3% of respondents report that a family member was diagnosed with cancer within the last month, and 10% within the last year. A key limitation relates to the anonymous nature of this survey: Participants were not prevented from completing the survey more than once if submitting form computers with different IP addresses, potentially influencing results. Moreover, an individual's risk factors can change over time, and this is not captured by the present survey methodology.
While the purpose of the present work is to report cancer risk patterns in a large group of patients voluntarily seeking self-education regarding cancer risk, future directions include analyses of the interactions of reported risk factors in an effort to develop targeted risk reduction interventions by clinicians and public health workers. As non-intrinsic risk factors for the development of cancer (those that are not related to random DNA replication errors) include both non-modifiable variables and modifiable behaviors, the results of this survey may allow for the exploration of interactions between multiple variables and behaviors, ultimately leading to better-targeted interventions by clinicians and public health campaigns. For instance, alcohol use and cigarette smoking, which synergistically increase the risk of p16-negative, HPV-negative oropharynx and other aerodigestive cancers, are modifiable behaviors. Yet cigarette smoking also leads to worse outcomes for p16-positive, HPV-related oropharynx cancer, as compared to the high rates of cure now achievable for HPV-related oropharynx cancer in non-smokers. 14 Prior HPV exposure is not a modifiable risk factor, but early HPV vaccination and avoidance of tobacco are behaviors that, if targeted at an early age, can meaningfully reduce future cancer risk. Other examples include sun exposure in fair-skinned individuals, smoking among occupationally exposed workers, and diet and obesity in individuals at increased risk of breast and colon cancers. Family histories could be examined to determine whether individuals at risk for hereditary cancer syndromes are receiving appropriate screening. In a future iteration of the OncoLink Reduce My Risk tool, patients who take this survey upon referral from their primary care physician could consent to longitudinal follow-up to determine the impact of risk reduction recommendations. Future work may also focus