The Impact of Receiving Predictive Genetic Information about Lynch Syndrome on Individual Colonoscopy and Smoking Behaviors

Personalized medicine refers to the tailoring of preventive, diagnostic, and therapeutic interventions to the characteristics of individuals, using advanced biomedical technologies (1). Its ability to identify genetic susceptibilities to preventable diseases, obtain unequivocal diagnostic results, and tailor drug therapies on an individual basis promises to revolutionize health care and improve population health (2, 3).

Many promises of personalized medicine have been realized, especially in cancer treatment (4). Its impact in cancer prevention, such as predictive genetic testing—assessing whether individuals with a family history, but no previous diagnosis, of a genetic disorder carry their family's deleterious genetic mutation and are at elevated risk of developing the disease in the future—has also been demonstrated. For example, positive results from predictive genetic testing for hereditary breast cancer or Lynch syndrome are associated with increased adherence to risk-reducing strategies, such as prophylactic surgery or surveillance (5), although behavioral change from predictive genetic information has generally been deemed less than expected (6, 7).

Building on this literature, this study investigated whether receiving a positive or negative result from predictive genetic testing for Lynch syndrome was associated with change in individual health behaviors in Australasians. The lifetime probability of developing colorectal cancer is estimated to be 54% to 74% for men and 30% to 52% for women with Lynch syndrome, compared with 5% to 6% for the general population (8–10). Various health behaviors have been demonstrated to reduce the risk of hereditary colorectal cancer in Lynch syndrome families, such as colonoscopy use (11, 12), smoking (13, 14), and other lifestyle behaviors, including diet (15). On the basis of data availability, this study focused on colonoscopy and smoking behaviors. Colonoscopy is recommended for the prevention and early detection of colorectal cancer for individuals at elevated risk (11), and its use has been shown to change upon genetic testing for Lynch syndrome (16–23). Smoking represents a broader array of health behaviors that also affect the incidence of colorectal cancer, but the impact of genetic testing for Lynch syndrome on smoking, or any other lifestyle behaviors, is yet to be assessed.

From a simple health economics model, we generated hypotheses on the impact of the new information that the advanced biomedical technology provides. We then tested the hypotheses against individual-level panel data. Our study distinguished individuals with a positive genetic predisposition (“carriers”) from those without (“noncarriers”), as well as individuals who received their genetic testing results (“receivers”) from those who did not (“nonreceivers”).

In what follows, we present our hypotheses, data, and empirical strategy. We then describe our results. The final section discusses and concludes.

We developed a model of utility maximization under uncertainty that incorporated Ehrlich and Becker's concepts of self-insurance and self-protection (24). The full model is presented in Supplementary Materials and Methods. Briefly, our model characterized the occurrence of colorectal cancer as a probabilistic event, in which carriers have a higher probability of developing colorectal cancer, compared with noncarriers (8, 9). The probability of developing colorectal cancer was defined as a combination of exogenously determined genetics and endogenously determined health behaviors, namely, colonoscopy use and smoking. Receivers, unlike nonreceivers, have an opportunity to change their behaviors, based on the results that are returned to them. Our model predicts that carriers who receive their genetic information are more likely to undergo colonoscopy but less likely to smoke, compared with carriers who do not receive their genetic information, as the opportunity cost of foregoing healthy behaviors rises with increasing risk. Our model also predicts that noncarriers who receive their genetic information are less likely to undergo colonoscopy but more likely to smoke, compared with noncarriers who do not receive their genetic information, as the opportunity cost of investing in healthy behaviors rises with decreasing risk. The predictions were tested empirically.

We used data from the Australasian Colorectal Cancer Family Registry (ACCFR) for the statistical analysis. Ethics approval for the analysis was obtained from the Office of Research Ethics at the University of Toronto (Toronto, Canada). Detailed descriptions of the ACCFR were provided by Newcomb and colleagues (25) and Jenkins and colleagues (26). Here, we provide an overview of the ACCFR registrant recruitment, data collection, and genetic testing.

Between 1997 and 2007, the ACCFR used several recruitment strategies. Individuals who were diagnosed with their first colorectal cancer between the ages of 18 and 59 years (1997–2001) or 18 and 49 years (2001–2006) were identified from the population-based Victorian Cancer Registry. Cases of colorectal cancer or other Lynch syndrome–related cancers (i.e., of the endometrium, stomach, small intestine, urinary tract, or central nervous system) were also identified from seven family cancer clinics across Australia and New Zealand. From these clinics, individuals at high risk of Lynch syndrome were also identified, using the Amsterdam criteria (27). All identified individuals, known as probands, were invited to register with the ACCFR. The probands' first- and second-degree relatives, as well as other relatives with previous diagnosis of any cancer, were also invited to register with the ACCFR (25). Therefore, the ACCFR registrants comprised individuals with previous diagnosis of colorectal cancer or other Lynch syndrome–related cancers and their first-degree, second-degree, and more distant relatives with or without any cancer.

The ACCFR registrants were asked to complete an interviewer-administered questionnaire at enrollment (“baseline”) and after 5 years (“follow-up”). The baseline questionnaire asked about the registrants' socio-demographic factors (e.g., age, sex, marital status, and education), health behaviors (e.g., colonoscopy use and smoking), and medical conditions (e.g., previous cancer diagnosis). The follow-up questionnaire obtained updates on the registrants' health behaviors and medical conditions. All reported incidents of Lynch syndrome–related cancers were verified against medical records (25, 26).

The ACCFR registrants were asked to provide a blood sample for the purpose of conducting research and not for the purpose of genetic testing per se. Instead, blood samples of the registrants whose families had previously identified deleterious mutations in the mismatch repair (MMR) or EPCAM genes were tested for their families' genetic mutations by the ACCFR. Therefore, the registrants did not choose to undergo genetic testing, and the availability of their results at the ACCFR was free of potential selection bias associated with the choice to undergo genetic testing. The registrants were informed of the availability of their results and referred to a genetic counseling clinic to receive formal retesting and their results free of charge (28–30).

The study population for the analysis included all registrants of the ACCFR that would have been eligible for predictive genetic testing for Lynch syndrome, having (i) no previous diagnosis of colorectal cancer at follow-up so that the testing would be truly predictive; and (ii) a family with a previously identified deleterious mutation in the MMR or EPCAM genes so that testing for that mutation would conclusively distinguish carriers from noncarriers. A total of 1,753 registrants of the ACCFR satisfied the conditions and had been tested and identified as either carriers or noncarriers of their families' deleterious mutations in the MMR or EPCAM genes. However, 1,246 of the 1,753 individuals had insufficient information on whether or when their genetic testing results had been disclosed to them and were excluded from the analysis. Therefore, the study sample for the analysis comprised 507 individuals who had sufficient information on whether or when their results had been disclosed to them.

We used logit regression to estimate the probability that an individual underwent colonoscopy or smoked between baseline and follow-up (i.e., yes/no), I_B-F or S_B-F, respectively, as a function of the carrier status (i.e., carrier/noncarrier), G; the receiver status at follow-up (i.e., receiver/nonreceiver), R_F; previous cancer diagnosis at follow-up (i.e., yes/no), H_F; the interaction of the carrier and receiver statuses and previous cancer diagnosis at follow-up, G × R_F, G × H_F, R_F × H_F, and G × R_F × H_F; a vector of the socio-demographic (i.e., age at follow-up, sex, marital status, and education) and colorectal cancer awareness (i.e., degree of relatedness to family probands) factors, D_F; and colonoscopy use or smoking at baseline (i.e., yes/no), I_B or S_B, respectively. In other words:

Table 1 provides the full variable list, including the scales and sources of the variables.

Colonoscopy use and smoking between baseline and follow-up (i.e., I_B-F and S_B-F), as well as before baseline (i.e., I_B and S_B), were defined in terms of propensity (i.e., yes/no) to undertake the behaviors. We adjusted for the respective behaviors at baseline (i.e., I_B or S_B), following suggestions by others to account for any baseline imbalance in the outcome variables (31). We repeated the analysis on colonoscopy use within 2 years of follow-up and smoking at follow-up, adjusting for colonoscopy use within 2 years of baseline and smoking at baseline, respectively, to ensure the relevance of the behaviors at those time points. We also conducted subgroup analyses only on the study participants who had undergone colonoscopy or smoked before baseline (i.e., I_B = yes or S_B = yes) to eliminate any baseline imbalance.

The carrier status (i.e., G) distinguished carriers, who carried any deleterious mutations in the MMR or EPCAM genes, from noncarriers, who did not carry their families' deleterious mutations in the MMR or EPCAM genes.

The receiver status at follow-up (i.e., R_F) distinguished 392 receivers, who had received their genetic testing results before follow-up, from 115 nonreceivers, who had either received their results after follow-up (n = 80) or never received them (n = 35), among the 507 study participants.

The receiver status at follow-up (i.e., receiver/nonreceiver) only allowed for the comparison of those who had received their results versus those who had not, which was not a pre- and postintervention study design. To address this shortcoming, we conducted a subgroup analysis that directly compared 43 receivers, who had received their results between baseline and follow-up, and 80 nonreceivers at follow-up, who had received their results after follow-up, in a pre- and postintervention study design.

Previous cancer diagnosis at follow-up (i.e., H_F) distinguished those who had been diagnosed with some non-colorectal cancer before follow-up from those who had not. We adjusted for this factor and interacted it with the carrier and receiver statuses at follow-up, following the observations of another study reporting that previous cancer diagnosis affected individual health behaviors differentially among receiver/nonreceiver carrier/noncarrier subgroups (32). None of the study participants had had colorectal cancer at follow-up, as we selected for individuals with no previous diagnosis of colorectal cancer at follow-up. We also conducted subgroup analyses only on the study participants with no previous cancer diagnosis at follow-up (i.e., H_F = no) to eliminate any endogeneity associated with the variable.

We controlled for the study participants' age at follow-up (i.e., years), sex (i.e., female/male), marital status (i.e., married or common-law/single), education (i.e., less than high school/vocational, training, or high school/college, university, or more), and degree of relatedness to their family probands (i.e., self or identical twin/first/second/third/higher degree, factors previously identified to affect individual health behaviors that modulate the risk of cancer (32–35). The degree of relatedness to family probands also addressed potential heterogeneity in the study participants' awareness of colorectal cancer from their family relationships.

All statistical tasks were performed with the Stata (version 12.1) software (36). The estimation results on the carrier and receiver statuses and previous cancer diagnosis at follow-up and their interaction terms were linearly combined to compare the receivers and nonreceivers in the various carrier/noncarrier cancer/no-cancer subgroups to test the hypotheses from the previous section.

Table 2 reports the descriptive statistics on the study participants and nonparticipants. The study participants were, on average, in their mid-40s at baseline, more often female, and predominantly married or in a common-law relationship and had high-school education or more. Before baseline, 71% and 47% of the study participants had ever undergone colonoscopy or smoked, respectively, whereas 47% and 17% had undergone colonoscopy within 2 years or were currently smoking, respectively. By baseline, 13% had been diagnosed with some non-colorectal cancer. Compared with the 507 study participants, the 1,246 nonparticipants were significantly older, more often male, more often single, with lower levels of education, more distant relatives to their family probands, and less likely to have undergone colonoscopy.

The study participants were stratified by the carrier and receiver statuses. The receivers were older than the nonreceivers, suggesting that the receivers enrolled in the ACCFR earlier than the nonreceivers. The carriers were younger than the noncarriers, which was expected, as identifying individuals with no previous diagnosis of colorectal cancer leads to selection for carriers who are younger than noncarriers. At baseline, the receivers had higher rates of colonoscopy use, compared with the nonreceivers, likely because the receivers were older than the nonreceivers and already under colon screening. Indeed, aging is a good predictor of colonoscopy use (37). The carriers had higher rates of previous diagnosis of non-colorectal cancer, compared with the noncarriers, which was expected, as the carriers' genetic mutations predisposed them to colorectal cancer, as well as other cancers (8). No other significant differences or indications of systemic differences were observed among the receiver/nonreceiver carrier/noncarrier subgroups.

Table 3 reports the estimation results of the logit regression models on the likelihood of undergoing colonoscopy or smoking between baseline and follow-up. Odds ratios (ORs) greater (or less) than one indicate higher (or lower) probabilities for undergoing colonoscopy or smoking versus not undergoing colonoscopy or smoking, respectively, between baseline and follow-up.

There were significant associations between the likelihood of undergoing colonoscopy between baseline and follow-up and the single and interaction terms of the carrier and receiver statuses and previous cancer diagnosis at follow-up. The study participants who discovered that they carried a genetic mutation predisposing to colorectal cancer (i.e., receiver carriers) were more likely to undergo colonoscopy compared with those who were also carriers but had not received their results (i.e., nonreceiver carriers). However, this was only true for the study participants with no previous cancer diagnosis (OR = 13.124, P < 0.001) and not for those with previous cancer diagnosis (OR = 1.502, P = 0.698). The study participants who discovered that they did not carry a genetic mutation predisposing to colorectal cancer (i.e., receiver noncarriers) were not different from those who were also noncarriers but had not received their results (i.e., nonreceiver noncarriers) in their colonoscopy use (OR = 1.213, P = 0.587 for no previous cancer diagnosis; OR = 0.702, P = 0.717 for previous cancer diagnosis).

There were also significant associations between the likelihood of smoking between baseline and follow-up and the single and interaction terms of the carrier and receiver statuses and previous cancer diagnosis at follow-up. The study participants who discovered that they did not carry a genetic mutation predisposing to colorectal cancer (i.e., receiver noncarriers) were more likely to smoke compared with those who were also noncarriers but had not received their results (i.e., nonreceiver noncarriers). This was because the receiver noncarriers were less likely to quit smoking, and not because the receiver noncarriers were more likely to take up smoking, compared with the nonreceiver noncarriers (see Supplementary Table S1). However, this was only true for the study participants with no previous cancer diagnosis (OR = 4.147, P = 0.037) and not for those with previous cancer diagnosis (OR = 0.025, P = 0.089). The study participants who discovered that they carried a genetic mutation predisposing to colorectal cancer (i.e., receiver carriers) were not different from those who were also carriers but had not received their results (i.e., nonreceiver carriers) in their smoking behavior (OR = 0.802, P = 0.702 for no previous cancer diagnosis; OR = 1.511, P = 0.800 for previous cancer diagnosis).

The ORs on the other explanatory variables were consistent with expectations. For colonoscopy, the age terms suggested increasing colonoscopy use until 57 years of age and decreasing colonoscopy use thereafter. Being female versus male was associated with a higher likelihood of undergoing colonoscopy. Closer relatives to family probands had a higher likelihood of undergoing colonoscopy, compared with more distant relatives. For smoking, the age terms suggested decreasing smoking with age. Being married or in a common-law relationship versus being single was associated with a lower likelihood of smoking. Higher education was associated with a lower likelihood of smoking. For both colonoscopy and smoking, each behavior at baseline was a significant and positive predictor of itself between baseline and follow-up. The area under the receiver operating characteristic (ROC) curve, Somers D_xy rank correlation, and Brier score indicated good predictive abilities of the models.

Supplementary Tables S2 and S3 report the results of the subgroup analyses, estimated only on the study participants who had undergone colonoscopy or smoked before baseline, respectively. Supplementary Table S4 reports the results of the subgroup analyses, estimated only on the study participants with no previous cancer diagnosis at follow-up. The results presented in Supplementary Tables S2–S4 were in agreement with those presented in Table 3, although the results in Supplementary Tables S2–S4 were lacking power, due to the smaller sample sizes.

Table 4 reports the estimation results of the logit regression models on the likelihood of undergoing colonoscopy within 2 years of follow-up or smoking at follow-up. Because some of the interaction terms involving previous cancer diagnosis at follow-up did not have any observations and could not be estimated, these models were estimated on the study participants with no previous cancer diagnosis at follow-up.

The results presented in Table 4 were in agreement with those presented in Table 3, although the results in Table 4 were lacking power, due to the smaller sample sizes.

Table 5 reports the estimation results of the logit regression model on the likelihood of undergoing colonoscopy between baseline and follow-up, in a subgroup analysis that directly compared the 43 receivers, who had received their results between baseline and follow-up, and the 80 nonreceivers at follow-up, who had received their results after follow-up, in a pre- and postintervention study design. Because some of the interaction terms involving previous cancer diagnosis at follow-up did not have any observations and could not be estimated, this model was estimated on the study participants with no previous cancer diagnosis at follow-up. Because none of the 43 and 80 study participants were first-degree relatives to their family probands, the degree of relatedness to family probands variable had four, instead of five, categories. For smoking, because the 43 receivers, who had received their results between baseline and follow-up, were lacking information on smoking at follow-up and automatically omitted from the estimation, the model could not be estimated.

The results presented in Table 5 were in agreement with those presented in Table 3, although the results in Table 5 were lacking power, due to the smaller sample size. For example, the carriers who had received their results between baseline and follow-up (i.e., receiver carriers) were more likely to undergo colonoscopy compared with the carriers who had not received their results until after follow-up (i.e., nonreceiver carriers; OR = 6.121, P = 0.088).

We found strong evidence of a higher likelihood of undergoing colonoscopy for individuals with a family history of Lynch syndrome, but no previous cancer diagnosis, upon learning that they were carriers of a genetic mutation predisposing to colorectal cancer. There was also strong evidence of a lower likelihood of quitting smoking for individuals with a family history of Lynch syndrome, but no previous cancer diagnosis, upon learning that they were noncarriers of a genetic mutation predisposing to colorectal cancer. These empirical findings were consistent with the hypotheses generated from our simple health economics model. They also agreed well with previous studies that reported substitutive effects between genetic endowments and personal health behaviors in economic (38–40) and clinical (16–23) literature. They were also congruent with a previous study that reported modulating effects of previous cancer diagnosis (32). The empirical findings were robust to factors analyzed in subsequent and subgroup analyses, involving the recent colonoscopy use and current smoking measures and the pre- and postintervention study design.

However, no empirical evidence was found to support the hypotheses that individuals with a family history of colorectal cancer are less likely to undergo colonoscopy upon learning their noncarrier status and also more likely to quit smoking upon learning their carrier status. These discrepancies may imply the receivers' misunderstanding of their genetic status and/or difficulty in accepting the new information. Indeed, some of the previous studies also found noncarriers who continued to undergo colonoscopy (18) and identified the family history of colorectal cancer, as opposed to new information obtained through predictive genetic testing, as the most significant attribute to one's perceived risk (41). Furthermore, a recent study using the ACCFR data identified individuals from colorectal cancer-causing mutation-carrying families who perceived their risk of having colorectal cancer as high and did not receive or refused genetic testing results (28). A lack of guidance on appropriate colon screening behaviors may have been another factor, and further research on the screening advice given to receivers by their physicians is required. The discrepancies may also imply the receivers' difficulty in overcoming an addiction and/or understanding that smoking is associated with high risk of colorectal cancer (13, 14). Alternatively, the lack of significant evidence may reflect the fact that the predicted change from “use” to “no-use” would not be captured in our data if the study participants who received their genetic testing results between baseline and follow-up had undergone colonoscopy or smoked after baseline but before receiving their results.

Our study had many strengths. To the best of our knowledge, it was the first to investigate the impact of receiving predictive colorectal cancer genetic information on individual smoking behavior and offered the longest follow-up duration of 5 years among the studies of its kind on colorectal cancer (16–23). However, there were several limitations.

Of the 1,753 individuals in the ACCFR who were potentially eligible for inclusion in our study, 1,246 had insufficient information on whether or when their genetic testing results had been disclosed to them and were excluded from the analysis. Our descriptive statistics identified significant differences in age, sex, marital status, education, degree of relatedness to family probands, and colonoscopy use between the study participants and nonparticipants. Being female and receiving family support have been shown to be positively associated with individual health behaviors that modulate the risk of colorectal cancer (32, 37). Consistent with these observations, the study nonparticipants, who were more often male and more distant relatives to their probands, compared with the study participants, were less likely to have undergone colonoscopy. Therefore, the findings of our study might have been biased toward predicting healthier behaviors than it would have, had it included the study nonparticipants from the study population. In other words, the higher likelihood of undergoing colonoscopy for individuals upon learning that they were carriers and the lower likelihood of quitting smoking for individuals upon learning that they were noncarriers might in fact be lower in a broader population.

Because the ACCFR registrants did not choose to obtain genetic testing, the availability of their results was free of potential selection bias associated with the choice to undergo genetic testing. Although there was potential selection bias associated with the choice to receive genetic testing results, our descriptive statistics found no indications of systemic differences between the receivers and nonreceivers other than age, which was adjusted for in all our regression models. Furthermore, the majority of the nonreceivers at follow-up (i.e., 80/115) did receive their results after follow-up, suggesting against inherent differences between the receivers and nonreceivers in their choice to receive genetic testing results. Rather, the receiver status at follow-up was an artificial distinction, made to take advantage of the differential timing in the receipt of the results. Nevertheless, this being an observational study, there remain concerns that the receiver/nonreceiver carrier/noncarrier subgroups were inherently different from each other on some unmeasured characteristics. Specifically, the lower likelihood of quitting smoking for individuals upon learning that they were noncarriers was driven by the particularly low rate of smoking in the nonreceiver noncarriers. Our study must be replicated on different populations, to test not only the generalizability of our study findings but also their robustness.

For the majority of the receivers (i.e., 343/392), the exact dates of their receipt were unavailable. As some of these receivers had likely received their results before baseline, baseline for the receivers did not necessarily represent “preintervention.” Therefore, the receiver status at follow-up only allowed for the comparison of those who had received their results versus those who had not, which was not a pre- and postintervention study design. Assuming that behavioral change from this new information generally occurs shortly after its receipt, we took a conservative approach by including these individuals in our empirical analysis and still discovered significant effects. Furthermore, our analysis on a pre- and postintervention study design subgroup found consistent results, although the results were lacking power, due to the smaller sample sizes. Future studies may benefit from larger sample sizes.

In conclusion, our empirical analysis identified intended and unintended consequences of predictive genetic testing for colorectal cancer, suggesting that the impact of personalized medicine on disease prevention is more intricate than generally expected. The exact consequences of providing individuals with new information about their genetic disease risk will likely differ from one disease to another, with variations in the efficacy of the intervention, penetrance of the genetics, effectiveness of the preventive measures, and individual attitudes toward risk. Therefore, our research findings highlight the need for a fulsome assessment on the benefits and costs of personalized preventive interventions from both the clinical and societal perspectives that take account of individual responses to the resulting new information.

No potential conflicts of interest were disclosed.

Conception and design: J.S.-M. Kim, P.C. Coyte, M. Cotterchio, L.A. Keogh, A. Laporte

Development of methodology: J.S.-M. Kim, P.C. Coyte, L.A. Keogh, A. Laporte

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Gaff

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.S.-M. Kim, P.C. Coyte, A. Laporte

Writing, review, and/or revision of the manuscript: J.S.-M. Kim, P.C. Coyte, M. Cotterchio, L.A. Keogh, L.B. Flander, C. Gaff, A. Laporte

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.S.-M. Kim

Study supervision: J.S.-M. Kim, P.C. Coyte

The authors are grateful to Dr. Mark A. Jenkins and Ms. Judi Maskiell for their assistance.

J.S.-M. Kim was funded by the Vanier Canada Graduate Scholarship and the Health Care, Technology, and Place Doctoral Fellowship from the Canadian Institutes of Health Research to pursue her Ph.D. degree, during which time the study was conceived and conducted. This work was supported by grant UM1 CA167551 from the National Cancer Institute and through cooperative agreements with the following CCFR centers: Australasian Colorectal Cancer Family Registry (U01 CA074778 and U01/U24 CA097735) and Ontario Familial Colorectal Cancer Registry (U01/U24 CA074783).

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.