Family History and Cancer Risk and Mortality

Subproject

Pooling Project of Prospective Studies of Diet and Cancer (DCCP)

Robert Macinnis-macinnis@unimelb.edu.au
Mary Beth Terry-mt146@columbia.edu

We plan to apply for NCI, ACS, NHMRC and other funding upon approval, but already have staff in place who can start work on this project.

Family history is a well-established risk factor for many cancers. For example, prior research has shown that women who have one affected first-degree relative are on average at a two-fold increased breast cancer risk relative to those with no first-degree family history, and this association increases up to four-fold for women with three or more affected first-degree relatives. An increased familial risk has been consistently found for prostate, ovarian, lung, colorectal, pancreatic and other cancers, and the increased risk can be up to 57-fold for those who have three or more affected family members(1-10).

However, the familial risk decreases as the age at diagnosis of affected relatives, as well as the age of the woman herself, increases. So categorizing family history as a binary (ever/never) construct captures only part of the great heterogeneity in both absolute and relative site-specific cancer risk, which depend on the age of the individual at risk, the age(s) at diagnosis of the affected relative(s) and the genetic relationship between them. Thus, the average risk difference and relative risk between those with a family history and those without likely underestimates the risk gradient possible from a better characterization of family history that takes into account the above factors, which can have large implications for accurate cancer risk assessment.

Furthermore, the implications of familial risk have been under-appreciated. Even the modest two-fold increase in risk typically associated with having an affected first-degree relative could not exist without there being a very strong underlying familial, including genetic, risk gradient which we refer to as familial risk profile(11,12). Yet, epidemiologic studies rarely account for familial risk profile through study design or analysis techniques (e.g., matching), leading to vast differences in the distribution of familial risk profile for cases compared to controls, especially in the upper tail.

While some of the heterogeneity in familial risk is explained by genetic risk factors that have been identified by linkage, candidate gene and genome-wide association studies (e.g., BRCA1, PALB2, FGFR2 for breast cancer)(13-18), the vast majority of those affected with a cancer family history, even if strong, do not have high- or moderate-risk mutations in the cancer susceptibility genes identified to date. Thus, there are substantial proportions of individuals at increased, if not high, risk of cancer due to familial factors; however, few studies have examined cancer risk factors for individuals at increased familial risk, let alone sought to see if risk associations differ across the spectrum of risk. Most prospective cohorts have been unselected for familial risk (some exceptions include the Breast and Colon Cancer Family Registries and the Sisters Study(19,20)). Given the highly skewed distribution of familial risk profile, the vast majority of individuals in these cohorts are below average risk and therefore pooled analyses across cohorts are necessary to ensure sufficient number of individuals at the higher end of the familial risk spectrum.

Thus, we propose to create the largest data set to date by using the cohort consortium to prospectively evaluate the impact of family history (by extent and type) on cancer risk and mortality. We also want to examine the association between risk factors and cancer risk stratified by predicted familial risk profile (across the spectrum of risk). As well as considering family history in terms of the conventional categories (e.g., number of affected first degree relatives), we will assess family history in a novel way that uses the variation in ages at diagnosis and relationships between relatives to create a continuously-distributed predicted familial risk profile. This will give us more power to study if the causes of family history are modifiers of established risk factors. We are developing these tools as part of existing NIH-funded studies in the Breast and Colon Cancer Family Registries by modifying the computer program underpinning the risk prediction model BOADICEA(21,22), and will adapt these to the data collected by cohorts within the cohort consortium.

Overall, our goal is to improve risk assessment and stratification and identify factors that could lead to reduced risk and increased screening efficacy of these site specific cancers across the spectrum of familial risk and, in particular, for people with strong or stronger family histories of these diseases.

Aim 1: To examine the associations between family history, both as conventional categories and as measured by predicted familial risk profile, and risk for cancers of the: a) breast, b) prostate, c) colon and rectum, d) ovary, e) lung and f) pancreas.

Aim 2: To assess whether the associations with established and suspected risk factors differ according to family history, and if they are modified by predicted familial risk profile, for cancers of the: a) breast, b) prostate, c) colon and rectum, d) ovary, e) lung and f) pancreas.

Aim 3: To evaluate the associations between family history of cancer, both in terms of conventional categories and as measured by predicted familial risk profile, and overall and cancer-specific mortality.

We will calculate age-specific incidences of site-specific cancers, and overall and cause-specific mortalities. We will examine the association between family history (e.g., ever/never status, predicted familial risk profile(21,22)), and cancer risk and mortality using multivariable Cox proportional hazards modeling, stratified on study. We will assess cancer-specific risk factors for individuals with strong or stronger family histories.

Few prospective studies have assessed family history and cancer risk with decent statistical power due to the 'exposure' not being common, especially for strong family histories. Most prior studies of such people have relied on highly selected clinic-based samples. Knowledge about risk factors for those at increased familial risk is important, and to date under-studied. Thus, we propose to examine familial aspects risk using the largest prospective dataset to date.

At least one of the cancers of interest

Age at diagnosis of the following cancer types: a) breast, b) prostate, c) colon and rectum, d) ovary, e) lung and/or f) pancreas.
Age at death (if deceased)
Age last known to be alive

Family history of at least one of the six selected cancers (breast, prostate, colon and rectum, ovary, lung, and pancreas)

Age, Sex, Ethnicity, Smoking history, Alcohol intake, Education, Body Mass Index, Waist Circumference, Waist to Hip Ratio, Physical Activity, Dietary variables (e.g. energy intake, meat, fiber, fruit/Vegetables, glycemic load/index, fat, etc), NSAID use, Hormone therapies (OC, MHT), Parity, Menopausal status, Age at menarche, Diabetes history, Surgical history (oophorectomy, mastectomy), Medical radiation exposure

No