Title: Combining Individual-Level Dynamics of Union Formation and - - PDF document

Title: Combining Individual-Level Dynamics of Union Formation and Dissolution and Fertility with Microsimulation Modeling to Infer the Joint Distribution of Family Pedigrees and Genetic Risks for Breast Cancer Note: This is the original extended abstract. The full paper has more substantial and new results. It is available by individual request, but not for circulation without permission Authors: Michael Wolfson, University of Ottawa, mwolfson@uottawa.ca et al. Background : The vast majority of breast cancer screening in Canada and many other countries is based

• n a woman’s age, typically starting age 50. However, women with certain genotypes can be at high

risk of breast cancer at earlier ages. An ability to ascertain a woman’s genetic risk at an earlier age would enable organized screening programs cost-effectively to offer routine mammographic screening at significantly earlier ages to high risk women, and possibly starting at later ages and / or lower frequencies (e.g. triennial rather than annual or biennial) to low risk women. Multiple rare but moderate to high risk mutations for breast cancer, together with relatively more common single nucleotide polymorphisms (SNPs), each indicating lower risks but in combination quite substantial risk, have now been identified. The combined effects of multiple SNPs can be summarised into a polygenic risk score (PRS). Taken together, these measurable genetic effects can explain about 45% of familial aggregation to breast cancer, so that a woman’s detailed family history (FH) still provides substantial incremental information on her risk, even given soon to be available genetic tests. A population-based assessment of the comparative benefits of using such genetic testing + family history (FH) information for risk-based rather than primarily age-based breast cancer screening is emerging as a significant health policy question. However, no data exist that are derived from a representative population sample for the multivariate distribution of rare breast cancer genetic variants jointly with PRS and FH, nor are such data likely to become available for a considerable time. Aim: In the absence of actual data, develop a population simulation model to estimate the joint distribution of heritable risks for breast cancer, as needed for prospective cost-effectiveness evaluation

• f risk- rather than primarily age-based breast cancer screening.

Methods: As part of a major Genome Canada-funded project, a Genetic Mixing Model (GMM) is being developed to estimate this joint distribution. GMM is an interacting agent continuous time Monte Carlo microsimulation model. It simulates key demographic events especially union formation and dissolution, nulliparity and parity-specific fertility, as well as relationships that reflect blended families and hence half siblings, and mortality. It also simulates genetic inheritance for each individual in synthetic populations of millions of individuals. For its union formation and dissolution transitions, GMM draws on Canadian data and patterns, previously estimated for a microsimulation model of Human Papilloma Virus (HPV) transmission.1 2 In these models, the main focus is the transmission of HPV via heterosexual contact. As such, a very detailed stochastic characterization of short and longer term couple formation has been estimated for

• Canada. For purposes of GMM, the virus transmission and vaccination modules have been removed.

1 Miller A, Gribble, S, Nadeau, C, Asakawa, K, Flanagan, WM, Wolfson, M, Coldman, A, Evans, WK, Fitzgerald, N,

Lockwood, G, Popadiuk, C. Evaluation of the natural history of the cervix, implications of or prevention. The Cancer Risk Management Model (CRMM) - Human PapillomaVirus and Cervical components. Journal of Cancer Policy 2015.

2 Van de Velde N, Brisson M, Boilya M-C. Understanding differences in predictions of HPV vaccine effectiveness: a

comparative model-based analysis. Vaccine 2010;28:7473–84.

Instead, couples in (first, and possibly also in subsequent) stable unions have been subjected to the risks

• f births based on the mother’s age and parity-specific fertility rates.

A critical input for GMM is a set of single year age- and parity-specific fertility rates. This level of detail in the fertility rates is essential to represent, in particular, realistic distributions of age differences between mothers and daughters, numbers of female siblings, and age differences between women and their siblings. (While sisters are clearly relevant for family histories of breast cancer, brothers’ histories

• f prostate, pancreatic, and (rarely) breast cancer, which are also simulated, and sisters’ histories of
• varian and pancreatic cancer, provide further relevant information.)

Parity-specific fertility rates are not published routinely, nor are the underlying birth records linked by mother’s identity to enable them to be generated directly. Thus, parity-specific fertility rates must be estimated from a combination of fertility rates by single year of age of mother independent of parity (i.e. the data routinely available), and census data on the distribution of family compositions. Use has also been made of a U.S. survey on completed fertility to inform half sibling frequencies. The LifePaths age- and parity- specific fertility rates being used in GMM are shown in the Figure to the left.3 The children born of these unions inherit genetic characteristics. For patterns of inheritance of risk, GMM draws on the underlying genetics of the BOADICEA algorithm (Lee et al. Genet Med, 2016) based

• n Mendelian inheritance for specific genetic

mutations, and a stochastic representation for the inheritance of the polygene representation of a collection of SNPs. Given data on demographic transition probabilities, and the algorithms for the inheritance of genotypes, GMM starts with a population of men and women with randomly assigned genetics and a representative distribution by age and sex, and then simulates the evolution of this population over a period long enough to have a sample of women where all their first and second degree relatives have also been simulated, and where each has a biologically appropriate genotype. An example of one such simulated pedigree is shown to the left. Preliminary Results: Starting with populations of 250,000, and following them for 200+ years, initial simulations resulted in an average age of a mother at first birth of 28.2 in GMM, compared to 28.5 in the published Statistics Canada data for 2011 [From 91-209-X]. The average ages at second birth also align reasonably well, although the GMM results are a bit low. GMM estimates that about 30% of women

3 Note that there is considerable variability in these rates at young ages for the birth of a fourth and subsequent

child (i.e. parity = 3 or more) – basically because this is a very rare event.

have no children at the end of child-bearing years. This is high compared to data from the World Fertility Report 2012, which suggests something more in the range of 20%. As a result, further calibration work is underway. Interestingly, 45% of ‘completed’ families (female-centric) with 2 or more children have at least one pair

• f half-siblings. This proportion, while perhaps appearing high, is not out of line with (incomplete) data

from the National Longitudinal Survey of Children and Youth.4 Occurrence of half siblings has implications for the predictive power of pedigrees, since half-siblings are less genetically informative for the proband than full siblings. GMM further has calculations for predicted breast cancer risk, based on the algorithms included in BOADICEA using various kinds of genotype and family history information provided by a woman, specifically: (a) family history for first degree relatives only, (b) family history for both first and second degree relatives, (c) a polygenic risk score, and (d) the presence or absence of one of the five rare genetic variants included in the augmented version of BOADICEA currently under development. Preliminary population estimates are shown in the Table below for women age 40. The columns in the table are for various combinations of the kinds of genotype and family history data that can be used to estimate a woman’s breast cancer risk. Specifically, the first four columns represent scenarios where only one of the four kinds of information is provided for risk stratification. The next five columns show the results for all the possible pairs of types of information, while the last two columns use all three kinds of information. Columns 1, 5, 8 and 10 assume the family history information is for first degree relatives only, while columns 2, 6, 9 and 11 assume information is provided for both first and second degree relatives. Along the rows, the percentage of the population identified as being at high risk is shown for each of four relative risk (RR) thresholds, expressed as multiples of the average risk for women at age 40. The figures in the table then show the percentages of all women age 40 estimated by the BOADICEA risk assessment tool to be at a higher risk for developing breast cancer compared to women’s average risk, where the extent of higher relative risk (RR) is either 2.0, 2.5, 3.0 or 3.5 times the average for 40 year old women. Based on GMM’s simulated joint distribution of rare but highly penetrant genetic variants, PRS, and FH, at all the relative risk (RR) thresholds shown, the PRS indicated the highest percentage of “high-risk”

• women. Next in the percentages identified as being at high risk was FH when RR > 2.0. However, the

rare genes identified greater percentages for the higher RR thresholds. When two kinds of information can be used for risk prediction, PRS plus FH for second degree relatives identifies the greatest number of women at high risk.

4 http://www.justice.gc.ca/eng/rp-pr/fl-lf/famil/2004_9/f1_10.html

Column 1 2 3 4 5 6 7 8 9 10 11 Family History - 1st Degree Only Family History - 1st & 2nd Degree Rare Dominant Genetic Variants Polygenic Risk Score RR > 2.0 1.6 1.8 1.1 4.4 5.5 6.0 5.3 3.1 4.0 6.8 7.8 RR > 2.5 0.2 0.5 1.1 2.1 2.8 3.0 2.9 1.5 2.1 4.0 4.6 RR > 3.0 0.1 0.2 0.4 1.1 1.5 1.7 1.8 1.1 1.5 2.6 3.0 RR > 3.5 0.1 0.1 0.4 0.7 0.9 1.0 1.3 1.0 1.2 1.9 2.1

However, being identified as being at high risk need not be the same as actually being at high risk. A key question is the accuracy of these risk estimates, for example how extensive are false positives and false negative high risk estimates? GMM also simulates breast cancer incidence in the years following risk assessment, using the full genotype on each woman, not only that which is observed and collected during the risk assessment. These results will be critical for evaluating the most cost-effective designs of a risk-based breast cancer screening program, and will be included in the paper.5

5 We plan to incorporate a set of non-genetic risk factors in a subsequent version of GMM.