Skip to main content

4) Breast cancer and BRCA1

This example is directly taken from Belman et al. (2020). It puts to the test shinyseg’s more advanced capabilities by requiring a penetrance model with age-dependent relative risks and the explicit declaration of unobserved phenotypes.

Pedigree table​

The data corresponding to the case is shown below:

 ped id fid mid sex phenotype carrier proband age
   1  1   0   0   1    nonaff       .       .  80
   1  2   0   0   2      BrCa       .       .  65
   1  3   1   2   2      BrCa     het       .  81
   1  4   1   2   1    nonaff       .       .  41
   1  5   0   0   2         .       .       .  89
   1  6   0   0   1    nonaff       .       .  80
   1  7   1   2   2    nonaff     het       .  75
   1  8   1   2   2    nonaff     neg       .  60
   1  9   4   5   2      BrCa     het       .  41
   1 10   4   5   2      BrCa     het       .  50
   1 11   4   5   2    nonaff     het       .  52
   1 12   6   7   2      BrCa     het       1  49
   1 13   6   7   1    nonaff     het       .  36
   1 14   6   7   2      BrCa     het       .  48

Notice that, together with breast cancer (BrCa), the legend includes ovarian (OvCa) and pancreatic cancer (PanCa), even though these are not present in any of the members of the pedigree. This is because they are still relevant for computing the penetrances and must be included in the analysis. The next section explains how to do so.

Penetrance​

This analysis employs the default Relative risk mode and autosomal dominant (AD) inheritance. When uploading the previous data to the app, the relative risk table will only allow parameter setting for breast cancer, as it’s the sole observed phenotype. It will also consist of a single row applying to both sexes.

However, in this case involving breast cancer and BRCA1, it is important to incorporate other cancer types to accurately calculate the penetrance. This can be achieved by entering their names into the Extra phenotypes input, located at the bottom of the Penetrance panel. Moreover, we need to declare sex-specific penetrances. This can be easily done by clicking on the cells of the sex column and changing both to any of the sexes. With that done, the table will appear as follows:

We are now ready to start filling in the table, using the same sources as in Belman et al. (2020):

  • For the baseline parameters (neg risk, neg mean, neg SD), we employed the UK 2008-2012 estimates from Cancer Incidence in Five Continents (CI5). The optimal parameters for each cancer type were identified by using shinyseg’s Assistant.

  • The hazard ratios(s) were adapted from COOL’s website. We specified the hazard ratios at ages approximately 1, 20, 40, 60, 80, and 100, which are then interpolated/smoothed by the app as shown below.

FLB​

Finally, we can Calculate the Bayes factor, which in this case points to inconclusive evidence for pathogenicity. While this conclusion aligns with that of Belman et al. (2020), we note a slight quantitative deviation (5.7 vs. 3.9) which is primarily due to how shinyseg utilizes model-based estimates instead of the raw data.

The payoff of shinyseg’s set up is that sensitivity analyses can now be easily performed by altering a reasonable number of parameters.