Project Abstract
Due to the rapid drop in the cost of DNA sequencing, a panel of multiple (25 to 125) cancer susceptibility
genes can be now tested at a fraction of cost and time of what was needed earlier for testing only a handful
of well-known cancer genes such as BRCA1 and BRCA2. As multi-gene panel tests are becoming accessible to
increasingly broader population, more and more patients carrying pathogenic germline mutations of various
genes are being identi¿ed. This trend is creating a major paradigm shift in hereditary cancer risk assessment
because the newly identi¿ed mutation carriers need advise about approporiate management strategies such as
targeted surveillance and preventive options. The counseling hinges crucially on accurate quantitative estimates
of age-speci¿c risks of developing cancers associated with a speci¿c gene whose pathogenic mutation a patient
carries, i.e., penetrance estimates. For several gene-cancer associations, a substantial amount of literature on risk
estimation is available and new studies are also becoming available at a fast pace. Yet a synthesis of evidence
from all relevant studies in the form of a robust meta-analysis of penetrance is generally lacking. A case in point
is PALB2-breast cancer association for which several studies report risk measures, however, there is no meta-
analysis of penetrance estimates. Thus there is an urgent need for meta-analysis of penetrances so that mutation
carriers in PALB2 and many other genes can be counseled appropriately in light of their age-speci¿c risks of
developing various cancers. A major challenge in this task is that studies typically vary in design (e.g., family-
or population-based) and hence type of risk measures reported (e.g., penetrance, relative risk, or odds ratio).
Synthesis of such heterogeneous risk measures is not possible using the standard meta-analysis approaches.
Moreover, for an accurate and robust estimation, the meta-analysis model should properly take into account
various sources of uncertainties that arise in such kind of synthesis. To ¿ll this gap, we propose a Bayesian
hiererchical model for meta-analysis as it allows seamless integration of results of different types by borrowing
information across the studies. At the same time, it accounts for uncertainties in a formal manner through
hierarchical priors. Our ¿rst speci¿c aim is to develop this Bayesian meta-analysis methodology, investigate its
properties, and compare it with existing approaches through simulations. The next speci¿c aim is to apply this
method to estimate the penetrance for PALB2 gene for breast cancer. Our ¿nal aim is to develop an R package
that implements the proposed Bayesian methodology and integrate it into the clinical decision support tool
ASK2ME for immediate clinical use.
