Assessing Uncertainty in Hereditary Risk Models for BRCA1/2 Related Cancer (Vortrag)
|Assessing Uncertainty in Hereditary Risk Models for BRCA1/2 Related Cancer (Vortrag)|
|Autor||Ekaterina Auer, Wolfram Luther|
|Ort||Szeged, Hungary (online)|
|vom||13. September 2021 bis 15. September 2021|
Public websites making general recommendations about preventive services for major diseases are becoming increasingly important in the healthcare area. Government agencies, leading universities, and independent foundations provide them nationally or internationally for the purpose of informing the population about the available possibilities. Additionally, such websites allow individuals and their families to assess their risk of contracting a particular disease based on various factors such as their age, origin, or genetic predisposition. With the help of this risk assessment, concrete recommendations can be made for individual prevention and risk mitigation.
In this contribution, we focus on BRCA1/2 related cancer. Complex prediction software relying on various kinds of mathematical stochastic models plays an essential role in the process of genetic counseling with the goal of determining either the gene mutation probability of a patient or their lifetime risk of cancer. Using personal information about individuals and their relatives as well as standardized case data from medical databases, a recommendation can be calculated and communicated via a suitable output interface (e.g., as a graph or a report). Often, such recommendations can be augmented based on opinions of a multidisciplinary team of experts who collaborate in a final meta-study on issues concerning, for example, benefits/harms of counseling or clinical treatment for specific disease patterns. Among the questions most meta-studies are raising is the issue of reliability of the generated recommendations, since they are considerably influenced by uncertainty. It is noted that existing verification and validation approaches usually account only for aleatory uncertainty and tend to disregard other kinds.
We consider genetic risk assessment and genetic counseling for BRCA1/2 related breast cancer from the point of view of reliable uncertainty handling. First, we provide a short overview of existing risk models, software tools as well as family history interfaces and repositories. We show how missing or conflicting information on mutation probabilities can be improved using Dempster-Shafer theory. Based on multi-criteria binary decision trees and interval analysis, we combine the referral screening tool RST designed to determine patients at risk of breast cancer with three further widely spread risk assessment tools for this purpose. The combined method has the advantage of assigning individuals to appropriate risk classes depending on their family history, taking into account epistemic uncertainty in the information about such factors as the age of onset in a relative, the degree of kinship or the relative's origin.