Polygenic scores - from risk models to clinical assays

A consistent message in our reports has been that to deliver PGS-based tests or assays into clinical care effectively and efficiently, policy-makers and health professionals need clarity about these technologies and the purpose of their use.

Sowmiya Moorthie

12 May 2022


Integration of polygenic scores (PGS) into clinical practice requires robust, validated mechanisms to generate these scores. A consistent message in our reports has been that to deliver PGS-based tests or assays into clinical care effectively and efficiently, policy-makers and health professionals need clarity about these technologies and the purpose of their use.

A recent paper from the GenoVA study provides a nice illustration of the practical steps and considerations essential to moving from models to clinical assays - a subject on which there has been very little discussion prior to this paper.

The paper describes the development of PGS-assay for use in clinical care for six conditions - coronary artery disease, atrial fibrillation, type 2 diabetes mellitus, colorectal, breast or prostate cancer. This assay is being used as part of the GenoVA study, which aims to determine the clinical effectiveness of using polygenic score testing in patients aged 50-70 years old, at high genetic risk for one of those six diseases.

The paper describes the process to develop the assay and to determine its analytical validity. As the authors illustrate, this is not a simple task and requires considerable effort.

The elements of a clinical assay for polygenic scores

Polygenic scores are a calculated measurement, usually the sum of the SNPs associated with a disease, weighted to reflect their relative influence. Polygenic score models enable this calculation, making them an integral part of any testing strategy that relies on PGS information.

There are several methods that can be used to obtain a polygenic score, based on how the genetic data is generated (e.g. microarray or sequencing), the polygenic score model that is applied to it and how this information is then interpreted.

Several models have been developed for different diseases, but the availability in research of such models is not the same as having an assay or test that can be used in clinical medicine. Model development is just one of a series of steps in creating a test suitable for use when making clinical decisions about patient care. And, while several polygenic score models have shown promise in research settings, these will need to be further validated and developed for use as clinical prediction algorithms.

One important short-coming to be addressed is that, to date, a large proportion of models have been developed using data only or primarily from populations of European ancestry. This makes them less predictive in those of non-European ancestry. As discussed in the paper, this has implications for implementation and raises questions as to whether the models should be used. In this study, additional statistical methods that correct for population structure, were used to enable application of existing models to those from other ancestries. Whilst not a perfect solution, it is a pragmatic solution in the short-term and enables studying the potential of this technology for a broader group of people. 

The research team were also keen to help clinicians and patients talk about using PRS when making medical decisions about screening and prevention. As corresponding author Jason Vassy, MD, MPH stated ‘As a primary care physician myself, I knew that busy physicians were not going to have time to take an entire course on polygenic risk scores.’ The team have designed ‘a lab report and informational resources that succinctly told the doctor and patient what they need to know to make a decision about using a polygenic risk score result in their health care’.

As illustrated by this paper and a key message in many of our reports, implementation of PGS analysis is not a simple task. The assay is much more than a risk model, and comprises of different elements that need to function together to provide consistent and robust results.

Applying evaluation frameworks to polygenic score analysis

There is a lot of enthusiasm for using polygenic scores as a biomarker within healthcare pathways, but, as with other biomarkers, they don’t provide a complete solution, nor do they provide definitive answers. We still need to develop the evidence-base for its use within specified care pathways. There also needs to be clarity about what the information from such analysis means and the value it adds to specific healthcare pathways.

Without a clear understanding of the proposed, defined use of a specified PRS test or assay its effective and responsible evaluation is not feasible.

The PHG Foundation continues to work on these challenges in collaboration with stakeholders. We are continuing to analyse the key concepts around PRS test evaluation frameworks and more effectively assess the potential of polygenic score analysis for healthcare. Following on from examining the concept of clinical utility and how it may apply to polygenic score analysis, we are now investigating the issues in demonstrating clinical validity of such assays and tests.

Genomics and policy newsletter

Sign up