Bringing order to polygenic score reporting
14 April 2021
Despite the rapid development of polygenic score models – hundreds of which have been developed for diseases ranging from diabetes to cardiovascular disease to cancer - and broad agreement that there is some ability to predict disease, inconsistencies in the way that models are reported could delay their introduction into healthcare.
Two recently released papers, one presenting a standards framework, the other a catalog of published polygenic scores (PGS) aim to address the problem with guidance on how to describe the development of polygenic score models and a repository which researchers can search for, and compare, the various models available. Although their use is voluntary, these tools should enable identification of the polygenic score models most suited for a particular clinical need and support translation from research into clinical practice.
What are polygenic score models – a reminder
Polygenic scores are the merging of the effects of multiple common genetic variants into a single number, usually to indicate the genetic risk a person is at of developing a named disease.
Considerable time and energy are being invested in developing polygenic score models and evidence is mounting that their predictive ability could be clinically useful, although the information they provide will have different implications for different conditions, depending on the underlying genetic architecture of the disease.
A decision on clinical utility (effectively, clinical ‘usefulness’) requires a clear specification of the purpose for using a score and full appreciation of the information that the specific polygenic score being used can provide, as well as the degree to which this information makes an actual improvement to clinical practice, such as a reduction in deaths from the disease.
Despite the rapid development of polygenic score models, the way they are reported in published research is inconsistent. Variations in details such as how the models are built and assessed make it very difficult to define their use beyond research and identify those that are best suited for use in clinical practice. The two papers could bring much needed order to the field of polygenic score research.
A library for polygenic score models
The PGS Catalog is a central, open access online resource of curated polygenic scores. Researchers can upload standardised summaries of key aspects of their polygenic score model, such as the genetic variants included in the model, as well as the research study design and its findings. Others can then use this information to help evaluate existing polygenic scores, make comparisons and assess their predictive ability.
While curating PGS Catalog, the developers uncovered issues around poor reporting of polygenic scores, highlighting the need for standards. Approximately 40% of the 231 polygenic score publications reviewed did not include adequate information to enable users to repeat the calculation in a new sample set, meaning researchers cannot determine if the polygenic score model would work in the same way when looked at in new samples.
So in a collaboration with the Clinical Genome Resource (ClinGen) Complex Disease Working Group and the PGS Catalog, Polygenic Risk Score Reporting Standards (PRS-RS) were developed.
Improving the quality of reporting
Reporting standards are essential for meaningful evaluations by ensuring different developers share the same information about their models. The PRS-RS are an update to previous guidance and consist of twenty two items of model development information that, at a minimum, researchers are recommended to include when publishing their work. These fall under the broad categories:
- background on the study type and risk model
- study population
- risk model development and application
- risk model evaluation
- limitations and clinical implications and
- data transparency and availability.
The standards also outline how to describe the datasets and statistical methods used to develop the polygenic score model as well as validation of the models and how to note any limitations.
These PRS-RS are necessary to enable comparison between different polygenic models, evaluate models for their use within specific settings and ensure reproducibility of results. They will provide transparency as to how scores are developed and simplify the identification of those that may be developed for clinical care.
The road ahead
As highlighted in our report, Polygenic score and clinical utility, demonstrating how accurately a model predicts what it says it will i.e. its analytic validity, is only the first step towards demonstrating its ability to improve health outcomes i.e. its clinical utility.
However adherence to the PRS-RS will improve consistency in the way model developers and researchers report their findings, allowing implementers to make comprehensive comparisons between models, in the PGS Catalog.
Both the PRS-RS and the PGS Catalog are positive developments that will contribute to the translation of polygenic scores from research into healthcare by enabling those wanting to implement polygenic scores to select the most promising ones to take to the next stage – such as trialling them in the clinic.
The use of these standards and submission to the PGS Catalog should be highly encouraged and supported. They will support broader efforts to bring the appropriate polygenic scores into clinical care which can provide benefit to patients, healthcare professional in their decision making and healthcare systems.