NEW YORK – New research published in Nature Medicine on Monday points to the possibility of developing clinical assays based on established polygenic risk scores (PRSs) and creating corresponding reporting strategies for several common conditions.
"Our work illustrates the generalizable development of a clinical PRS assay for multiple conditions and the technical, reporting, and clinical workflow challenges for implementing PRS information in the clinic," co-senior and corresponding author Jason Vassy, a researcher affiliated with Veterans Affairs Boston, Harvard Medical School, Brigham and Women's Hospital, Ariadne Labs, and his colleagues wrote, calling the PRS assay and related reporting methods "feasible (but non-trivial) next phases in PRS implementation."
In developing their workflow, the VA team sought to address two key challenges in the clinical implementation of PRS.
"On the laboratory side, how does a clinical laboratory develop a clinically valid assay like a DNA test to take an individual's DNA sample and generate these polygenic risk scores, write a report to interpret what those risk scores mean for that individual patient, and return that to the doctor," Vassy said in an interview.
"And then on the receiving end," he added, "what's the doctor — especially the primary care physician who may not be trained in genetics — do with that information?"
Vassy and his colleagues focused on PRSs for half a dozen common conditions, including atrial fibrillation, coronary artery disease, type 2 diabetes, colorectal cancer, prostate cancer in men, and breast cancer in women. Using established PRSs for each of the conditions, together with genotyping data from a randomized clinical trial called the Genomic Medicine at Veterans Affairs (GenoVA) study, they came up with array-based assays that could be used in the clinic.
The team further validated the PRSs and rationale for the related assays with genotyping, ancestry, and clinical data for another 36,423 participants in the Mass General Brigham Biobank, making additional tweaks to PRS-based clinical assays being developed at the Mass General Brigham Laboratory for Molecular Medicine that took ancestry and population structure into account.
Although these data sources offer the advantage of large numbers of individuals' genetic information, these data are typically skewed by poor ethnic diversity, with most genomic data coming from people with European ancestries.
Vassy and his colleagues tried to remove as much of this bias as possible through principal component analysis, a statistical technique for summarizing the content of large datasets through smaller "summary indices" that simplify analysis.
"There was significant skew by reported race in Mass General Brigham Biobank," Vassy said, "[so] what we had to do was additionally adjust patient samples for the first n number of principal components. And we found that brought the entire MGB population back into aligned distributions from which we felt more comfortable attributing a certain amount of risk to [patients]."
This method offers something of a trade-off between information and comparability (lower principal components effectively explain less of the overall dataset), but Vassy describes this tradeoff as conservative.
Vassy also noted that the VA team is attempting to "oversample" participants of more diverse backgrounds as their study moves forward, both in terms of ethnic ancestries, and in terms of gender representation.
When they prospectively applied the PRS assay pipeline to blood or saliva samples from 227 GenoVA participants, the researchers identified a small but significant subset of individuals who appeared to have at least double the risk of developing each of the conditions considered. When it came to prostate cancer risk, for example, more than 15 percent of individuals had genetic risk scores putting them at more than twice the population risk.
PRS cutoffs associated with a similar level of risk for colorectal cancer were found in nearly 6 percent of participants, while PRSs linked to high risk of the other conditions turned up at intermediate frequencies: 7 percent of participants were classified at high risk of coronary artery disease, 8 percent appeared to have enhanced genetic risk of T2D, 11 percent had high atrial fibrillation risk, and 13 percent of women had PRSs putting them in the high breast cancer risk group.
Reports conveying findings from the PRS assays included educational materials aimed at both the physician and the patient that distinguished between high and average risk of a given disease, which were designed to inform clinical care and decision making.
In reporting risk scores, the researchers chose to report polygenic risk separately from monogenic risk. Although ideally both would be reported together, as they provide complementary information, Vassy commented that reporting them together could prove confusing for patients, given that the science of PRSs remains an active area of research.
"I think the next wave of this kind of study will take that integrated approach," he said, "but we didn't want to risk the importance of a pathogenic finding being lost in the noise of a polygenic risk score. We wanted to really distinguish those two types of genetic risk because it may not be obvious to the average patient."
Another potential point of confusion in reporting scores, particularly to patients or even physicians without strong genetic training, is the difference between relative and absolute risk. Relative risk provides a comparison between an individual's PRS results and a population average, while that individual's absolute risk represents their personal probability of acquiring or avoiding some illness. The two measures can differ significantly.
Similar to their logic in reporting polygenic and monogenic risks separately, the VA team opted to only report relative risk, out of consideration for the current state of research into the clinical utility of PRSs.
"We didn't think at the time that we had accurate enough risk prediction models to report absolute risk," Vassy said.
Reporting the absolute risk stemming from a PRS could also mislead patients, he explained, as it could get conflated with one's total absolute risk for getting or avoiding a given condition, as this depends on many other components, including environmental and lifestyle factors.
"Research groups are actively grappling right now with trying to develop those absolute risk prediction models," he said, "and it was just beyond the scope for our study."
Vassy and his colleagues are in the process of returning results from the PRS assays to the early participants in the GenoVA trial and their doctors, in an effort to understand if and how such findings influence clinical management decisions, disease diagnoses, and related care over two years of follow-up. They are also continuing to enroll individuals from the VA Boston Healthcare System into the trial, with the goal of enrolling 1,000 individuals in all.
Results from the current analysis suggest that as PRS performance improves, the implementation processes described in the study can serve as generalizable models for laboratories and health systems looking to apply them toward patient care. The next phases of the study should shed some light on how well that works and what obstacles might lie ahead.
"We're still actively recruiting, and we're going to be following up the outcomes of these individuals," Vassy said. "What did their doctors do with the information? Did they order additional screening tests? Did patients change their health behaviors [or] change any medications?"
Perhaps most importantly, Vassy will be looking at whether any of the participants considered at risk for a disease actually ends up getting diagnosed with it.