Skip to main content
Premium Trial:

Request an Annual Quote

Myriad Genetics Generates More Data on Ability of PRS to Identify Who Will Develop Breast Cancer


NEW YORK – After comparing the combined power of its polygenic risk score (PRS) and a clinical algorithm to predict which patients will get breast cancer against real-world incidence, Myriad Genetics has greater confidence that its clinico-genomic algorithm could potentially improve breast cancer screening and prevention strategies over standard methods.

In a prospective, longitudinal validation study presented at the San Antonio Breast Cancer Symposium last week, researchers from Myriad and multiple institutions followed for roughly one year approximately 130,000 women between the ages of 18 and 84, who received PRS results and who were referred for clinical genetic breast cancer testing. Researchers used hospital and medical claims data to identify patients who developed breast cancer and then compared observed cancer rates against expected cancer rates.

Researchers calculated expected breast cancer rates using a combined risk score (CRS) consisting of Myriad's RiskScore, a 149-SNP PRS test, and the Tyrer-Cuzick model, an established risk tool that assesses a woman's 10-year and lifetime risk of developing breast cancer using clinical features. The claims data also allowed researchers the opportunity to gauge whether the CRS improved risk estimates over just the Tyrer-Cuzick model.

Myriad markets the PRS currently for estimating the five-year and lifetime risk of breast cancer for women who haven't had it before and who are negative for high-risk mutations in breast cancer-associated genes detected by its next-generation sequencing MyRisk Hereditary Cancer test. The Salt Lake City-based company aims to set its PRS apart from others that have been developed largely using data from patients of European ancestry and last year, presented data showing that its score can provide accurate breast cancer risk estimates for women of different ancestral backgrounds.

"We've had a lot of women get our test at this point," said Thomas Slavin, Myriad's chief medical officer, explaining that with the latest prospective validation study the company wanted to strengthen the evidence base for the PRS and show that its test does, in fact, predict who is most likely to get cancer.

"[The study] is a validation that our test is doing what it should be doing," Slavin said.

Over a median follow-up time of 21.1 months, Myriad's researchers observed 340 breast cancer events, which the CRS model appeared well calibrated to predict. A ratio of observed-to-expected cancer incidence equaling one would denote that the risk prediction model perfectly predicted who actually got breast cancer. But when the ratio exceeds one, that indicates the risk prediction model said fewer people would get cancer than actually did and when the ratio is less than one, then the risk prediction model said more people would get cancer than actually did.

In the overall cohort, the ratio of observed incidence to the CRS-based expected incidence was 1.11. In the highest-risk decile, the ratio of observed to CRS-based incidence was 0.91 but only 0.67 when the observed incidence was compared to predictions by Tyrer-Cuzick alone.

"That just gets at the ability of [the model] to predict the observed breast cancers from what was expected," Slavin said.

Researchers also compared breast cancer risk predictions according to CRS against just Tyrer-Cuzick. Among the 16,448 women designated as high risk by Tyrer-Cuzick, Myriad's CRS model reclassified 32.6 percent of them as low risk, while 7.9 percent of the 113,610 women labeled as low risk by Tyrer-Cuzick were reclassified as high risk by the CRS model. In all, the CRS model found 15,986 women — 12.3 percent of the total population — above the high risk threshold, 123 of whom experienced breast cancer events.

Slavin explained that this meant that the estimated risk for women who were reclassified using the CRS better matched the observed incidence of breast cancer in the cohort than using the Tyrer-Cuzick model alone.

Jonathan Mosley, an associate professor of clinical pharmacology at Vanderbilt University Medical Center, whose research often focuses on PRS, commented that while the use of a prospective cohort is a strength of this study, the low overall event rate seen in the study — 340 breast cancer cases out of approximately 130,000 women, or roughly 0.2 percent of the total — is far lower than the 13 percent average lifetime risk for breast cancer in the general population.

This could make it hard to gauge the actual performance of the CRS, Mosley said, as the study data suggest that risk may be underestimated in the lowest risk group, where the observed-to-expected incidence ratio was highest at 1.83. "Underestimating risk in the lowest risk thresholds could lead to inappropriate down-classification of risk," he said.

Although the cancer events observed in this study, along with their distribution across women of different backgrounds, were too small to confidently analyze the multi-ancestry aspect of the company's PRS, Slavin said that the overall results fall in line with those of a larger study that Myriad published with academic collaborators in November in JCO Precision Oncology.

In that study, which followed close to 300,000 women, the multi-ancestry PRS significantly associated with breast cancer among both European and non-European women.

In the current study, he said, "we still had 20,000 people of non-European ancestry, so if [the data] was off, then we would have seen it in the [results]."

Amid calls to improve equity in healthcare, companies like Myriad and others have committed to developing multi-ancestry PRS tools. Earlier this year, MyOme presented proof-of-concept data on its own multi-ancestry breast cancer PRS, which it hopes to commercialize next year. Invitae and Allelica also recently partnered to do the same.

Slavin calls such multi-ancestry PRS "game changers," and noted that these algorithms have utility beyond breast cancer. "You can use that model now for diabetes, colon cancer, cardiovascular disease," he said. "To weight things by ancestry, using the underlying genetic information, is really a very innovative way to do polygenic risk scores."