NEW YORK – Polygenic risk score (PRS) performance in non-European populations appears to be bolstered by bringing in information on an individual's family health and disease history, according to new research by a Broad Institute- and Harvard University-led team.
"[I]ncluding family history improves the accuracy of polygenic risk scores, particularly in diverse populations," senior and co-corresponding author Alkes Price, a biostatistics, medical and population genetics, and epidemiology researcher affiliated with the Harvard T.H. Chan School of Public Health and the Broad Institute, and his colleagues suggested.
Using simulated data, along with data for UK Biobank participants, the researchers gauged their ability to predict 10 complex conditions ranging from cancer or depression to type 2 diabetes (T2D), hypertension, or heart disease using PRSs with or without the inclusion of family history insights — work they shared in Cell Genomics on Wednesday.
Genetic risk scores used for the study were initially trained with clinical and array-based genotyping data for some 409,000 UK Biobank participants of British ancestry, the team explained, and tested on tens of thousands more individuals, including 42,000 participants with European ancestry from beyond Britain, 7,000 individuals of South Asian descent, and 7,000 individuals with African ancestry.
Along with conventional PRSs, the investigators brought in participants' family history data to come up with family history-informed PRSs known as PRS-FH scores, comparing the complex disease predictions possible with PRSs, family history alone, or PRS-FHs established using either logistic or liability threshold statistical modeling methods.
In addition, the authors noted that "[w]e have publicly released open-source software implementing both methods as well as model parameters."
The team noted that the average prediction accuracy for T2D, hypertension, and depression PRSs was 5.8 percent in the non-British Europeans and 4 percent in individuals of South Asian ancestry, but just 0.53 percent in African ancestry participants — falling short of predictions possible with models based on family history alone.
In the latter analyses, family history led to average predictions that came in at 8 percent in non-British Europeans, 8.6 percent in South Asians, and nearly 10 percent in African individuals, the researchers reported.
By combining genetic risk scores and family history into PRS-FH scores, meanwhile, the team achieved average predictions that reached 13 percent accuracy in the Europeans born outside Britain, 12 percent in South Asian individuals, and 10 percent in individuals with African ancestry.
"We determined that PRS-FH increases prediction accuracy as compared to PRS alone and across a broad set of simulations and empirical analyses, including analyses incorporating covariates," the authors reported.
The researchers noted that the PRS-FH approach was able to enhance complex disease prediction due to unexpectedly low levels of overlap between individuals' genetics and the history of disease in families, suggesting the two can provide complementary disease risk information.
Even so, they acknowledged that accurate family history of disease may be hard to come by in some cases or for certain conditions. Among other potential limitations of the study, the authors noted that extra training in given populations will be needed to refine PRS-FH models, which may misclassify disease risk related to shared environmental conditions.
"Despite [these and other] limitations," they concluded, "we anticipate that PRS-FH will attain large increases in prediction accuracy in future studies, particularly in diverse populations."