Polygenic risk scores in healthcare contexts: what's the scope? An interview study of European healthcare providers and researchers' perspectives on ethical challenges.

The First Baby Born After Polygenic Embryo Screening

During the past 10 years, large-scale genetic studies have identified hundreds of novel genetic variants for heart disease and other forms of cardiovascular (CV) disease and their risk factors.1 Although highly successful in identifying novel genomic loci, genomic research has been criticized for its high costs, slow translation to clinical care, and many unfulfilled promises. It is now clear that the promised timeline to reap the genomic benefits for medicine was too short, and the benefits themselves, to some degree, were exaggerated.2 But major progress continues to be made on several fronts in translating genomics to medicine. In this issue of Arteriosclerosis, Thrombosis, and Vascular Biology , 3 articles demonstrate 1 way by which the CV genomics community is using genomic discoveries to further our understanding of fundamental issues in the prediction and prevention of CV disease. See accompanying articles on pages 2233, 2261, and 2267 Isaacs et al,3 using lipid genetic scores, add to the mounting evidence that lifelong alterations of both total cholesterol and low-density lipoprotein cholesterol, but not high-density lipoprotein cholesterol, promote atherosclerosis and vascular plaque, leading to a higher rate of CV events. Because these natural Mendelian randomization experiments take advantage of the lifelong nature of the genetic exposure4,5 and are devoid of confounding and reverse causality, they provide important …

Transethnic Transferability of a Genome-Wide Polygenic Score for Coronary Artery Disease.

Using adult identified bone mineral density (BMD) loci, we calculated genetic risk scores (GRS) to determine if they were associated with changes in BMD during childhood. Longitudinal data from the Bone Mineral Density in Childhood Study were analyzed (N = 798, 54% female, all European ancestry). Participants had up to 6 annual dual energy X-ray scans, from which areal BMD (aBMD) Z-scores for the spine, total hip, and femoral neck were estimated, as well as total body less head bone mineral content (TBLH-BMC) Z-scores. Sixty-three single-nucleotide polymorphisms (SNPs) were genotyped, and the percentage of BMD-lowering alleles carried was calculated (overall adult GRS). Subtype GRS that include SNPs associated with fracture risk, pediatric BMD, WNT signaling, RANK-RANKL-OPG, and mesenchymal stem cell differentiation were also calculated. Linear mixed effects models were used to test associations between each GRS and bone Z-scores, and if any association differed by sex and/or chronological age. The overall adult, fracture, and WNT signaling GRS were associated with lower Z-scores (eg, spine aBMD Z-score: βadult = -0.04, p = 3.4 × 10(-7) ; βfracture = -0.02, p = 8.9 × 10(-6) ; βWNT = -0.01, p = 3.9 × 10(-4) ). The overall adult GRS was more strongly associated with lower Z-scores in females (p-interaction ≤ 0.05 for all sites). The fracture GRS was more strongly associated with lower Z-scores with increasing age (p-interaction ≤ 0.05 for all sites). The WNT GRS associations remained consistent for both sexes and all ages (p-interaction > 0.05 for all sites). The RANK-RANKL-OPG GRS was more strongly associated in females with increasing age (p-interaction < 0.05 for all sites). The mesenchymal stem cell GRS was associated with lower total hip and femoral neck Z-scores, in both boys and girls, across all ages. No associations were observed between the pediatric GRS and bone Z-scores. In conclusion, adult identified BMD loci associated with BMD and BMC in the pediatric setting, especially in females and in loci involved in fracture risk and WNT signaling.

Background: Understanding the joint effect on breast cancer risk of rare pathogenic variants in cancer predisposition genes and polygenic risk scores (PRS) from common variants can enable a precision approach to breast cancer management. Previous work found that PRS were associated with breast cancer risk in BRCA1 and BRCA2 mutation carriers recruited from cancer genetics clinics (Kuchenbaecker et al. JNCI 2017). The joint effects of pathogenic variants and PRS have not been studied in samples drawn from the general population. There is also no existing study evaluating the effect of PRS in mutation carriers in genes other than BRCA1/2. Here we evaluate the joint effects of PRS and pathogenic mutations in established cancer predisposition genes in a population-based case-control sample. Method: We analyzed 53,199 European-ancestry individuals (20,730 controls and 21,272 cases) drawn from 7 cohorts and 2 population-based case-control studies in the CAnceR RIsk Estimates Related to Susceptibility” (CARRIERS) consortium. Targeted sequencing was performed using dual bar-coded QIAseq. Mutation calling was conducted with Haplotype Caller and Vardict. We sequenced 18 breast cancer predisposition genes: ATM, BARD1, BRCA1, BRCA2, BRIP1, CDH1, CDKN2A, CHECK2, FANCC, MLH1, MSH2, MSH6, NF1, PALB2, PTEN, RAD51C, RAD51D and TP53. The PRS was calculated based on 128 common variants using effect estimates from the largest published breast cancer GWAS. The PRS was standardized to a mean of 0 and standard deviation of 1. Logistic regression was performed to assess the combined effect of identified mutation and common variants (including main and interaction effects for carrier status and PRS). Results: A total of 1,770 pathogenic mutations were observed. About 1% of the study sample had mutations in either BRCA1 (n=209) or BRCA2 gene (n=305), and 2.2% had mutations in other genes (n=1,176). The effect of PRS in BRCA1 carriers was OR= 0.97 (95%CI: 0.62, 1.52); however, this is not statistically significantly different from the PRS in non-carriers [OR=1.23 (95%CI: 1.20, 1.25)] or previous estimates in carriers [OR: 1.14, 95%CI: 1.11-1.17]. The effect of PRS in BRCA2 carriers was OR=1.87 (95%CI: 1.31, 2.78) which was statistically significantly different from that of the non-carriers. Among the mutation carriers in genes other than BRCA, the effect of PRS on breast cancer was OR=1.00 (95%CI: 0.88,1.15). Conclusion: Consistent with previous studies, we did not find evidence that the effect of the PRS among BRCA1 carriers was statistically significantly different than non-carriers. We found some evidence that the PRS effect may be larger among BRCA2 carriers than non-carriers. Our results also suggest that PRS may not have a strong effect on breast cancer risk in mutation carriers of other predisposition genes; further work is needed for verification. Citation Format: Chi Gao, Chunling Hu, Steven N. Hart, Rohan Gnanaolivu, Kun Y. Lee, Jenna Lilyquist, Nicholas J. Boddicker, Bruce Eckloff, Raed Samara, Josh Klebba, Paul Auer, Leslie Bernstein, Mia Gaudet, Hoda Anton-Culver, Amy Trentham-Dietz, Julie R. Palmer, Song Yao, Christopher Haiman, Janet E. Olson, Susan Domchek, Jeffrey Weitzel, David Goldgar, Katherine L. Nathanson, Eric C. Polley, Fergus J. Couch, Peter Kraft. The joint effects of polygenic risk scores and pathogenic variants in cancer predisposition genes on breast cancer risk in the general population: results from the CARRIERS study [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 4177.

Introduction: The potential use of polygenic risk scores (PRSs) in clinical practice is tempered by concern about their portability among diverse populations. To prevent disparities in genomic medicine, there is an urgent need to conduct genome-wide association studies in non-European ancestry cohorts. Methods: We conducted whole genome sequencing (WGS) with 30x coverage on coronary heart disease patients (n=1,067, mean age ± SD = 59.96 years ± 10.99; 70.32% males) and controls (n=6,170, mean age ± SD = 40.02 years ± 12.56; 43.45% males) in a Middle Eastern cohort to compare the performance of available PRSs for CHD (LDpred, metaGRS, lassosum, and P+T) and identify common variants associated with CHD (via generalized linear mixed models). Results: Excepting lassosum, all PRSs performed well. LDpred and metaGRS performed similarly (AUC= ~0.685) and outperformed P+T (AUC=0.667). Based on the OR per 1 SD increase (OR 1sd ), P+T (OR 1sd =1.85 [1.69-2.02], P =3.69x10 -41 ) outperformed other PRSs (OR 1sd =1.61 [1.48-1.74], P =3.02x10 -31 for LDpred and OR 1sd =1.61 [1.49-1.75], P =9.47x10 -31 for metaGRS). After binning PRSs into 10 deciles, the odds of CHD in the top decile compared to all others was highest for metaGRS (3.87 [3.07-4.88]) and LDpred (3.45 [2.74-4.341]). Thirty-two known GWAS loci (e.g., ABCG8 , CELSR2 , and SLC22A4 ) were replicated in our analysis with P &lt;0.05. Seven suggestive new loci/genes ( P &lt;10 -6 ) with relevant biological function were identified (e.g., CORO7 , RBM47 , and PDE4D ). The well-established 9p21 locus was not replicated. Conclusions: Genome-wide PRSs derived from European ancestry cohorts performed well in a Middle Eastern cohort. Further studies are needed to develop and validate an ancestry specific PRS and to confirm the suggestive loci/genes.

Polygenic scores (PGS) capture the polygenic nature of common complex diseases or traits by summarising the effects of a large number of genetic variants in a single real number. Most PGS are weighted sums of individual allele dosages of single‐nucleotide polymorphisms (SNPs), with weights corresponding to allelic log‐odds ratios estimated in SNP‐based genome‐wide association studies (GWAS). PGS are valuable in medical research because they efficiently account for the genetic background of study participants. In addition, PGS are considered to have potential as a tool for personalised disease risk assessment, prevention and treatment. However, the predictive capabilities of PGS are limited, especially in diseases with low heritability but, when combined with clinical and environmental data, PGS may improve the clinical utility of the latter, although the exact gain of this enhancement remains to be determined in appropriate studies. The clinical use of PGS also poses several ethical challenges, including issues of genetic privacy and risk communication. In addition, there is still unequal access to PGS due to the fact that non‐European population groups have so far been underrepresented in GWAS. Key Concepts Polygenic scores (PGS) aggregate the contribution of a large number of genetic variants to the risk for a complex disease or non‐clinical phenotype in a single real number. PGS are valuable tools for research to identify overlaps of the genetic aetiology of complex diseases or to assess the presence of gene–environment interactions. Although PGS bear the potential to guide personalised prevention and treatment for complex diseases, this claim most often still requires validation in real‐world settings. Sufficiently accurate prediction of disease is not feasible with PGS alone, but a combination of PGS can notably improve the predictive performance of other biomarkers or clinical information. The clinical use of PGS can have serious ethical, legal and social implications, including issues like genetic privacy and general accessibility. Since the effectiveness of PGS as a clinical and research tool varies across different ancestries, more diversity is required in genetic studies.

Researchers aim to develop polygenic risk scores as a tool to prevent and more effectively treat serious diseases, disorders and conditions such as breast cancer, type 2 diabetes mellitus and coronary heart disease. Recently, machine learning techniques, in particular deep neural networks, have been increasingly developed to create polygenic risk scores using electronic health records as well as genomic and other health data. While the use of artificial intelligence for polygenic risk scores may enable greater accuracy, performance and prediction, it also presents a range of increasingly complex ethical challenges. The ethical and social issues of many polygenic risk score applications in medicine have been widely discussed. However, in the literature and in practice, the ethical implications of their confluence with the use of artificial intelligence have not yet been sufficiently considered. Based on a comprehensive review of the existing literature, we argue that this stands in need of urgent consideration for research and subsequent translation into the clinical setting. Considering the many ethical layers involved, we will first give a brief overview of the development of artificial intelligence-driven polygenic risk scores, associated ethical and social implications, challenges in artificial intelligence ethics, and finally, explore potential complexities of polygenic risk scores driven by artificial intelligence. We point out emerging complexity regarding fairness, challenges in building trust, explaining and understanding artificial intelligence and polygenic risk scores as well as regulatory uncertainties and further challenges. We strongly advocate taking a proactive approach to embedding ethics in research and implementation processes for polygenic risk scores driven by artificial intelligence.

Purpose: Tamoxifen and raloxifene, are primary prevention strategies for women at high risk of breast cancer (Visvanathan, 2013). Recent advances in genetic studies of breast cancer risk have identified common susceptibility loci that explain 14% of familial risk for breast cancer in the general population (Michailidou, 2013). However, it is not known if these loci are risk factors for breast cancer among high-risk women treated with SERMs for breast cancer prevention. We hypothesized that the large risk reduction associated with SERMs, coupled with the fact that several breast cancer loci correlate with family history, may limit the contribution of these common genetic loci to breast cancer in this high risk population. We present the first report to evaluate 75 established breast cancer susceptibility loci, in the context of a polygenic risk score (PRS), as a risk factor for breast cancer among high risk women taking raloxifene and tamoxifen for breast cancer prevention. Methods: We conducted a matched case-control study of 594 cases (i.e., participants who developed breast cancer while on SERM therapy) and 1,171 matched controls selected from the 33,000 participants enrolled in the NSABP P-1 and P-2 breast cancer prevention trials. Genotypes of 75 single nucleotide polymorphisms (SNPs) were available from a genome-wide association study conducted at the RIKEN Center for Genomic Medicine. We formed a quantitative PRS from reported per-SNP odds ratios (OR) for the 75 susceptibility loci. Conditional logistic regression was used to examine the PRS as a risk factor for breast cancer and to assess whether the PRS and breast cancer association differed by treatment type, family history, or other clinical characteristics. Analyses also examined associations of PRS with invasive vs. in situ cancer and ER-positive vs. ER-negative cancer. Results: The PRS ranged from 3.98 to 7.74, and a one unit change in PRS was associated with a 42% increase in breast cancer (OR=1.42; 95% CI: 1.18-1.70; P = 0.0002). There was evidence of a stronger association of PRS with breast cancer among women with no first-degree family history (OR=1.62, 95% CI: 1.18-2.21) compared to those with a positive family history (OR=1.32, 95% CI: 1.06-1.66) (Pintx&amp;lt;0.05). The PRS also appeared a stronger risk factor for ER-positive (OR=1.59, 95% CI: 1.25-2.02, P &amp;lt; 0.0002) vs. ER-negative (OR=1.05, 95% CI: 0.68-1.62, P=0.84) breast cancer, although differences did not reach statistical significance (Pintx=0.10). PRS and breast cancer associations were similar across tamoxifen and raloxifene treatments, age at trial entry, 5-year predicted Gail model risk, hysterectomy status, BMI, presence of atypical hyperplasia and invasive vs. in situ cancer. Conclusion: A polygenic risk score composed of 75 loci was a risk factor for ER-positive breast cancer, especially in the absence of a first-degree family history of breast cancer. Further, the PRS associations with breast cancer were similar for women taking tamoxifen or raloxifene for prevention. These data suggest that common genetic variation adds information on risk of ER-positive breast cancer in a high-risk population receiving SERMs. Citation Format: Celine M Vachon, Daniel J Schaid, James N Ingle, Matthew P Goetz, Donald L Wickerham, Michiaki Kubo, Erin E Carlson, Soonmyung Paik, Norman Wolmark, Yusuke Nakamura, Liewei Wang, Richard M Weinshilboum, Fergus J Couch. The contribution of common genetic variation to breast cancer risk among women receiving tamoxifen or raloxifene within the National Surgical Adjuvant Breast and Bowel Project (NSABP) P-1 and P-2 trials [abstract]. In: Proceedings of the Thirty-Seventh Annual CTRC-AACR San Antonio Breast Cancer Symposium: 2014 Dec 9-13; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2015;75(9 Suppl):Abstract nr P6-10-03.

AimsRecent studies have revealed that the interplay between polygenic risk scores (PRS) and large copy number variants (CNV; >500kb) is essential for the etiology of schizophrenia (SCZ). To replicate previous findings, including those for smaller CNV (>10kb), the PRS between SCZ patients with and without CNV were compared.MethodsThe PRS were calculated for 724 patients with SCZ and 1178 healthy controls (HC), genotyped using array‐based comparative genomic hybridization and single nucleotide polymorphisms chips, and comparisons were made between cases and HC, or between subjects with and without ‘clinically significant’ CNV.ResultsFirst, we replicated the higher PRS in patients with SCZ compared to that in HC (without taking into account the CNV). For clinically significant CNV, as defined by the American College of Medical Genetics (‘pathogenic’ and ‘uncertain clinical significance, likely pathogenic’ CNV), 66 patients with SCZ carried clinically significant CNV, whereas 658 SCZ patients had no such CNV. In the comparison of PRS between cases with/without the CNV, despite no significant difference in PRS, significant enrichment of the well‐established risk CNV (22q11.2 deletion and 47,XXY/47,XXX) was observed in the lowest decile of PRS in SCZ patients with the CNV.ConclusionAlthough the present study failed to replicate the significant difference in PRS between SCZ patients with and without clinically significant CNV, SCZ patients with well‐established risk CNV tended to have a lower PRS. Therefore, we speculate that the CNV in SCZ patients with lower PRS may contain ‘genuine’ risk; PRS is a possible tool for prioritizing clinically significant CNV because the power of the CNV association analysis is limited due to their rarity.

The decline of academic medicine

Polygenic risk scores (PRS) have emerged as a potential tool for predicting complex genetic traits and disorders, which may complement traditional rare variant testing. As genome-wide association studies (GWAS) expand, PRS predictive accuracy improves, yet its role in clinical genetics remains undefined. Here, we discuss four scenarios for PRS integration into diagnostic workflows: (1) PRS as a first-tier screen to stratify patients for rare variant testing; (2) parallel testing with whole-genome sequencing (WGS) to capture both rare and common variant contributions; (3) selection between PRS and rare variant testing guided by clinical characteristics; and (4) PRS application in rare variant-negative cases to identify likely polygenic etiologies. We highlight different trade-offs of each approach, which include costs, turnaround time, diagnostic efficiency, and risk of secondary findings. While PRS shows promise in conditions with both monogenic and polygenic contributions, challenges remain in defining risk thresholds, equal accuracy across (non-European) ancestries, and integrating PRS into clinical decision-making. Although not yet standard practice, we envision PRS is likely to play an increasing role in genetic diagnostics, necessitating collaboration between clinicians and laboratory geneticists to optimize its application.

The integration of artificial intelligence (AI) with medical genetics is transforming healthcare by addressing the analytical challenges posed by the vast complexity of multi-omics data. This review explores the synergistic convergence of these fields, highlighting AI’s transformative role in enhancing diagnostic precision, enabling non-invasive molecular profiling through imaging-genetics, and advancing predictive and personalized medicine via polygenic risk scores and pharmacogenomics. AI is also emerging as a powerful generative tool in therapeutic design, accelerating drug discovery, protein engineering, and precision gene editing. However, this powerful synergy introduces significant ethical, regulatory, and biosecurity challenges, including data privacy, algorithmic bias, and the dual-use risks of AI-enabled genetic engineering. The future envisions a responsible co-evolution, with multimodal AI and the concept of the Digital Twin driving precision medicine, underpinned by interdisciplinary collaboration to ensure fairness, transparency, and societal trust. This article charts the current landscape and proposes actionable directions, emphasizing the need for robust governance to harness AI’s potential while mitigating its risks for the benefit of human health.

Weighing the evidence on costs and benefits of polygenic risk-based approaches in clinical practice: A systematic review of economic evaluations.

Testing the association between genetic scores for Attention Deficit Hyperactivity Disorder (ADHD) and health conditions, can help us better understand its complex etiology. Electronic health records linked to genetic data provide an opportunity to test whether genetic scores for ADHD correlate with ADHD and additional health outcomes in a health care context across different age groups. We generated polygenic scores (ADHD‐PGS) trained on summary statistics from the latest genome‐wide association study of ADHD (N = 55,374) and applied them to genome‐wide data from 12,383 unrelated individuals of African‐American ancestry and 66,378 unrelated individuals of European ancestry from the Vanderbilt Biobank. Overall, only Tobacco use disorder (TUD) was associated with ADHD‐PGS in the African‐American ancestry group (Odds ratio [95% confidence intervals] = 1.23[1.16–1.31], p = 9.3 × 10−09). Eighty‐six phenotypes were associated with ADHD‐PGS in the European ancestry individuals, including ADHD (OR[95%CIs] = 1.22[1.16–1.29], p = 3.6 × 10−10), and TUD (OR[95%CIs] = 1.22[1.19–1.25], p = 2.8 × 10−46). We then stratified outcomes by age (ages 0–11, 12–18, 19–25, 26–40, 41–60, and 61–100). Our results suggest that ADHD polygenic scores are associated with ADHD diagnoses early in life and with an increasing number of health conditions throughout the lifespan (even in the absence of ADHD diagnosis). This study reinforces the utility of applying trait‐specific PGSs to biobank data, and performing exploratory sensitivity analyses, to probe relationships among clinical conditions.