Toward Population-Level Brain AVM Screening: Construction and Evaluation of a Two-Step Genetic Risk Model Incorporating Rare Mendelian Variants and a Novel Polygenic Risk Score
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Introduction
Brain arteriovenous malformation (bAVM) is present in 0.01% of the population and has devastating neurologic consequences when ruptured. This study reports and evaluates a novel two-step genetic risk stratification model for bAVM, taking into account both rare Mendelian and common genetic variants, to identify patients with high genetic risk who can then receive screening MRI.
Methods
Participants in the Penn Medicine Biobank (PMBB) with both exome sequencing and imputed (TOPMED) genotyping array data were included in this study. Participants were stratified into a “Mendelian cohort” and a “non-Mendelian cohort.” Participants with a pathogenic or high probability loss-of-function variant in any known Mendelian AVM-related gene formed the Mendelian cohort, while the remainder formed the non-Mendelian cohort. To define bAVM cases, chart review was performed for all participants in the Mendelian group and for all other participants with at least one occurrence of the International Classification of Diseases (ICD)-10 code Q28.2 (“Arteriovenous malformation of cerebral vessels”) in their EHR. Within this reviewed group, those with MRI-, CTA-, or angiogram-confirmed brain AVMs were considered bAVM cases, those with negative imaging and all those participants lacking the ICD-10 Q28.2 code were considered controls. Those with ICD code Q28.2 but with inadequate imaging were excluded from analysis. A polygenic risk score (BAVM-PRS) was created using genetic data from a prior, external genome-wide association study of 1706 European-ancestry participants (515 bAVM cases, 1191 controls). Performance of the BAVM-PRS among non-Mendelian PMBB participants was assessed using logistic regression, density plots, receiver operating characteristics (ROC), and threshold-based analysis. Using this unbiased genotype-first approach to identify participants at high risk for bAVM, the prevalence of bAVM among both Mendelian risk and high-risk BAVM-PRS PMBB participants was calculated.
Results
A total of 56,332 participants from PMBB were included and stratified into a Mendelian cohort (n=86; median [interquartile range, IQR] age 55.6 [40.7, 64.5] years, 47.7% male, 67.4% European-ancestry, 30.2% African-ancestry) and a non-Mendelian cohort (n=56,246; median [IQR] age 56.4 [41.5, 66.8] years, 48.7% male, 73.9% European-ancestry, 21.4% African-ancestry). There were 4 (4.7%) confirmed bAVM cases in the Mendelian cohort and 55 (0.1%) in the non-Mendelian cohort. In the non-Mendelian cohort, BAVM-PRS Z-score was significantly higher among cases than controls (median [IQR] 0.303 [-0.218, 1.291] vs.-0.099 [-0.677, 0.566], p=0.002). Overall discrimination was moderate as assessed by ROC (Area-under-curve was 0.694, 95% CI 0.633-0.756). Density plots stratified by ancestry showed separation between cases and controls at high BAVM-PRS values for both European-and African-ancestry participants. Using a two-step risk stratification approach, flagging all Mendelian variant-carrying participants regardless of BAVM-PRS and all non-Mendelian participants with a BAVM-PRS Z-score >2.7 as high risk for bAVM, we identified a high risk population of 681 participants where prevalence of brain AVM was 1.0%. Within this high risk group, the number needed to screen (NNS) with brain MRI to diagnose one bAVM was 122, and our two-step risk stratification model demonstrated an overall capture rate of 12% with a sensitivity of 10% for bAVM in the PMBB population.
Conclusion
A two-step genetic risk model substantially improves bAVM screening efficiency, with a NNS of 122 to identify a single bAVM. Compared to a NNS of ∼1,250 when screening the general PMBB population, these sequential genetic filters represent a 10-fold increase in diagnostic efficiency. Similar approaches in other sequenced populations may allow for the efficient and early identification of bAVM in asymptomatic, high-risk individuals before they suffer the devastating sequelae of bAVM rupture.
