Improving precision health approaches through large-scale EHRs and biobanks

National Human Genome Research Institute

Investigators

Linked publications & trials

Abstract

During the current reporting period we have focused on these major projects: 1. PheWAS Studies The All of Us Research Program (All of Us) is a multi-site, national study with the goal of recruiting at least one million participants to further precision medicine. To evaluate how well All of Us data replicated known epidemiological associations, we performed a PheWAS using both EHR and survey-defined smoking status as a well-studied health exposure. For both exposure measures, we replicated most matched-phenotype association effects in published meta-analyses, with 73 phenotypes replicated and 59 replicated with effect sizes whose confidence intervals overlapped. This work was recently accepted for publication in JAMIA. Variation in outcomes for patients infected with COVID-19 is still poorly understood. We conducted a PheWAS to find common phenotypes associated with COVID-19 in the National COVID Cohort Collaborative (N3C) enclave, which is collected as a case and control design based on COVID-19 status and compared these results with a PheWAS in All of Us. We found highly significant findings from PheWAS analyses to be symptoms most notably associated with COVID-19 morbidity and mortality. A manuscript is in development. 2. Study of blood pressure (BP) and hypertension in large-scale biobanks Nearly half of adults in the US have hypertension. We examined the genetic determinants of BP traits in European-ancestry cohorts in a combined analysis of over 1 million individuals. This study yielded >2,100 independent genetic loci, including 113 novel associations, which explain 40% of the heritability in BP levels. A polygenic risk score (PRS) generated from these data revealed clinically meaningful differences in BP (12.9 mm Hg for systolic BP, 95% CI 11.514.2 mm Hg, p=9.08e-73) and hypertension risk (OR 5.41; 95% CI 4.12 to 7.10; P=9.71e-33) between top and bottom deciles in an independent dataset. The PRS generated from our European meta-analysis was associated with increasing BP in an African-American ancestry sample (N=21,843) from the All of Us cohort: e.g. with sex-adjusted differences between top and bottom deciles of the PRS distribution of 5.4 mm Hg for SBP (95% CI 4.2 to 6.6 mm Hg, p=1.15e-19) and increased sex-adjusted odds of hypertension (OR 1.52; 95% CI 1.3 to 1.78; p=8.11e-8). Taken together, these data demonstrate the utility of conducting both trans-ancestry and ancestry-specific analyses to understand the genetic architecture of complex human disease which may lead to robust therapeutic targets. This work is currently in revision at Nature Genetics. 3. Evaluating drug efficacy in EHR data We used EHR records in All of Us to identify differences in anti-hypertensive drug effectiveness by self-identified race and ethnicity by developing a simple generalizable method to evaluate drug-response effects in EHR data relying on OMOP-formatted data. We used a self-controlled case study design to determine medication effectiveness by calculating the reduction in SBP measurements after medication start. We analyzed 19 commonly used anti-hypertensive medications in 7,465 All of Us participants. Black and Hispanic participants were started on anti-hypertensive medications at higher SBP than White participants. Seven out of 19 anti-hypertensive drugs were less effective in Black and six were less effective in Hispanic, compared to non-Hispanic White participants. A manuscript is in development. 4. Drug effect studies in All of Us We are performing drug-effect studies in drug-induced syndrome of inappropriate antidiuretic hormone secretion (SIADH), idiosyncratic drug-induced liver injury (DILI), and clopidogrel-associated adverse cardiac events in All of Us. SIADH can be caused by a number of medications and is associated with iatrogenic morbidity (e.g., hyponatremia, neurologic complications, falls and fractures). We quantified this risk by performing an observational cohort study in All of Us with some common medications associated with SIADH. We found that duloxetine, venlafaxine, and escitalopram were associated with the highest long-term risk of hyponatremia, and bupropion was associated with the lowest risk. Antibiotics are a known cause of idiosyncratic drug-induced liver injury (DILI). This year, we published a study showing quantified frequency of acute DILI for common antibiotics in All of Us using a validated phenotyping algorithm. Our final case counts were similar to other published results, demonstrating value of All of Us. We intend to study genomic predictors in the future. Antiplatelet therapy with clopidogrel is effective at reducing subsequent events in patients with a recent myocardial infarction (MI) or percutaneous coronary procedure, but its efficacy is influenced by CYP2C19 variants. As a first test of the ability to study pharmacogenomic interactions in All of Us, we evaluated the association of CYP2C19 variants with clopidogrel response in 1,937 participants meeting study definition. As expected, CYP2C19*2 was associated with a higher risk of recurrent cardiovascular events in trans-ancestry analysis (hazard ratio (HR): 1.53, 95% CI: 1.43-1.63, P=6.6e-38). Further analysis is ongoing. 5. Other GWAS in All of Us With the release of whole genome sequencing data for nearly 250,000 All of Us participants this past April, we have launched several genome-wide association in primary hypothyroidism (PH), Graves disease, endometriosis, and uterine fibroids, among others. We have significant findings in PH, Graves, and uterine fibroids. Importantly, these analyses include diverse populations. Our findings for PH, Graves, and uterine fibroids are also being combined with other data sets in larger consortia. 6. Phenotype Risk Scores (PheRS) We have previously demonstrated PheRS as a scalable approach to determine pathogenicity of rare variants in VUMC data (Bastarache, Science 2018). Using genome sequencing data from 81,054 participants with EHR data in All of, we annotated all variants in the 78 genes with clinically actionable phenotypes in the ACMG 3.1 list. We classified these variants according to the ACMG variant interpretation guidelines and identified 6,736 individuals harboring P/LP variants (homozygotes and compound heterozygotes if applicable for autosomal recessive disorder-related genes and heterozygotes for autosomal dominant disorder-related genes) for 69 genes (ACMG positive). PheRS were significantly elevated in ACMG positive individuals for 12 diseases (P< 0.0007). We noted a higher prevalence of ACMG positive individuals for 15 genes among African ancestry individuals and 10 for admixed American ancestry individuals, including BRCA2, TSC2, and HNF1A. We also explored the potential of PheRS as a tool that could aid clinicians to accelerate the diagnosis of Cystic Fibrosis (CF) if implemented as decision support. We analyzed the health records of 965,626 individuals with ongoing care at VUMC. Among the 400 subjects with the highest PheRS for CF, 62% had already been diagnosed with CF. For 26 undiagnosed individuals who had DNA available, we conducted further clinical review and sequencing analysis of CFTR and SERPINA1. This led to the potential diagnosis of two individuals, one with CF and another with alpha-1-antitrypsin deficiency. Moreover, we discovered that six individuals had pathogenic variants (two in CFTR and four in SERPINA1), a more than three-fold enrichment for P/LP variants compared to expected. This paper was recently published in Genetics in Medicine.