Abstract: FR-PO104
Novel Data-Driven Phenotyping to Support Genome-Wide Association Study (GWAS) Exploration in AKI
Session Information
- AKI: Epidemiology, Risk Factors, Prevention - I
November 03, 2023 | Location: Exhibit Hall, Pennsylvania Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Acute Kidney Injury
- 101 AKI: Epidemiology, Risk Factors, and Prevention
Authors
- Matheny, Michael Edwin, VA Tennessee Valley Healthcare System, Nashville, Tennessee, United States
- Hellwege, Jacklyn N., VA Tennessee Valley Healthcare System, Nashville, Tennessee, United States
- Birkelo, Bethany, VA Tennessee Valley Healthcare System, Nashville, Tennessee, United States
- Parr, Sharidan, VA Nebraska-Western Iowa Health Care System, Omaha, Nebraska, United States
- Vincz, Andrew J., VA Tennessee Valley Healthcare System, Nashville, Tennessee, United States
- Richardson, Trey Howard, Vanderbilt University Medical Center, Nashville, Tennessee, United States
- Hung, Adriana, VA Tennessee Valley Healthcare System, Nashville, Tennessee, United States
- Velez edwards, Digna R., VA Tennessee Valley Healthcare System, Nashville, Tennessee, United States
- Siew, Edward D., VA Tennessee Valley Healthcare System, Nashville, Tennessee, United States
Group or Team Name
- Million Veterans Program Initiative.
Background
Traditional phenotyping may have limited sensitivity for detecting meaningful phenotypes. Machine learning may detect patterns of variables for novel phenotype identification. We applied this approach to identify phenotypes of acute kidney injury (AKI).
Methods
A cohort of VA patients hospitalized from 2002-2019 was aggregated. AKI was defined as KDIGO Stage 1 or greater during hospitalization. 5,515 data features including individual demographics, laboratory tests, medications, and billing codes were used to calculate longitudinal curves anchored on each patient hospitalization, and endophenotypes inferred with Independent Component Analysis (ICA). Ten iterations from 200,000 randomly selected hospitalizations with AKI were analyzed, resulting in 11,985,029 learning instances. GWAS was performed on 167,051 MVP patient hospitalizations with and without AKI were scored for each phenotype. Each patient was represented only once with that patient’s hospitalization being randomly selected. Single variant linear regression analyses were performed using imputed data, adjusting for sex, age, the top 10 principal components, stratified by HARE race/ethnicity, followed by meta-analysis. Phenotypes were retained if they had at least 1 significant SNPs and inflation between 0.98-1.1.
Results
2,137 distinct phenotypes were identified. 50 phenotypes with data patterns potentially relevant to AKI were selected by expert adjudication. Nine of 50 phenotypes were retained. We selected two phenotypes, one clinically consistent with patterns of acute glomerulonephritis (AG) and one with cardiorenal syndrome (CS). The former was associated with serum complement and albumin levels, hyperlipidemia, and acute/chronic hepatitis C. The latter was characterized by acute on chronic heart failure, elevated BNP, and acute renal failure (ARF) codes. The presence of ARF in both phenotypes enriched for intrinsic/more severe injury. Two loci with AG (HLA and COL4A2), and one locus (FANCL) with CS reached genome-wide significance.
Conclusion
An entirely data-driven process can identify both well-established and potentially novel endophenotypes of AKI that can be explored for clinically provocative features. These findings show promise for novel phenotype discovery and significant genetic association detection.
Funding
- NIDDK Support