Medical records-based chronic kidney disease phenotype for clinical care and "big data" observational and genetic studies

Abstract

Chronic Kidney Disease (CKD) represents a slowly progressive disorder that is typically silent until late stages, but early intervention can significantly delay its progression. We designed a portable and scalable electronic CKD phenotype to facilitate early disease recognition and empower large-scale observational and genetic studies of kidney traits. The algorithm uses a combination of rule-based and machine-learning methods to automatically place patients on the staging grid of albuminuria by glomerular filtration rate ("A-by-G" grid). We manually validated the algorithm by 451 chart reviews across three medical systems, demonstrating overall positive predictive value of 95% for CKD cases and 97% for healthy controls. Independent case-control validation using 2350 patient records demonstrated diagnostic specificity of 97% and sensitivity of 87%. Application of the phenotype to 1.3 million patients demonstrated that over 80% of CKD cases are undetected using ICD codes alone. We also demonstrated several large-scale applications of the phenotype, including identifying stage-specific kidney disease comorbidities, in silico estimation of kidney trait heritability in thousands of pedigrees reconstructed from medical records, and biobank-based multicenter genome-wide and phenome-wide association studies.

Fig. 1. Electronic CKD diagnosis and staging algorithm.

a Flowchart of the National Kidney Foundation (NKF) criteria-based algorithm composed of three parts: data pre-filtering, G-staging, and A-staging b G-Stage Classifier for staging of CKD based on estimated glomerular filtration rate (eGFR), and c A-Stage Classifiers for staging of CKD based on albuminuria. UACR Urine Albumin-to-Creatinine Ratio, UPCR Urine Protein-to-Creatinine Ratio, A24 24-h urine collection for albumin, P24 24-h urine collection for protein, UA Urinalysis, SG Specific Gravity.

Fig. 2. Comorbidity heatmaps for 239,332 CUIMC patients algorithmically placed on the A-by-G grid.

The prevalence of a comorbidity within each cell is provided, with the shaded color scale varying from red (highest prevalence) to green (lowest prevalence). The arrows correspond to the direction of effect and P values the statistical tests of comorbidity gradients across the grid. The analysis excludes individuals with missing urine tests and those with ESRD on dialysis or after transplant. Models based on logistic regression was used for binary traits and Poisson regression for ICD counts. All models were adjusted for age and sex and P value <6.25 × 10⁻⁴ is considered as significant after Bonferroni correction. NS not significant.

Fig. 3. EHR-based observational heritability (h_o²) of renal function and albuminuria.

a h_o² of eGFR (quantitative trait) in families with any CKD, moderate CKD (G-stage 3 or greater) and advanced CKD (G-stage 4 or greater) b h_o² of albuminuria (A2 and A3, dichotomous) and severe albuminuria (A3, dichotomous). Bars correspond to 95% confidence intervals around the point estimates.

Fig. 4. Combined GWAS-PheWAS approach for moderate CKD (G3 or greater).

Manhattan plots for a eMERGE Europeans (7536 cases, 17,841 controls) with a genome wide-significant signal at the UMOD locus (red); b eMERGE African-Americans (702 cases, 2029 controls) with a genome wide-significant signal at the APOL1 locus (red); regional plots for the c UMOD and d APOL1 loci; eMERGE-based PheWAS plot for the top SNPs at the e UMOD (n = 78,638 Europeans) and f APOL1 (n = 16,976 African Americans) loci; upward triangles refer to increased risk; downward triangles indicate reduced risk; horizontal doted lines refer to Bonferroni-corrected significance thresholds.