Bias-corrected serum creatinine from UK Biobank electronic medical records generates an important data resource for kidney function trajectories

Abstract

ABSTRACTBACKGROUNDLoss of kidney function is a substantial personal and public health burden. Kidney function is typically assessed as estimated glomerular filtration rate (eGFR) based on serum creatinine. Emerging electronic Medical Records (eMR) in UK Biobank present a promising resource to augment the data on longitudinal eGFR based on study center visits (SC; n=15,000). However, it is unclear whether eMR-based creatinine values can be used for research on eGFR trajectories.METHODSWe derived eMR-based serum creatinine values (various assays/labs, Jaffe or enzymatic) from UK Biobank “GP-clinical”. We compared these with SC-based creatinine in individuals with both measurements available in the same calendar year (n=70,231; 2007-2012).RESULTSWe found a multiplicative bias for eMR-based creatinine that was large, factor 0.84, for 2007, and decreased over time (0.97 for 2013). Deriving eGFR based on SC- and bias-corrected eMR-creatinine (CKD-Epi 2021) yielded 454,907 individuals with ≥1eGFR assessment (2,102,174 assessments). This included 206,063 individuals with ≥2 assessments (median=6.00 assessments) for a time between 1stand last assessment of up to 60.2 years (median time=8.7 years). We enriched the dataset with eMR-recorded kidney-relevant events from “GP-clinical” (Acute Kidney Injury, End stage Kidney Disease, Nephrectomy, Dialysis, Kidney Transplant, Pregnancy, and Diabetes). We illustrated the suitability of this data: e.g. we found an annual eGFR decline of 1.04 mL/min/1.73m²/year (95%-CI=1.03-1.05), in line with literature and a four-fold steeper decline following Acute Kidney Injury.CONCLUSIONSIn summary, our bias-correction of eMR-based creatinine values enabled a 4-fold increase in the number eGFR assessments in UK Biobank suitable for kidney function research.