Evaluation and Refinement of Clinical Equations in Nephrology

Abstract

Chronic kidney disease (CKD) is a common and heterogeneous condition in which clinical decisions depend on assessing both current kidney function and future risk of kidney failure using estimated glomerular filtration rate (eGFR) and kidney failure risk equations. This dissertation examines three limitations of these assessments: competing equations produce different eGFR estimates and downstream classifications; estimation error propagates to uncertainty in threshold-based decisions; and current risk equations underutilize routinely collected repeat laboratory measurements. I address these limitations by evaluating and refining equations according to their clinical consequences rather than their statistical performance alone, through population-aware evaluation, uncertainty-aware interpretation, and trajectory-aware risk assessment. First, using a large multiethnic population, I evaluated current and previously recommended creatinine-based eGFR equations across CKD staging, kidney failure prediction, and CKD prevalence. Equation choice substantially shifted CKD stage assignment and estimated disease burden across populations, while producing much smaller effects on kidney failure prediction. Equations therefore differed primarily in calibration against established thresholds while discrimination remained largely preserved, particularly when eGFR was contextualized with additional predictors such as UACR. Second, I translated eGFR uncertainty into expected discordance at clinically meaningful GFR and kidney failure risk thresholds and quantified confirmatory testing trade-offs at the population level. Threshold-adjacent eGFR uncertainty emerged as a broader source of potential confirmatory testing burden than guideline-recognized clinical indications alone. Simple proximity-based testing closely approximated probabilistic prioritization at GFR thresholds, but capturing more expected discordance required progressively greater testing with declining yield. This work reframes eGFR uncertainty according to whether it is likely to change a clinical decision. Finally, I evaluated whether routinely collected laboratory history could improve kidney failure risk assessment without requiring new biomarkers or sacrificing interpretability. In over 270,000 adults with CKD, I compared static and longitudinal models while holding population, laboratory inputs, and modeling framework constant, isolating the prognostic contribution of laboratory histories. Repeated measurements improved predictive performance across laboratory settings and time horizons. Much of this gain was captured by simple and clinically recognizable concepts including more robust estimates of current kidney function, historical baseline, and variability. These findings demonstrate that established risk equations can be extended using measurements already collected during routine care and concepts clinicians may already incorporate informally. By linking population-aware evaluation, uncertainty-aware interpretation, and trajectory-aware risk assessment, this dissertation provides a clinically grounded framework for evaluating and refining kidney function and kidney failure risk assessment.