Cystatin C equations shift CKD staging without changing outcome prediction
Replacing the Chronic Kidney Disease Epidemiology Collaboration equation with the European Kidney Function Consortium cystatin C-based equations changed how chronic kidney disease was staged in women, but not in men. The reclassification moved women into milder CKD categories, yet prediction of kidney failure remained unchanged. That disconnect sits at the center of the debate over what an estimated glomerular filtration rate equation can legitimately claim to measure.
The problem is not merely statistical. Estimating equations are built to approximate filtration, but they are not direct measurements of it. In clinical practice, the number printed beside an eGFR result often carries more weight than it deserves, especially when treatment thresholds, referral decisions, and prognostic labels are attached to it. A change in equation can alter staging without necessarily improving the biological truth behind the number.
Prediction of kidney failure does not validate a GFR equation.
Why staging shifts do not prove better measurement
The central argument here is methodological and clinical at the same time. A kidney failure endpoint is influenced by more than filtration alone, including decisions to start dialysis, access to care, local practice patterns, and the clinical thresholds used by nephrologists. Dialysis initiation is a clinical decision, not a direct readout of filtration. When an equation predicts that outcome, it may be capturing a mix of kidney function, physician behavior, and disease severity rather than measuring glomerular filtration with greater fidelity.
That distinction matters because validation by outcome prediction can be seductive. If a formula predicts a future event well, it is tempting to call it the better estimate of GFR. Yet prediction and measurement are not the same task. A model can be useful for risk stratification while still being imperfect as a physiologic estimator, especially when the endpoint depends on human decisions as well as biology.
In this analysis, the shift in staging was sex-specific. Women were reclassified into milder CKD stages under the cystatin C-based equations, whereas men were not. The pattern raises immediate questions about how sex interacts with biomarker behavior and equation structure. It also suggests that the choice of estimating formula can change the apparent burden of CKD in a subgroup without necessarily altering the model’s ability to predict what happens next.
Creatinine and cystatin C both carry non-GFR noise
No estimating equation is insulated from biology outside filtration. Creatinine is affected by muscle mass, diet, and other nonrenal determinants. Cystatin C is also influenced by factors beyond GFR, so a cystatin C-based equation is not a pure solution to creatinine’s limitations. At the bedside, either marker can mislead when non-GFR determinants dominate the signal.
This is why a population equation should be treated as a tool, not an oracle. In large cohorts, average performance can look excellent even when individual patients are misclassified. A frail patient with low muscle mass, a person with inflammation, or someone with altered protein turnover may be assigned an eGFR that reflects biology other than filtration. The result can be overdiagnosis in one patient and under-recognition in another.
The practical consequence is easy to miss. A patient who moves from a more advanced CKD label to a milder one under a different equation may appear to have improved kidney health without any real physiologic change. The reverse can happen as well. These are not trivial bookkeeping differences; they shape follow-up intensity, counseling, and whether a patient is perceived as nearing kidney failure.
Cross-sectional design leaves longitudinal performance unanswered
The analysis was cross-sectional, which means it tested reclassification and outcome association at a single point in time rather than tracking how well the equations performed over years. Longitudinal performance remains untested here. That limits how far the findings can be extended into clinical forecasting, because kidney disease is dynamic and estimation errors may accumulate as body composition, inflammation, and illness evolve.
A cross-sectional comparison can show that one equation moves more patients between categories, but it cannot show whether those movements are clinically correct over time. It cannot determine whether the reclassified women truly had milder disease, or whether the new equation simply shifted the threshold at which disease is labeled. It also cannot determine whether earlier or later intervention based on that reclassification improves patient outcomes.
For nephrologists, the unanswered issue is not whether equation choice changes numbers. It clearly does. The real issue is whether those numbers reliably track the patient’s true renal trajectory in a way that improves care. Cross-sectional prediction of a hard endpoint does not settle that question, especially when the endpoint is partly shaped by treatment decisions.
Bedside interpretation should stay tied to physiology
The bedside lesson is to resist turning an eGFR equation into a final verdict. The equation should be read as an estimate filtered through population data, assay behavior, and non-GFR determinants of its biomarkers. When creatinine and cystatin C disagree, the disagreement itself may be informative, because it can point to atypical body composition, inflammation, or other factors that distort one marker more than the other.
That kind of uncertainty is exactly why no equation should be asked to do more than it can. A formula may help stage chronic kidney disease, estimate prognosis, and support discussions about referral or monitoring. It cannot, by itself, prove the true filtration rate in every patient. It cannot, by itself, validate the timing of dialysis. And it cannot be judged solely by how well it predicts a clinical endpoint that depends on more than filtration.
Schultheiss and colleagues place the field back on firmer conceptual ground by separating estimation from prediction. The new equations may alter CKD staging, especially in women, but unchanged kidney-failure prediction does not establish superior measurement. For clinicians, the message is straightforward: use estimation equations as population tools, interpret them in physiologic context, and avoid equating a predictive score with a direct measure of glomerular filtration.