Cystatin C equations change staging in women without improving prediction
Schultheiss et al. revisited a familiar problem in nephrology: how best to estimate glomerular filtration rate when no laboratory equation can directly measure it. Their comparison of the Chronic Kidney Disease Epidemiology Collaboration equation with the European Kidney Function Consortium cystatin C-based equations produced a clear pattern. Women were reclassified into milder chronic kidney disease stages, while men were not, yet the ability to predict outcomes did not improve. The result is clinically useful precisely because it separates two ideas that are often conflated in everyday practice: estimating kidney function and predicting what happens next.
The distinction matters at the bedside. A patient can move from one estimated stage to another because the equation has changed, not because the kidney has changed. A lower-risk label may feel reassuring, but it is still a product of model structure, biomarker selection, and the population used to build the equation. In this analysis, the shift toward milder categories among women did not translate into better outcome discrimination. The equation altered classification, not clinical truth.
That outcome has a practical consequence for interpretation. If two equations rank risk similarly, better prediction alone does not prove that one more accurately estimates true glomerular filtration rate. Prediction is about prognosis; filtration is about physiology. Those are related, but they are not interchangeable, and they answer different questions in nephrology clinics and research studies.
Prediction is not measurement, and dialysis is not an endpoint for equation validation
The commentary takes direct aim at a recurring shortcut in the field: using kidney failure or dialysis initiation as a surrogate test for whether an equation is “right.” Dialysis initiation is a clinical decision, not a direct filtration measurement. It reflects symptoms, volume status, metabolic complications, patient preference, local practice, and access to care, in addition to the underlying loss of renal function. Because of that, an equation can appear to perform well against dialysis-based outcomes while still missing the underlying biology of glomerular filtration.
This is where the logic of validation becomes slippery. A formula that predicts future kidney failure well may still be imperfect as a measure of actual GFR. Conversely, an equation that tracks measured filtration more closely may not always provide the strongest risk prediction for downstream events. The two goals overlap, but they should not be collapsed into one another. The commentary argues that outcome prediction cannot settle the question of which equation is the best estimate of GFR, because the clinical endpoint used for validation is itself shaped by human decision-making.
The issue is not academic. Treatment decisions, referral timing, and staging discussions often follow from the estimated GFR number placed in the chart. When an equation reclassifies women into milder stages, clinicians may be tempted to interpret the shift as a correction of overdiagnosis or a better alignment with physiology. The data presented here do not support that leap. Reclassification occurred, but improved prognostic performance did not follow.
Non-GFR determinants can distort creatinine and cystatin C at the bedside
The commentary also reminds readers that both creatinine and cystatin C are vulnerable to influences outside filtration. That limitation is especially important when equations are applied to individuals rather than populations. Muscle mass, inflammation, body composition, and other non-GFR determinants can alter biomarker levels and shift estimated kidney function away from the patient’s true physiologic state. An equation can therefore look more precise than it is, especially when used as if it were a direct readout of nephron filtration.
This problem becomes sharper when clinicians compare creatinine-based and cystatin C-based estimates as if one were inherently “truer” than the other. In reality, each biomarker carries a different bias profile. Cystatin C may reclassify some patients downward or upward depending on the clinical context, but the direction of change does not automatically reveal better measurement. The bedside challenge is to recognize when discordance reflects disease biology and when it reflects the biomarker’s own non-GFR determinants.
The commentary’s message is not that equations are unhelpful. They remain essential population tools for staging, risk stratification, and trial design. Their limits become visible when they are asked to do more than they can do. GFR estimation equations can rank groups, but they cannot by themselves prove what a patient’s true filtration is. They can also not resolve the gap between statistical prediction and clinical judgment.
A better test of an equation would need longitudinal physiology, not a single cross-sectional comparison
A key weakness of the comparison is its cross-sectional design. Longitudinal performance remains untested. Without follow-up across time, it is difficult to know whether the observed reclassification reflects stable physiologic improvement in estimation or merely a different snapshot of the same population. Cross-sectional analyses are useful for describing how often equations disagree, but they are limited when the clinical question is whether one estimate better tracks true kidney function over time.
That limitation is central to the article’s argument. If the goal is to estimate GFR, the evaluation must reach beyond associations with future events and ask whether the equation follows measured physiology across changing clinical states. If the goal is prediction, then the model should be judged as a prognostic tool, not disguised as a filtration assay. Confusing those roles has long complicated nephrology research, and this commentary returns the discussion to first principles.
The practical takeaway is sharp. When an estimated GFR changes because the formula changes, the number has changed less than the disease state may seem to have changed. Clinicians should read the estimate as one input among several, not as a direct substitute for physiology or as proof that one equation has settled the matter. The more pressing unanswered question is not which equation predicts kidney failure best, but which one most faithfully estimates filtration in real patients across time and context.