This is a bit embarrassing to admit, but as a nephrologist with epidemiology training, I've never been particularly excited by glomerular filtration rate (GFR) estimating equations.
Don't get me wrong—it's a fun statistical exercise to take a few factors (serum creatinine, age, sex, and race) and try to estimate the GFR measured by inulin or iohexol clearance. After all, the latter measurements are rarely performed in clinical practice because of their inconvenience and cost. A simple equation is much, well, simpler. But have these endeavors changed my practice? I'm not so sure.
I could argue that we should only be outcomes-focused and that GFR is really only important insofar as it predicts when a patient will need dialysis or transplant or a change to the dosing of a particular drug. This is a bit deterministic, though, and deep down, I see the value of trying to get at a true physiologic measurement. Creatinine, transformed by an eGFR equation, becomes more than just a biomarker; it actually means something.
But my enthusiasm along these lines can only go so far. Efforts to improve existing equations are interesting from a statistical perspective—maybe even a history of medicine perspective. But clinically? Let's just say that my clinical lab (and yours too, in all likelihood) is still using the quite antiquated Modification of Diet in Renal Disease (MDRD) Study equation, when most of us agree that the Chronic Kidney Disease-Epidemiology Collaboration (CKD-EPI) equation[1] is the gold standard.
Of course, none of the equations we've developed are perfect. Two relevant subgroups have been notoriously difficult to model by GFR estimating equations: patients with relatively normal GFR,[2] and elderly persons.[3]
Because of wide variations in diet, muscle mass, comorbidities, and (presumably) creatinine generation, GFR estimating equations are less reliable in an elderly population than in younger cohorts. This has prompted a minor boom in the estimating equation industry, with several new equations popping up in recent years.
An article in JAMA Internal Medicine[4] does yeoman's work in evaluating four contenders for the eGFR-of-the-elderly crown and concludes (spoiler alert) that the CKD-EPI equation is probably just fine. I've broken down some of the results according to each equation.
eGFR Equation | Source Population | Includes a Race Variable? | RMSE | P30 |
---|---|---|---|---|
CKD-EPI | North American/European population | Yes | 0.195 | 78% |
Lund-Malmö Revised (LMR)[5] | Swedish population | No | 0.185 | 82% |
Full Age Spectrum (FAS)[6] | European population | No | 0.188 | 79% |
Berlin Initiative Study (BIS-1)[7] | Older adults | No | 0.189 | 78% |
eGFR = estimated glomerular filtration rate; CKD-EPI = Chronic Kidney Disease-Epidemiology Collaboration; P30 = percentage of estimates within 30% of the measured value; RMSE = root mean square error |
Are Any of These Equations Actually Good?
Before we close the book on this one, it's worth asking just how you decide how good your estimating equation is. There's no one metric here, and the authors of the JAMA Internal Medicine article classify the equations using no less than six measures of "accuracy." Fear not—we won't go through them all.
Although the root mean square error will be familiar to those of us who have dabbled in linear regression, the most clinically intuitive accuracy metric is the P30, defined as the percentage of estimates that are within 30% of the true value. All four equations had P30 values of around 80%, meaning there was no clear winner. But more interesting (to me, at least) is just how bad that is, if you think about it.
Imagine that a patient has a measured GFR of 60 mL/min/1.73 m2. By the P30 metric, we'd consider our equation a success if it estimated the GFR anywhere between 42 and 78. In one case, that's fairly severe chronic kidney disease; in the other case, we might think of it as "normal variation," particularly in an older population. And now realize that 20% of the time, we'll be estimating results that are even farther afield.
If that diminishes your enthusiasm for GFR estimating equations, I feel you. We can, however, be encouraged by the area under the curve measurement of accuracy, which exceeded 0.9 in all cases. What this means is that, given two individuals, one of whom has a measured GFR < 45 and one > 45, the estimating equation will identify the lower individual in about 90% of cases.
Now, I wrote that all the equations performed similarly, but CKD-EPI had one hand tied behind its proverbial back: The cohort did not contain information on participant race. As the only equation that uses race, we might expect that the CKD-EPI performance would have been better had it been given all the information it requires.
In sum, all four equations are using the same raw data points. They just combine them in different ways. Advances in prediction often require adding data points to the equation. The best way to improve the accuracy of GFR estimating equations may be by adding a new biomarker, cystatin C.[8] Yes, it's an additional measurement, but if we really want good numbers, it's probably our best bet.
But do we really need a better measurement? I think one of the reasons I've never been excited about the accuracy of our GFR estimating equations is that they were all built in isolation, one measurement per patient. But that's not how we practice. Although the estimated GFR is important to my decision-making, it's not nearly as important as the change in creatinine (and, ergo, GFR) in the individual patient.
I'm glad that our tools, including GFR estimating equations, are being rigorously evaluated. Fortunately, in this case, the old standby CKD-EPI equation seems to be as good as any of the others. Now if only we can get our clinical labs to report it.
Follow Medscape on Facebook, Twitter, Instagram, and YouTube
Medscape Nephrology © 2019 WebMD, LLC
Any views expressed above are the author's own and do not necessarily reflect the views of WebMD or Medscape.
Cite this: Estimating GFR in the Elderly: Which Equation Is Best? - Medscape - May 24, 2019.
Comments