The MDRD formula and validation

Sir, MacGregor et al . are quite correct to point out that the MDRD equation will improve identification and management of chronic kidney disease. 1 However, the MDRD four-variable formula is not as well-validated as one might hope. Firstly, despite the initial MDRD equation being based on 1628 samples, 2 there were insufficient patients with K/DOQI stages 1 and 2, as pointed out by MacGregor et al . However, this may be responsible for the significant negative bias of 29% reported in healthy persons. 3 Secondly, the four-variable equation was published as an abstract, 4 although the seven-variable paper 2 was subject to full peer review. Thirdly, the authors unnecessarily log-transformed the data, to reduce the increasing absolute difference between the reference method and the estimated values with increasing glomerular filtration rate (GFR). As a result, the model used may not be optimal or mathematically justifiable. Indeed, the formula was developed 4 by empirical means, with no attempt to use an appropriate mathematical model. Such models may be acceptable as rough sketches of relationships, but one should not depend on them for making precise clinical decisions. Empirical models, and in particular those built using log-transformed data, are weak, because they can result in the propagation of errors. Fourthly, the reference method used, iothalamate, is positively biased when compared to the gold standard GFR method (inulin 3–5 ml/min at low levels of GFR and 15–25 ml/min in healthy subjects 5–8 ). Finally, there is the issue of creatinine standardization, which MacGregor et al . mentioned: the constant factor in the four-variable MDRD equation has been changed from 185 to 176 9 for creatinine methods that have been calibrated to be traceable to isotope dilution mass spectroscopy. It was subsequently suggested by the UK National External Quality Assessment Service (UKNEQAS) that slope and intercept adjusters for the creatinine methods could be used to approximate non-IDMS traceable creatinine results to an IDMS standardized method. 10 We have shown that depending on which equation is used, there is the potential for a significant effect on the classification of chronic kidney disease primary care. 11 Estimated GFR is potentially an excellent method of detecting and monitoring chronic kidney disease, as it is quick, cheap and simple. Answers to some of our questions will not negatively affect eGFR. However, as long as they remain unanswered, eGFR may remain open to criticism.