Our paper included two related but separate meta-analyses. The first analysis focused on nondiabetic renal disease and the second was limited to diabetic renal disease. We clearly stated that the methods and robustness of the results differed for the two meta-analyses.

The meta-analysis of nondiabetic renal disease included only randomized, controlled trials and compared the proportion of patients who developed renal failure or died. We concluded that the result was robust and indicated no heterogeneity among studies. Dr. Parving apparently agrees. However, Shah and colleagues suggest that the range in the level of renal function at baseline might compromise the relevance of the results. We strongly disagree. Other studies have shown that reduction in glomerular filtration rate is associated with widespread renal parenchymal damage [1, 2]. Even in the MDRD study, which had the highest mean glomerular filtration rate at baseline (38 mL/min per 1.73 m²), 27 of 294 patients in the usual-protein diet group (92%) developed renal failure or died during a mean follow-up of only 2.2 years. Thus, we consider even a moderate reduction in glomerular filtration rate to reflect serious, life-threatening renal disease. Further, experiments with animals have shown that the application of a low-protein diet at the onset of renal disease (mild renal insufficiency) slows progression [3], whereas the application of a low-protein diet in humans with severe renal insufficiency alleviates symptoms of uremia. We agree that it would be interesting to determine whether a low-protein diet is beneficial to specific subgroups of patients. However, such analyses would require data from individual patients and probably more patients than those in the five studies that we examined.

On the other hand, we specifically mentioned that the results of the meta-analysis of studies of diabetic renal disease were not as robust. Fewer randomized, controlled studies have been done; therefore, we also included two nonrandomized crossover studies. In these studies, few patients developed renal failure or died; we therefore used surrogate outcomes that we based on the stages of diabetic nephropathy. For example, the study by Dullaart and colleagues [4] included 30 patients with normo- or microalbuminuria; thus, urinary albumin excretion was used to assess progression of nephropathy. These methodologic differences are a limitation of the second meta-analysis. Further, we agree with Shah and colleagues that all clinical investigations, including meta-analyses, should contain interpretations of both statistical and clinical significance. Although we observed a statistically significantly reduced risk for worsening renal function (P 0.001), we interpreted the results as a "strong indication, but not conclusive proof." On the basis of clinical judgment, we recommended dietary protein restriction to certain patients with diabetic renal disease.

Dr. Parving also questions the interpretation of changes in renal function based on measurement of serum creatinine levels or creatinine clearance. Levey and colleagues [5] have shown that a low-protein diet affects production and secretion of creatinine in addition to glomerular filtration rate. However, because a low-protein diet causes a decline in creatinine secretion, the apparent beneficial effect on creatinine clearance in the studies that we included probably reflects a beneficial effect on glomerular filtration rate.