The recent meta-analysis of low-protein diets in diabetic and nondiabetic renal disease by Pedrini and colleagues [1] contains examples of three limitations of meta-analysis in clinical medicine. First, the criteria used for study selection were inconsistent. For nondiabetic nephropathy, the meta-analysis included only randomized trials that had end points of end-stage renal disease or death. The meta-analysis of diabetic nephropathy, however, included two observational studies, and only one of three randomized trials used intention-to-treat analysis. In addition, the studies of diabetic nephropathy used surrogate outcomes of glomerular filtration rate, urinary albumin excretion rate, and creatinine clearance. Thus, the studies of nondiabetic nephropathy were held to a higher standard than were the studies of diabetic nephropathy.

Second, the lack of attention to clinical distinctions of patients in different studies threatens the clinical relevance of the results. Among studies of nondiabetic nephropathy, the mean serum creatinine level ranged from 1.9 to 7.4 mg/dL; among studies of diabetic nephropathy, urinary protein excretion ranged from 48 to 5700 mg/d. Although the point estimates for the effectiveness of low-protein diets may be mathematically correct, these average results among patients with disparate severities of illness may not be accurate for individual patients.

Third, the statistical focus of a meta-analysis can be a problem rather than a virtue. In the meta-analysis of 1413 patients with nondiabetic nephropathy, the overall P value was 0.007. In the meta-analysis of 108 patients with diabetic nephropathy (which used only surrogate outcomes), the corresponding P value was 0.00036. Despite the enhanced statistical significance of the second meta-analysis, the authors state that "these studies included fewer patients and used more varied study designs and surrogate end points. In our view, these results are not as strong a justification for the use of dietary restriction protein [sic] in routine clinical practice." Thus, when interpreting results, the authors resort to their best judgment-a time-honored practice that is, in our view, desirable but that meta-analysis was intended to replace.