In reply

Sepe et al express concern about potential confounding among the studies in our meta-analyses and call for discussion of the rules to select reports based on their finding only a 10-mg/dL (0.3-μmol/L) difference in mean plasma fibrinogen between the 13 peritoneal dialysis (PD) and 54 hemodialysis (HD) patients in their small, select sample—less than any value contained in set 1 (67 to 343 mg/dL [2.0 to 10.1 μmol/L]).1

Two studies we cited that measured plasma fibrinogen serially in patients initiating PD or switching from PD to HD1 should alleviate their concern about confounding. These observations were consistent with the estimates from the meta-analyses and not susceptible to confounding by patient characteristics (except for changes in proteinuria over time, if any). The analysis limited to nondiabetics was simply an exploratory sensitivity analysis done to look at patients who probably had little proteinuria, not to minimize confounding by diabetes. The meta-analysis of set 2 was done to validate the findings of set 1 in an independent set of publications, not to control for time of publication.

Other than residual proteinuria, which conceivably could be higher in new PD patients if they retain more renal function than comparable new HD patients, we are unaware of a systematic association between dialysis modality and the factors known to affect plasma fibrinogen (eg, inflammation and smoking). Chance differences in these factors may explain the smaller difference found by Sepe et al. Another possible explanation is sampling error. In small samples, even if unbiased, chance or random variation will often account for outlying mean values. Irrespective of its cause, the observation by a reader of a smaller difference is more likely to stimulate correspondence than would any of the other hypothetical comparisons performed by other readers. This undoubtedly produces selection bias and makes it impossible to interpret the data found by Sepe et al. Nonetheless, rules for selection of reports are an important aspect of meta-analysis, a subject that has been described extensively; below are suggestions for those who may be interested in pursuing this aspect of meta-analysis.2, 3