The authors reply
In our analysis, we used the inverse variance random effect method that “…incorporates the assumption that different studies are estimating different, yet related, intervention effects” (3) and considers potential heterogeneity. The generic inverse variance method, proposed by Dr. Parienti (1), is generally recommended whenever it is not possible to enter data in the usual form of dichotomous data. In the current meta-analysis, however, no study was excluded due to lack of dichotomous data. Additionally, only three of the included studies in the internal jugular-subclavian comparison provided data on dwelling time while no study provided dwelling time per patient, which could have permitted catheter-days calculation. Furthermore, the analysis of the count data as rates is not always the most appropriate approach, since the assumption of constant underlying risk may not be suitable (3).
Dr Parienti (1) added in his analysis two trials randomized for other than the central venous catheterization (CVC) site outcome (Timsit et al  and Mimoz et al ). Regarding the study by Timsit et al (4), in our CRBSI risk comparison between internal jugular-femoral, we included data from the study by Timsit et al (6), a study that combined data from two preceding randomized for other than CVC site outcome trials, one of which was the study by Timsit et al (4). Occurrence rates or ratios (risk or hazard ratios) of CRBSI in subclavian catheters were not reported in the studies by Timsit et al (4, 6). Consequently, similar data could not have been retrieved and included in our internal jugular-subclavian CRBSI metaanalysis. Also, Dr. Parienti (1) refers to the study by Mimoz et al (5), which is a randomized trial for other than the CVC insertion site outcome that presented occurrence rates on CRBSIs for internal jugular and subclavian sites. The study by Mimoz et al (5) was not included in our meta-analysis as occurrence rates were only reported in the appendix section. Nevertheless, we reran our analysis on internal jugular-subclavian CRBSI risk by adding the study by Mimoz et al (5) and found no difference from our initial results (risk ratio [RR], 1.32 [0.65–2.69]; I2 = 59% [Fig. 1) vs RR, 1.16 (0.57–2.35); I2 = 58% [Fig. 3B in our meta-analysis ). Consequently, the study by Mimoz et al (5) could not have influenced our results and certainly could not have altered our conclusions.
Five, not seven, CRBSIs were reported in the study by Arvaniti et al (7); zero events were found in internal jugular and femoral and five in subclavian catheters. Since we had zero events in both arms in the internal jugular-femoral comparison, we proceeded in a zero events correction.
We agree with the remarks by Dr. Parienti (1) on the important heterogeneity (I2 = 58%) and the considerable weight of observational studies (83.3%) on internal jugular-subclavian CRBSI comparison, which has already been reported in our study (2).
We consider that analysis of data from randomized controlled trials and observational studies does not provide, yet, sufficient evidence to conclude that selection of subclavian compared to internal jugular is associated with reduced CRBSI risk in ICU patients.