Re: Efficacy of influenza vaccination in HIV-positive patients: a systematic review and meta-analysis

Sir,

In their recent reassessment of the efficacy of influenza vaccines in HIV-infected individuals, instead of recognizing analytical differences as the source of rather negligible differences between their analyses and ours [1], Anema et al. [2] incorrectly claim that in our study “numerical values were incorrectly entered into the meta-analysis, leading to skewed outcomes”. Unfortunately they do not specify what numbers were supposedly incorrectly entered into our meta-analysis, or what outcomes were skewed – even Anema et al. [2] found a similar summary risk difference of −0.27 in re-analysing our data. As we specified in our original report, the data were independently abstracted by two of the authors (JA and LK). We have double-checked our meta-analysis and still could not identify any incorrectly entered number.

Differences between the two analyses include (1) their choice of risk ratios, also referred to as “relative risks” (and not risk differences) as a measure of effect, (2) their decision to exclude a case–control study and (3) their choice of different influenza outcome measures, particularly for the study by Tasker et al. [3]. These are legitimate analytical differences, not necessarily limitations, that merit discussion. We choose to use risk differences because these have the inherent advantage of better reflecting the absolute effect of an intervention, as numbers needed to treat can be easily inferred from them. Furthermore, although Fine et al. [4] designed and referred to their study as a case–control study, acknowledging the relative acuteness of influenza disease (particularly in the context of an outbreak), the influenza cases could be assumed to be incident cases, rendering this study similar to the other two observational nonrandomized prospective studies and allowing one to estimate a risk difference. Finally, the study by Tasker et al. [3] presented results for more than one outcome. In our meta-analysis, we chose to use the more specific outcome based on a definition of a case as a “laboratory-documented influenza A infection (by culture or serologic examination)”. While this is certainly not the only possible case definition, it is at least as legitimate as any other choice that could have been used.