Review of 'Head to head comparisons in performance of CD4 point–of-care assays: A Bayesian meta-analysis (2000-2013)'

The authors conducted a comprehensive systematic review and meta-analysis of published data on the accuracy of rapid, point-of-care CD4 devices. They appropriately used the mean bias of a Bland-Altman analysis as the primary measure of comparison, which is a more appropriate measure than correlation coefficients for comparison of continuous variables with non-normal distributions. They found Alere’s Pima machine to be accurate, and performed as well on capillary blood as compared to use on venous blood samples. In general, the authors have done and excellent study, but they have overstated their findings in several sections (described below).

The authors imply (in the abstract and text) that the Pima was better on capillary blood as compared to venous blood, as suggested by a simple comparison of the mean bias. However, the CI range encompasses the mean value, so there is no significant difference. The authors should change their terminology to say results from capillary blood ‘were equivalent to’ or ‘not significantly different from’ results from venous blood. This revision should also be made in the second paragraph of the Discussion. The final recommendation – that we should be testing capillary blood given easier operational use – remains unchanged.

Given that only 2 studies reported data for PointCareNOW (and with very different results), I suggest the authors be more cautious about drawing conclusions of this device (perhaps that one study was poorly executed). Within the Results (Bland-Altman section), rather than say the “PIMA device was superior”, without a direct comparison, the authors would be better served to describe the data – ‘Although data are limited, PointCareNOW had a higher mean bias than Pima.’ The second sentence of the Summary should be revised to ‘based on 2 studies, the PointCareNOW performed poorly in one study and sufficiently well in a second study, which highlights the need for additional data’. In the Conclusion, the paper would be better served to highlight that data for the PointCareNOW were limited and needs more data before being recommended for use (rather than stating it is inferior to Pima).

In the third paragraph of the Discussion, “current POC devices” should be “PIMA”. There was no clear evidence that the devices other than Pima worked better on capillary blood (perhaps equivalent to, but not better).

The authors should add to the limitations the variability of these studies. They have likely used different gold standard CD4 reference tests, were conducted in settings with different levels of infrastructure (perhaps some were done in a clinic), machines likely had different calibration procedure before testing, etc.