Response: A case of rich fruit

Response

We thank Mark Crowther and his colleagues for the comments that they have made on our paper.¹  While we welcome their opinions, it is worth noting that systematic reviews and meta-analyses are areas where opinions differ, controversy remains, and comments on published work are common.

Their first point is that our search strategy was incomplete. They observe that a search using both text words and MeSH headings inclusive of conference abstracts results in an increased yield. It is unknown, however, if any of these studies would have been incorporated given our stringent inclusion criteria.

Secondly, Crowther and his colleagues state that our exclusion criteria may have led to bias. While these criteria were clearly delineated and judged by the authors to be clinically and epidemiologically appropriate, we acknowledge the restrictions they may have placed on our review, including the preclusion of potentially valuable data from studies published in non-English language journals and those using serological diagnoses. The former criterion was not possible given our available resources. The latter may have excluded some high-quality studies, but its removal would likely have resulted in the inclusion of many studies of lower quality; the sensitivity and specificity ranges of the enzyme-linked immunosorbent assay (ELISA) test are greater than for the ¹³C-UBT test and do not discriminate between patients with active infection and those whose infection has been eradicated.² 

Their third point relates to homogeneity. We did have reservations about quantitatively synthesizing data because of the heterogeneity of results. However, we thought the studies to be sufficiently similar in scope to justify the pooling of their results. While visual inspection of the funnel plot (Stasi et al, Figure 2) does not show the perfect inverted V consistent with the absence of publication bias, the figure does show some decrease in spread with decreasing standard error. It would be implausible to think that a review such as this would be entirely free of bias. The heterogeneity noted is likely to be the result of small sample sizes (and in this analysis many studies are small) and biologic factors. Importantly, our exploration of this variability uncovered the striking correlation between country-specific infection prevalence and response rate illustrated in Figure 4.¹ 

We defend our use of the DerSimonian-Laird method to pool data across studies. Crowther and his colleagues suggest that this method increases the weight given to large observational studies, which may be more susceptible to bias. As a random-effects computation, it gives greater weight to smaller studies than conventional fixed-effects methods.³  Like most pooling methods, the DerSimonian-Laird method does partly weight studies by sample size, which we feel is appropriate. Although it is possible for large-scale observational studies to be susceptible to greater bias, this tendency is largely based on the methodology used, which was controlled for by the threshold established by our inclusion criteria. Furthermore, our investigation dealt almost exclusively with studies of small numbers of patients within a single center, a situation in which random fluctuation assumes a greater, potentially more dangerous role in impacting results.

In summary, we believe the methodology for our systematic review to be sound and support its finding of Heliobacter pylori eradication as a viable, noninvasive, and low-cost treatment for ¹³C-UBT–positive adults with immune thrombocytopenia, particularly in countries with a high prevalence of infection pending the outcome of large-scale randomized trials.