Letter: the effect of sirolimus on recurrence and survival in liver transplant recipients with hepatocellular carcinoma—authors’ reply

We read with interest the Letter by Grigg et al1 regarding our study.2 The authors have raised data extraction and methodological errors in the meta-analysis that we would like to clarify.

Grigg et al, with regard to Zimmerman et al3, raise concern regarding the change in recurrence and recurrence-related mortality numbers in the pooled analysis. At the outset, we stated clearly in our methods that the numbers that were used for the pooled analysis were extrapolated from the Kaplan-Meier plots, which is standard protocol for carrying out meta-analysis when the data are incongruent.4 In Figure 1 (page 636), the Kaplan-Meier plot of the HCC recurrence-free survival (RFS) shows the correct number of events in both groups.3 The sirolimus (SRL) group RFS falls at 84% (recurrence rate of 16%; 7/45) and, for the calcineurin inhibitor (CNI) group, 71% (recurrence rate of 29%; 15/52). The correct numbers were extracted as per standard meta-analysis protocol4 and, as we did not reiterate this in the results, that may have led to confusion with regard to validity of the data.

Grigg et al also question the overall mortality and RFS data from Chinnakotla et al.5 There is a transcription error in the data extraction that does not change the overall outcome of the meta-analysis, as shown by the revised forest plot.

With regard to the RFS statement by Grigg et al, we believe this is incorrect. Chinnakotla et al clearly state time to recurrence probability (page 1839, Figure 3) from which we calculated the 1, 3, and 5-year RFS as 94%, 85%, and 80%, respectively.

Grigg et al also state that we should have applied a random-effects model. As clearly mentioned in our methods section, we assessed the heterogeneity among the included studies using the I² statistic. With regard to type of statistical model used, it has been universally established that, if the number of studies is small and the within-studies variance is large, the test based on the Q statistic may have low power even if the between-study variance is substantial.6 We used a fixed effect model, as the Q statistic variance was 0% and as smaller studies would be given very little weight. Heterogeneity among the transplant studies is not uncommon, but routine use of random-effects in a small study size is unjustified as the results cannot be generalised. The previous meta-analysis referenced in our review also used a fixed effect model.7

The authors also comment on the heterogeneity statistics on our recurrence-related mortality stating that we had used the I² results based on the overall mortality. This is incorrect as we did use the recurrence-related mortality from the two studies included in the pooled analysis.3, 5 We have carried out the heterogeneity analysis again and did not find any difference for either recurrence-related or overall mortality.

Therefore, in summary, the methodology used in our meta-analysis is supported by strong statistical evidence6 with validation from previous similar studies7 and the minor transcription error in data extraction has not altered the outcome of the pooled analysis.