Sensitivity analyses allowed more appropriate and reliable meta-analysis conclusions for multiple outcomes when missing data was present

Abstract

Objective: A major problem for meta-analysis of multiple outcomes is the unavailability of some estimates from published and unpublished studies. Dissemination bias, in how and what outcomes are reported or published, may be causing this incompleteness. This article illustrates these problems and presents possible sensitivity analyses to allow the most reliable conclusions.

Study design and setting: In a systematic review of prognostic marker MYC-N in neuroblastoma, meta-analysis for overall survival (OS) and disease-free survival (DFS) was of interest. Only 17 published studies enabled extraction of both outcome estimates, 25 enabled only DFS, 39 enabled only OS, and 70 enabled neither outcome. Unidentified unpublished studies may also exist. We assessed the robustness of the pooled estimates to the problem of missing information. Because OS and DFS estimates seemed to be related, we used the known outcome estimates to predict estimates known to be missing, and combined this approach with existing methods for assessing dissemination bias.

Results: The results of the sensitivity analyses suggested that the original meta-analysis results were likely to be an overestimate of the true OS and DFS effect-sizes but strengthened the belief that MYC-N is a potentially important prognostic marker in neuroblastoma.

Conclusion: Sensitivity analyses in meta-analysis allow more appropriate and reliable conclusions when problems such as unavailable estimates and dissemination bias are present.