Adjusting for publication biases across similar interventions performed well when compared with gold standard data

To extend, apply, and evaluate a regression-based approach to adjusting meta-analysis for publication and related biases. The approach uses related meta-analyses to improve estimation by borrowing strength on the degree of bias.

The proposed adjustment approach is described. Adjustments are applied both independently and by borrowing strength across journal-extracted data on the effectiveness of 12 antidepressant drugs from placebo-controlled trials. The methods are also applied to Food and Drug Administration (FDA) data obtained on the same 12 drugs. Results are compared, viewing the FDA observed data as gold standard.

Estimates adjusted for publication biases made independently for each drug were very uncertain using both the journal and FDA data. Adjusted estimates were much more precise when borrowing strength across meta-analyses. Reassuringly, adjustments in this way made to the journal data agreed closely with the observed estimates from the FDA data, while the adjusted FDA results changed only minimally from those observed from the FDA data.

The method worked well in the case study considered and therefore further evaluation is encouraged. It is suggested that this approach may be especially useful when adjusting several meta-analyses on similar interventions and outcomes, particularly when there are small numbers of studies.