The use of non-randomized studies in the Cochrane reviews on drugs and alcohol

Article type
Authors
Vigna-Taglianti F, Mathis F, Foxcroft D, Amato L, Davoli M, Faggiano F
Abstract
Background: The inclusion of non-randomized studies in systematic reviews is under debate. In fact, results of meta-analyses produced by observational studies are often different or inconsistent with those coming from RCTs. The Cochrane Drug and Alcohol Group (CDAG) set up general recommendations on the study design to be used in the systematic reviews within the group, as well as on the quality appraisal. Recommendations are given for cases in which observational studies have to be included and how. In drug abuse research several areas are suspected to lack in RCTs. In these cases, to use non-randomized studies seems to be a matter of necessity and not of choice.

Objectives: To assess the proportion of the CDAGÂ’ reviews including non randomized studies, the quality, and the contribution of non randomized studies to the pooled results.

Methods: Systematic reviews published in the Issue 2 of the Cochrane Library on behalf of the CDAG (n=26) were reviewed. Number of randomized and non-randomized studies, as well as quality scores obtained by non-randomized studies were recorded. Their contribution to the results was evaluated by checking the proportion of outcomes including at least a non-randomized study. Excluded studies were also reviewed, registering the relative proportion of non randomized studies.

Results: Eighteen out of 26 CDAG reviews (69%) declared non-randomized designs among criteria for considering studies; 10/26 reviews (38%) included non-randomized studies. In these reviews, non-randomized studies were on average 30% of the included studies. Where quality scores were specified (7 out of 10 reviews including non-randomized designs), 55% of non-randomized studies were evaluated as moderate or low quality studies. On average, about 33% of the outcomes produced by these reviews included non-randomized studies. Where study design was specified, an average of 68% of the excluded studies were non-randomized studies.

Conclusions: Few reviews included non-randomized studies; their contribution to the scientific results of the CDAG reviews seems very little. The inclusion of non-randomized studies appear to be justified only when rare outcomes (e.g. death) or specific objectives are evaluated (e.g. effectiveness when ethical constraints hinder random allocation), or there is a clear lack of RCTs.