Critical review of meta-analyses of randomized clinical trials in hepatogastroenterology

Tags: Poster
Auperin A, Pignon JP, Poynard T, Pignon JP

Objective: This study aimed to identify meta-analyses performed in hepatogastroenterology, to assess their quality and to identify simple criteria of good quality.

Methods: Meta-analyses of randomized clinical trials in hepatogastroenterology published as full papers before 1993 were included. They were identified by computerized and hand searches. 27 items, adapted from Sacks (N Eng. J Med. 1987; 316:450), were evaluated in each publication, independently by 2 observers, and coded 2 if adequate, 1 if partial and 0 if missing or unknown. A quality score was defined as the sum of the rating of the 27 items. Meta-analyses were categorized into 3 classes according to the score (poor quality for the 25% lowest scores, high quality for the 25% highest and middle for the remaining). Quality criteria were identified by discriminant analysis through these 3 groups.

Results: From 1981 to 1992, 62 publications were identified, including 42% in the two last years. They reported 180 meta-analyses of which 43% concerned peptic ulcers, 13% esophageal varices and 6% digestive cancers. The quality score ranged from 14 to 40, the median was 29. 25% of the studies obtained a score lower than 23 and 25% a score higher than 34. Several stages of meta-analysis were poorly performed: identification and selection of trials, study of trial quality, data extraction and achievement of sensitivity analyses, subgroup and indirect analyses. Some of them led to biases and questions about the validity of results. Five criteria were significantly associated with high quality: presence of a protocol, assessment of trial quality, only randomized trials pooled, achievement of sensitivity analyses and peer-reviewed publication.

Discussion: In hepatogastroenterology, there is an exponential increase in publication of meta-analyses. Their quality is heterogeneous. To distinguish between poor and high quality studies, five criteria are proposed