One hundred and sixty-one studies, with a total of 28,862 patients, were included in the review.
The overall pooled incidence rates were as follows: operative mortality 3.3% (95% confidence interval, CI: 2.9, 3.6), all endoleaks 22.8% (95% CI: 20.6, 25.2), all conversions to open repair 5.4% (95% CI: 3.1, 4.0; this value seems to be in error since it lies outside of the 95% confidence limits), post-operative rupture 1.3% (95% CI: 1.1, 1.7). The annual rates were as follows: type 1 endoleaks 8.4% (95% CI: 5.7, 12.2), type 2, 3 and 4 endoleaks 10.2% (95% CI: 7.4, 14.1), post-operative rupture 0.6% (95% CI: 0.5, 0.8).
The rates of operative mortality, post-operative rupture and total number of endoleaks all fell significantly over time (p<0.05). Estimates from the regression analysis indicated that from 1992 to 2002 approximate mortality rates fell from 7.5% to 1.4%, approximate rupture rates fell from 5.0% to 0.4%, and the approximate total endoleak rate fell from 43% to 13.5%. No other variables had a statistically significant effect over all three outcomes.
Further analyses were reported in the paper.
There was statistically significant heterogeneity between the studies for all outcomes (p<0.01), which the authors attributed to the large number of small studies, and the inclusion of results from the surgeons' 'learning curve' period. All the studies were hospital-based; 39 were prospective and hospital-based and 113 were retrospective. The authors noted that the effect of study quality on the results was hard to interpret but reported that there was little difference in all-mortality rates between the two different types of study. They also noted that many studies were poorly reported and that there could have been repetition of data between the studies, which could bias results as there is a greater risk that early results were counted twice.
Publication bias was also hard to evaluate on account of study heterogeneity in the funnel plots.