Bayesian hierarchical meta-analysis model for medical device evaluation: application to intracranial stents

The aim of this study was to propose a statistical model that takes into account clinical data on earlier versions when evaluating the latest version of an implantable medical device (IMD).

We compared the performances of a bayesian three-level hierarchical meta-analysis model with those of a bayesian random-effects model through a simulation study. Posterior mean estimates of the success rate for each IMD version were computed as well as the probability that the latest version improved in effectiveness. Models were compared using the Deviance Information Criterion (DIC), the estimated bias and the standard deviation of the mean success rates. Sensitivity analyses to the choice of the priors were performed. These methods were applied to the evaluation of an intracranial stent used to treat wide-necked aneurysms.

When IMD versions did not differ in effectiveness, the best-fitting model was the random-effects model. By contrast, when there was a version effect, the hierarchical model was selected in more than 95 percent of the cases. It provided precise estimations of success rates of each IMD version and allowed detecting an improvement in effectiveness of the latest version, with a low influence of the choice of the priors. No evidence of benefit from the latest version of the intracranial stent was found.

In the setting of IMD assessment, comparison of DIC between the two proposed models appeared useful for detecting version effects. In that case, bayesian hierarchical meta-analysis model may help the decision maker by providing useful information on the latest version of IMD compared with the previous versions.