High dimensional variable selection with clustered data: an application of random multivariate survival forests for detection of outlier medical device components