Just had another look at the EMView trial results. As a...

Just had another look at the EMView trial results. As a reminder, the data from EMView were assessed in terms of the sensitivity and specificity along two arms, blood and ischaemia. Today’s results revisit the ischaemia aspect, but nothing more as yet on haemorrhage – the latter is the more important of the two, so it’ll be interesting to see if anything new on haemorrhage is presented in the next few weeks. Failure to release an update (on bleeds) might have no significance, though it may be that it hasn’t been possible to improve on (??match) the original results.

The EMView haemorrhage figures showed sensitivity of 92%, with detection of 12 out of 13 bleeds. Relatively low statistical power due to the low numbers, but promising, though we’ll probably need haemorrhage detection (sensitivity) of close to 100%.This is the key clinical question, with stroke treatment guidelines stipulating the exclusion of a bleed on imaging as the criterion for thrombolytic treatment. With any test, increased sensitivity generally leads to decreased specificity due to an increase in false positives – finding a bleed that isn’t there due to lowering the bar for detection. EMVision will be trying to increase sensitivity without bringing the specificity down more than absolutely necessary, and I think it’s probably here that an 80% figure might be the cut-off for authorities like the FDA.

If close to 100% sensitivity in bleed detection is possible for EMV, then suboptimal (>80%) bleed specificity is likely to be acceptable – in practice this would mean that a low single figure percentage of all stroke patients (<20% of 1 in 9, roughly) would be wrongly labelled as having a bleed, delaying thrombolytics until they reached hospital and were able to undergo a CT. Significantly outweighed by the benefits of potential at-home (‘time is brain’) diagnosis and treatment of the much more common ischaemic stroke.