Ali Saad, MD

Ay H, Arsava EM, Andsberg G, Benner T, Brown Jr RD, Chapman SN, et al. Pathogenic Ischemic Stroke Phenotypes in the NINDS-Stroke Genetics Network. Stroke. 2014

This group created an online tool where you can enter data containing a patient’s demographics, neuroimaging, vessel imaging, cardiac workup, and other diagnostics.The tool then uses an algorithm to determine the most likely stroke phenotype (by TOAST criteria) and cause. Users were stroke attendings, fellows, or neurology residents and from the US and Europe. They found that there was good inter-rater reliability between users for phenotype (kappa 0.73, 95%CI:0.70-0.75) and etiology (kappa 0.72, 95%CI:0.69-0.75). They also found that cause and phenotype did not necessarily correlate, as expected, because a patient might have a stroke risk factor found on work up that was not the immediate cause of the incident stroke. you can try the tool out for free right here.

So what’s the big deal? The fact that we can create such an algorithm for stroke diagnosis is a good thing. It means stroke neurology, as a field, has (for the most part) adequate randomized control trials to make evidence-based decisions AND has easy-to-use and reliable diagnostic tools (neuro and vascular imaging). as long as the user has the necessary data to input and some basic knowledge of neurology, they have a good chance of figuring out the stroke type and thus guide treatment. the same cannot be said for all neurological subspecialties. 

Is this a bad thing? Are we becoming automatons rather than critical thinkers? As with most technologies, there is the risk of falling into that behavior. The good thing about it is, it is proof that we have standards of practice that come from evidenced-based medicine that anyone can use anywhere on any patient. It levels the playing field by providing a reliable diagnostic tool to anyone with an internet connection.

With the advent of evidence-based medicine and algorithmic protocols, algorithms will likely be created for other diseases. These algorithms can then be converted into online tools and eventually virtual (or physical) robot doctors. For those interested, check out this podcast, Will computers replace doctors?

Robots aside, does this mean the vascular neurologist = internist + online tool or ER physician + online tool? although the tool can give an etiology, treatment choice still depends on several things the tool doesn’t address. This includes patient preference and provider experience with direct acting oral anticoagulants, choice of immunosuppressant for things like CNS vasculitis, and consideration of patient comorbidities in prescribing some treatments. a vascular neurologist would be more experienced with such literature and decision making.

Some takeaway clinical pearls from this study: patients over age 50 are 4 times more likely to have cardioembolic stroke and 6 times less likely to have stroke from other uncommon causes compared to those under age 50. The incidence of large artery athero peaked between ages 50-70, but the incidence of cardioembolic stroke as the likely etiology continuously increased with age. the chance of meeting criteria for “Evident” as a level of confidence for stroke subtype was between 40-50% across all stroke etiologies. limitations of the study include a population that was around 80% white and no use of extended cardiac monitoring.

How does this tool change current practice? I believe it is best used primarily in 2 situations: training purposes for residents/fellows and when you’re stumped about the diagnosis. going down the list of stroke etiologies and diagnostic tools reminds you of all the tools you have in your belt and makes you think of things that may not have been on your initial differential. when someone has a flame shaped ICA dissection on CTA and ipsilateral watershed infarcts, filling out the checkboxes in the tool seems unnecessary and cumbersome.