Applying machine learning to a regional STEMI dataset: employing neural networks to unravel new patterns to explain etiopathology of stroke after aspiration thrombectomy.

Background:
Widespread adoption of evidence‐based guidelines and treatment pathways in STEMI patients during the last two decades has considerably improved survival and decreased the risk of recurrent MI, however, there is a plateauing in outcomes survival since 2008. We aim to employ neural networks produced by machine learning to potentially unravel new patterns that will explain etiopathology of stroke after aspiration thrombectomy.

Methods:
The Minnesota Heart Institute Foundation (MHIF) STEMI database is a large prospective regional STEMI registry consisting of 180 variables recorded as binary semi‐quantitative data of over 5000 patients encompassing over 15 years. Where the initial proof‐of‐concept study mostly covered a broad swath of 30+ pre‐admission variables in a holistic, all‐encompassing representation of the STEMI dataset, the plan in the ongoing project is to pursue more targeted pattern detection. Specifically, we will be looking for variables that are related to the occurrence of cerebrovascular accidents (CVAs) after aspiration thrombectomy (AT). Following a division of STEMI patients into CVA and non‐CVA groups, two main analytical routes will be pursued in parallel.

Results:
Combining machine learning and advanced data visualizations produced for STEMI data. Figure 1

Conclusions:
Proof of concept achieved by first ever successful mapping of stemi data utilizing geographic principles, machine learning and advanced data visualization techniques. Further advanced data mining will be applied to understand etiopathology of stroke after aspiration thrombectomy.