Outcome Prediction of Endovascular Treatment in Acute Stroke using Novel Machine-Learning Approaches

Dr. med. Fanny Quandt, Department of Neurology, University Medical Center Hamburg–Eppendorf
Prof. Dr. med. Götz Thomalla, Department of Neurology, University Medical Center Hamburg–Eppendorf

Prediction of clinical outcome after endovascular thrombectomy (ET) may facilitate superior patient selection and help decide which patients should be transferred to a center able to perform ET. Moreover, post-interventional decisions whether to pursue maximal treatment or de-escalation is often influenced by the expected long-term outcome. Attempts have been made to establish pre-interventional outcome scores, such as the PRE and HIAT2 score, after ET in large vessel occlusion stroke (Rangaraju et al., 2015; Sarraj et al., 2013). Scores, however, show a ceiling effect in prognostic accuracy with an area under the curve of 0.8 at most. Also incorporating post-interventional variables have been shown to enhance outcome prediction (Rangaraju et al., 2014). Prognostic scores are usually informed by univariate analysis comprising of a relatively simple algorithm. The use of machine learning algorithms and of pattern recognition with deep learning in particular has developed immensely over the past years and is starting to gain currency in medicine. Deep learning holds the advantage that it finds non-linear patterns in multidimensional data and does not require prior dimensionality reduction of the data. First studies have investigated outcome prediction using machine learning approaches in patients undergoing ET after stroke (Alawieh et al., 2019; Nishi et al., 2019; van Os et al., 2018). Machine learning algorithms, however, benefit from a large data set, hence we expect a great advantage from analyzing data from multiple trials.

Novel machine-learning algorithms can predict long-term outcome of endovascular treatment in acute stroke with high accuracy.

It is the aim to predict, on a single patient basis, whether a patient will have a good outcome after endovascular treatment, defined as modified Rankin Scale 0-2 at 90 days. First, we would predict outcome based on baseline data prior to ET, secondly, we would analyze how inclusion of post-treatment data up to the time of hospital discharge would enhance long-term outcome prediction. In order to do so, we employ novel machine learning approaches. Classification will be performed with established Support Vector Machines but also with deep learning neural networks as implemented in the open-source TensorFlow library developed by the Google Brain Team. As prediction is data-driven, all available data of patients with ET without prior feature selection would be included. For example, pre-treatment data comprises e.g. patients’ characteristics, medical history, occluded vessel, ASPECTS, prior drug use, as well as time metrics up to the point of intervention. Post-interventional data would comprise, e.g. NIHSS at 24 h, mTICI grade and intracranial hemorrhage. Raw imaging data is not required. Prediction accuracy will be calculated using a cross-validation and will be compared to the performance of already published outcome scores. Post-hoc analysis will explore which features contribute the most to achieve a good prediction accuracy.

Funding will be provided through the Clinical Stroke and Imaging Research Laboratory by Götz Thomalla, at the Department of Neurology, University Medical Center Hamburg-Eppendorf.

Alawieh, A., Zaraket, F., Alawieh, M.B., Chatterjee, A.R., Spiotta, A., 2019. Using machine learning to optimize selection of elderly patients for endovascular thrombectomy. J. NeuroInterventional Surg. 11, 847–851. https://doi.org/10.1136/neurintsurg-2018-014381

Nishi, H., Oishi, N., Ishii, A., Ono, I., Ogura, T., Sunohara, T., Chihara, H., Fukumitsu, R., Okawa, M., Yamana, N., Imamura, H., Sadamasa, N., Hatano, T., Nakahara, I., Sakai, N., Miyamoto, S., 2019. Predicting Clinical Outcomes of Large Vessel Occlusion Before Mechanical Thrombectomy Using Machine Learning. Stroke 50, 2379–2388. https://doi.org/10.1161/STROKEAHA.119.025411

Rangaraju, S., Aghaebrahim, A., Streib, C., Sun, C.-H., Ribo, M., Muchada, M., Nogueira, R., Frankel, M., Gupta, R., Jadhav, A., Jovin, T.G., 2015. Pittsburgh Response to Endovascular therapy (PRE) score: optimizing patient selection for endovascular therapy for large vessel occlusion strokes. J. NeuroInterventional Surg. 7, 783–788. https://doi.org/10.1136/neurintsurg-2014-011351

Rangaraju, S., Liggins, J.T.P., Aghaebrahim, A., Streib, C., Sun, C.-H., Gupta, R., Nogueira, R., Frankel, M., Mlynash, M., Lansberg, M., Albers, G., Jadhav, A., Jovin, T.G., 2014. Pittsburgh Outcomes After Stroke Thrombectomy Score Predicts Outcomes After Endovascular Therapy for Anterior Circulation Large Vessel Occlusions. Stroke 45, 2298–2304. https://doi.org/10.1161/STROKEAHA.114.005595

Sarraj, A., Albright, K., Barreto, A.D., Boehme, A.K., Sitton, C.W., Choi, J., Lutzker, S.L., Sun, C.-H.J., Bibars, W., Nguyen, C.B., Mir, O., Vahidy, F., Wu, T.-C., Lopez, G.A., Gonzales, N.R., Edgell, R., Martin-Schild, S., Hallevi, H., Chen, P.R., Dannenbaum, M., Saver, J.L., Liebeskind, D.S., Nogueira, R.G., Gupta, R., Grotta, J.C., Savitz, S.I., 2013. Optimizing Prediction Scores for Poor Outcome After Intra-Arterial Therapy in Anterior Circulation Acute Ischemic Stroke. Stroke 44, 3324–3330. https://doi.org/10.1161/STROKEAHA.113.001050

van Os, H.J.A., Ramos, L.A., Hilbert, A., van Leeuwen, M., van Walderveen, M.A.A., Kruyt, N.D., Dippel, D.W.J., Steyerberg, E.W., van der Schaaf, I.C., Lingsma, H.F., Schonewille, W.J., Majoie, C.B.L.M., Olabarriaga, S.D., Zwinderman, K.H., Venema, E., Marquering, H.A., Wermer, M.J.H., Investigators, the M.C.R., 2018. Predicting Outcome of Endovascular Treatment for Acute Ischemic Stroke: Potential Value of Machine Learning Algorithms. Front. Neurol. 9. https://doi.org/10.3389/fneur.2018.00784