AI models can be trained to predict outcomes for patients receiving thrombolysis for acute ischemic stroke (AIS), according to a new study published in the Journal of the Neurological Sciences.
“Thrombolysis with intravenous administration of recombinant tissue plasminogen activator is the most accessible and effective treatment for AIS,” wrote lead author Chen-Chih Chung, Tapei Medical University, and colleagues. “Early thrombolytic treatment leads to a higher survival rate and more favorable outcomes.”
However, the researchers added, some patients don’t respond as well to thrombolysis, and predicting those outcomes is a “challenging but necessary” part of the treatment process. Chung and colleagues aimed to design “reliable models” for the prediction of major neurologic improvement (MNI) within 24 hours of thrombolysis as well as long-term patient outcomes.
To develop their artificial neural network (ANN)-based prediction models, the researchers explored data from 196 AIS patients who were treated with intravenous thrombolysis from 2009 to 2017 at a single facility. One model (ANN Model 1) was trained to predict MNI 24 hours following treatment. Another model (ANN Model 2) was trained to predict outcomes three months following treatment.
Overall, “after adequate training,” ANN Model 1 achieved an area under the ROC curve (AUC) of 0.944, accuracy of 94.6%, sensitivity of 89.8% and specificity of 95.9%. ANN Model 2, meanwhile, achieved an AUC of 0.933, accuracy of 88.8%, sensitivity of 94.7% and specificity of 86.5%.
“These models may have clinical value in assisting decision-making,” the authors wrote. “Further research is must be conducted on their predictive value and diagnostic accuracy while taking into account invasive adjuvant strategies.”
In addition, blood pressure (BP), heart rate, glucose level, consciousness level, National Institutes of Health Stroke Scale (NIHSS) score and a history of diabetes mellitus (DM) were all determined to be associated with MNI. Age, glucose level, BP, hemoglobin A1c, history of DM, stroke subtype and NIHSS score were all “significant factors” affecting long-term stroke outcomes.
BP was found to be especially crucial in both prediction models, with both low and high BP appearing to impact patients’ short-term and long-term outcomes after thrombolysis.
“Patients with lower BP during the acute phase of AIS are associated with brain injury and poor outcome” they wrote. “A higher initial BP may imply a later reduction in BP in response to thrombolysis and a better neurological improvement. The dynamic changes in BP during AIS are associated with impaired cerebral autoregulation, reperfusion injury, edema, and hemorrhagic transformation.”