EVALUATION OF THE USE OF ARTIFICIAL INTELLIGENCE AS A PRE-HOSPITAL TOOL FOR ACUTE STROKE DIAGNOSIS (ID 1060)
Abstract
Background And Aims
The rising pressure on emergency medical services, highlighted throughout the pandemic, has revealed a need for alternative methods of pre-hospital emergency care. We aim to evaluate the digital symptom triage application (app) Mediktor in the acute stroke diagnosis.
Methods
A retrospective cohort study of consecutive patients with a suspected acute stroke attended at the emergency room (ER) of a tertiary hospital by a neurologist between January 1st – December31, 2020, were evaluated. Retrospectively, another neurologist completed the digital symptom Mediktor app questionnaire according to the data of the emergency medical record, and demographic and clinical data were analyzed.
Results
150 patient cases were analyzed, mean age 70.87 ± 14.5 years; 50.7% women. In the ER, 134 cases (89.35%) were given the final diagnosis of stroke and 16 cases (10.7%) were labelled as stroke mimics. Mediktor had an accuracy of 75% in the main diagnosis (sensitivity: 77%; specificity: 56%). An adjusted regression model pointed at stroke severity assessed with the NIHSS score (OR = 0.85; 95% CI, 0.75-0.95) as an independent predictor of stroke diagnostic accuracy. After adjusting for other predictors, no significant differences were found in relation to anterior circulation stroke, the presence of vascular risk factors or previous history of stroke.
Conclusions
These findings suggest that digital symptom triage apps like Mediktor, while outperformed by a neurologist, can be a complementary tool in the emergent stroke screening. The development of these technologies and their implementation among patients and families can improve stroke diagnosis and consequently acute stroke care.
Trial Registration Number
Not applicable
lesiondetectionwithaconvolutionaldeeplearningmodel_poster/thumbnails/0.png)
