Background and Significant of the study: Stroke is a second leading cause of death and a leading cause of adult disability worldwide with occurrences of 400-800 strokes per 100,000. In view of the disabilities and long-term health consequences occurred as a result of ischemic stroke the need for an accurate prediction of functional independence is paramount for setting therapeutic goals, implementing home adjustments and community support tailored to patients’ needs, and informing patients about their prospects and prognosis. Accurate and efficient prediction of functional outcome of patients with acute ischemic stroke has the potential to enhance clinical care as well as improve the quality of stroke research. The purpose of this study was to develop a random forest model to predict the functional outcome of patients diagnosed with acute ischemic stroke at discharge from hospital.
Materials and Methods: We conducted a retrospective study on 6,524 patients extracted from Taiwan Stroke Registry (TSR). Patients were categorized into two groups: those who receive tPA (n=1,632) and non-tPA (n=4,892). We developed two Random Forest models to predict the functional outcome in both groups. The outcome of patients at discharge was measured using Modified Rankin Scale (mRS). The mRS was dichotomized into good outcome (mRS 0-2) and poor outcome (mRS 3-5) to provide a binary outcome measure. An “outcome” was defined as patient’s functional independent at discharge.
Results: A total of 6,524 ischemic stroke patients were enrolled in this study. In tPA group, there were 39% patients with good outcome (431 males and 212 females) while 61% had poor outcome (608 males and 381 females). The overall mean age of patients who had tPA was 64.5 (SD ± 11.4). The average age of patients with good and poor outcome were (62.8 ± 11.7 vs 65.6 ± 11.1) respectively. In non-tPA group, the overall mean age of patients was 66.0 ± 10.9, the mean age of patients with good and poor outcome were 62.7 ± 11.7 and 67.0 ± 10.5 respectively. However, 23% of patients in non-tPA group had good outcome while 77% had poor outcome. The accuracy of our model in tPA group was 68.8% with an AUC of 71% and the accuracy in non-tPA group was 76.9% with an AUC of 74%
Conclusion: RF was shown to predict the functional outcome of patients diagnosed with acute ischemic stroke. However, using RF to predict the functional outcome of ischemic stroke may be potentially beneficial in implementing early preventive and therapeutic measures. NIHSS scores at arrival were identified as the most informative variable associated with the functional outcome of patients who had tPA and non-tPA at discharge.
