| 摘要: | 中風是導致全球死亡和殘疾的主因之一。中風對患者及其家人產生廣泛的負面身體和經濟影響。其中，缺血性中風佔總中風的百分之八十。當前，通過靜脈內注射血栓溶解劑進行溶栓治療是急性缺血性中風的最有效的治療方法。早期接受溶栓治療可得到更高的生存率和更好的預後。 隨著近年醫學治療的進展，使靜脈內注射血栓溶解劑之後的血管內治療也成為中風患者的常規治療方法。然而，儘管治療技術取得了這些進步，但由於狹窄的治療時間窗口，大多數患者仍未接受靜脈血栓溶解劑或血管內治療。另外，儘管血栓溶解劑的優點很明顯，但個別患者接受治療的效果卻有所不同。只有一部分接受治療的患者的神經系統功能得到了改善，也只有一部分患者得到良好的長期預後。治療後產生症狀性腦出血的風險也削弱了治療的預期益處。溶栓後症狀性腦出血暗示臨床預後較差，導致早期死亡及增加三個月後的死亡率。對急性中風的患者的治療效果進行精準及個人化的預測極具挑戰性，但對於協助臨床醫生進行後續治療策略而言是必要的。
現今已經開發了各種計分系統來預測急性中風後的預後和治療結果。在不同的評分系統中，針對患者的特性，使用了各種不同預測模型，但是，這些傳統預測模型的臨床應用通常受到計分方法的複雜性或預測模型的中等程度準確性的限制。我們的研究使用基於機器學習的演算法來開發預測模型，以預測接受不同治療和不同年齡層的急性中風患者的短期和長期預後、發生併發症的風險及接受不同治療的反應。針對2009年至2018年之間的331名急性缺血性中風接受血栓溶解劑治療的患者，利用相關的八項臨床指標生成預測模型。並使用五倍交叉驗證(five-fold cross-validation)對模型的可推論性(generalizability)進行驗證。每個模型的正確性根據驗證組的準確性、精確度、敏感性，特異性和ROC曲線下面積（AUC）進行評估。模型經過適當的訓練後，預測症狀性腦出血的預測模型之AUC為0.941，準確性、敏感性、特異性分別為91.0％，85.7％和92.5％。預測三個月死亡率模型之的AUC為0.976，準確性、敏感性和特異性分別為95.2％，94.4％和95.5％。另外，針對196位急性缺血性中風患者，利用類神經網路建立之預測接受血栓溶解劑後24小時的主要神經功能改善（major neurologic improvement, MNI）及三個月後的長期功能恢復預測模型，顯示與24小時的主要神經功能改善相關的參數為血壓、心率、血糖、意識等級、美國國立衛生研究院中風量表（NIHSS）分數及是否有糖尿病史。與三個月功能預後相關的預測因子為年齡、意識水平、血壓、血糖水平、糖化血色素指數、糖尿病病史、中風類型和NIHSS分數。經過交叉驗證後。預測24小時主要神經功能改善的類神經網路模型可達到0.944的AUC。準確性，敏感性和特異性分別為94.6％，89.8％和95.9％。預測三個月功能預後的類神經網路模型之AUC為0.933，準確性，敏感性和特異性分別為88.8％，94.7％和86.5％。本研究發展的類人工神經網絡的模型對接受靜脈血栓溶解治療後患者的的短期主要神經功能改善、三個月的神經功能症狀、症狀性腦出血的併發症及和三個月的死亡率具有很高的預測性能和可靠性。生成的模型還可以準確預測不同年齡及接受不同急性中風治療策略的效果和預後；因此，本研究在臨床上將有助於急性中風患者的治療決策。
Stroke accounts for substantial premature death and disability worldwide. Consequently, stroke has a wide-ranging negative physical and economic impact on patients and their families. Ischemic stroke accounts for 80% of total stroke. Currently, application of thrombolysis with intravenous administration recombinant tissue plasminogen activator (IV-tPA) is the most accessible and effective treatment for acute ischemic stroke (AIS). Early thrombolytic treatment leads to a higher survival rate and more favorable outcomes. Advances in AIS therapy have made endovascular therapy (EVT) following IV-tPA a routine management for patients with AIS. However, despite these advances in therapeutic techniques, majority of patients with AIS do not receive IV-tPA or EVT because of the narrow treatment time windows. In addition, although the advantages of tPA are clear, the individual treatment response varies. Only a part of treated patients experiences major neurologic improvement (MNI) and have a favorable long-term outcome. The intended benefits of tPA are also dampened by the risk of symptomatic intracerebral hemorrhage (sICH). Post-thrombolysis sICH herald poor clinical outcome and account for most early excess deaths, with a high 3-month mortality rate. Individualized prediction of the outcome of AIS is challenging but necessary to assist the clinicians in conducting subsequent treatment strategies.
Various scoring systems to predict the prognosis and therapy outcome after AIS have been developed. In different scoring systems, depending on the cohort peculiarity, various predictors have been used, however, clinical application of these prediction models is often limited by the complexity of the scoring methods or the mild-moderate accuracy of the prediction models. Current study used machine learning-based algorithms to develop prediction models to estimate the short-term and long-term outcomes, complications, and treatment responses in different treatments and aged patient groups with AIS. We
developed artificial neural network (ANN)-based models after evaluating the predictive value of pre-treatment parameters associated with sICH and mortality in a cohort of 331 patients between 2009 and 2018. The ANN models were generated using eight clinical inputs and two outputs. After adequate training, the AUC of the ANN model for predicting sICH was 0.941, with accuracy, sensitivity, and specificity of 91.0%, 85.7%, and 92.5%, respectively. The AUC for predicting 3-month mortality was 0.976, with accuracy, sensitivity, and specificity of 95.2%, 94.4%, and 95.5%, respectively. The ANN model to predict MNI achieved an AUC of 0.944, with the accuracy, sensitivity and specificity of 94.6%, 89.8% and 95.9%, respectively. The AUC for the model to predict the three-month functional outcome was 0.933, with the accuracy, sensitivity and specificity of 88.8%, 94.7% and 86.5%, respectively. The generated ANN models exhibited high predictive performance and reliability for predicting major neurological improvement, 3-months functional outcomes, sICH and 3-month mortality after thrombolysis. The ANN models also accurately predicted the outcomes and responses to different treatments for AIS; thus, their proposed clinical application to aid outcome prediction and decision-making for the patients with AIS. |
