Chen Miao, Zhou Liya, Fan Limin, Li Yanhong, Wang Hairong
Department of Emergency, The First Affiliated Hospital of Hainan Medical University Haikou 570102, Hainan, China.
Department of Emergency, Xinhua Hospital Affiliated to Shanghai Jiaotong University, School of Medicine Shanghai 200092, China.
Am J Transl Res. 2023 Jul 15;15(7):4735-4745. eCollection 2023.
Through the comparison of different prediction models, we hope to find a promising statistical method to evaluate the prognosis of patients with acute ischemic stroke (AIS) after thrombolytic therapy.
Data of 518 patients who received thrombolytic therapy were retrospectively collected in this study. Among them, 362 patients met the eligibility criteria, so their data such as age, sex, smoking history, previous medical history, clinical and laboratory indicators were analyzed. According to the 3 month follow-up results, 266 patients were included in a good prognosis group (modified Rankin Scale (mRS) score ≤2) and 96 in a poor prognosis group (3≤mRS≤6). All variables with P<0.05 in univariant and multivariant analyses were assigned in logistic regression model and artificial neural network (ANN) model to predict neurological prognosis, and the performance of the two models were compared.
Age, NIHSS scores, the serum concentration of immediate glucose, APTT and MBP at admission were found to be the predictive factors through ANN and logistic regression analysis. The binary logistic regression model revealed that the percentage correction, the precision, recall and F1 score of the regression model were 79%, 69.23%, 37.50% and 48.65%, respectively. While those of ANN were 79.98%, 69.70%, 37.25%, and 49.66%, correspondingly.
ANN model is as effective as a logistic regression model in predicting the prognosis of AIS after thrombolytic therapy with rt-PA. Moreover, ANN is slightly superior to logistic regression in accuracy, precision and F1 score.
通过比较不同的预测模型，我们希望找到一种有前景的统计方法来评估急性缺血性卒中（AIS）患者溶栓治疗后的预后。
本研究回顾性收集了518例接受溶栓治疗患者的数据。其中，362例患者符合纳入标准，因此对他们的年龄、性别、吸烟史、既往病史、临床和实验室指标等数据进行了分析。根据3个月的随访结果，266例患者被纳入预后良好组（改良Rankin量表（mRS）评分≤2），96例被纳入预后不良组（3≤mRS≤6）。将单因素和多因素分析中P<0.05的所有变量纳入逻辑回归模型和人工神经网络（ANN）模型以预测神经功能预后，并比较两种模型的性能。
通过人工神经网络和逻辑回归分析发现，年龄、美国国立卫生研究院卒中量表（NIHSS）评分、入院时即刻血糖血清浓度、活化部分凝血活酶时间（APTT）和髓鞘碱性蛋白（MBP）是预测因素。二元逻辑回归模型显示，回归模型的校正百分比、精度、召回率和F1分数分别为79%、69.23%、37.50%和48.65%。而人工神经网络的相应指标分别为79.98%、69.70%、37.25%和49.66%。
在预测rt-PA溶栓治疗后AIS的预后方面，人工神经网络模型与逻辑回归模型一样有效。此外，人工神经网络在准确性、精度和F1分数方面略优于逻辑回归。
