Pan Tianyue, Jiang Xiaolang, Liu Hao, Liu Yifan, Fu Weiguo, Dong Zhihui
Department of Vascular Surgery, Institute of Vascular Surgery, Zhongshan Hospital, Fudan University, Shanghai, China.
National Clinical Research Center for Interventional Medicine, Shanghai, China.
Front Cardiovasc Med. 2022 Feb 9;9:783336. doi: 10.3389/fcvm.2022.783336. eCollection 2022.
The current scoring systems could not predict prognosis after endovascular therapy for peripheral artery disease. Machine learning could make predictions for future events by learning a specific pattern from existing data. This study aimed to demonstrate machine learning could make an accurate prediction for 2-year major adverse limb event-free survival (MFS) after percutaneous transluminal angioplasty (PTA) and stenting for lower limb atherosclerosis obliterans (ASO).
A lower limb ASO cohort of 392 patients who received PTA and stenting was split to the training set and test set by 4:1 in chronological order. Demographic, medical, and imaging data were used to build machine learning models to predict 2-year MFS. The discrimination and calibration of artificial neural network (ANN) and random forest models were compared with the logistic regression model, using the area under the receiver operating curve (ROCAUC) with DeLong test, and the calibration curve with Hosmer-Lemeshow goodness-of-fit test, respectively.
The ANN model (ROCAUC = 0.80, 95% CI: 0.68-0.89) but not the random forest model (ROCAUC = 0.78, 95% CI: 0.66-0.87) significantly outperformed the logistic regression model (ROCAUC = 0.73, 95% CI: 0.60-0.83, = 0.01 and = 0.24). The ANN model the logistic regression model demonstrated good calibration performance ( = 0.73 and = 0.28), while the random forest model showed poor calibration ( < 0.01). The calibration curve of the ANN model was visually the closest to the perfectly calibrated line.
Machine learning models could accurately predict 2-year MFS after PTA and stenting for lower limb ASO, in which the ANN model had better discrimination and calibration. Machine learning-derived prediction tools might be clinically useful to automatically identify candidates for PTA and stenting.
当前的评分系统无法预测外周动脉疾病血管内治疗后的预后。机器学习可以通过从现有数据中学习特定模式来预测未来事件。本研究旨在证明机器学习能够准确预测下肢动脉粥样硬化闭塞症（ASO）患者经皮腔内血管成形术（PTA）和支架置入术后2年的无主要肢体不良事件生存率（MFS）。
将392例接受PTA和支架置入术的下肢ASO患者队列按时间顺序以4:1的比例分为训练集和测试集。利用人口统计学、医学和影像学数据构建机器学习模型，以预测2年MFS。分别使用接受者操作特征曲线下面积（ROCAUC）的DeLong检验和校准曲线的Hosmer-Lemeshow拟合优度检验，将人工神经网络（ANN）和随机森林模型的区分度和校准度与逻辑回归模型进行比较。
ANN模型（ROCAUC = 0.80，95%CI：0.68 - 0.89）显著优于逻辑回归模型（ROCAUC = 0.73，95%CI：0.60 - 0.83，P = 0.01），而随机森林模型（ROCAUC = 0.78，95%CI：0.66 - 0.87）未显著优于逻辑回归模型（P = 0.24）。ANN模型和逻辑回归模型表现出良好的校准性能（P = 0.73和P = 0.28），而随机森林模型校准效果较差（P < 0.01）。ANN模型的校准曲线在视觉上最接近完美校准线。
机器学习模型能够准确预测下肢ASO患者PTA和支架置入术后2年的MFS，其中ANN模型具有更好的区分度和校准度。机器学习衍生的预测工具可能在临床上有助于自动识别PTA和支架置入术的候选患者。
