Da Binlin, Chen Huan, Wu Wei, Guo Wuhua, Zhou Anru, Yin Qin, Gao Jun, Chen Junhui, Xiao Jiangqiang, Wang Lei, Zhang Ming, Zhuge Yuzheng, Zhang Feng
Department of Gastroenterology, Nanjing Drum Tower Hospital, Affiliated Hospital of Medical School, Nanjing University, Nanjing, Jiangsu, China.
Research Institute of General Surgery, Jinling Hospital, Affiliated Hospital of Medical School, Nanjing University, Nanjing, Jiangsu, China.
EClinicalMedicine. 2024 Dec 16;79:103001. doi: 10.1016/j.eclinm.2024.103001. eCollection 2025 Jan.
Although numerous prognostic scores have been developed for patients with cirrhosis after Transjugular intrahepatic portosystemic shunt (TIPS) placement over years, an accurate machine learning (ML)-based model remains unavailable. The aim of this study was to develop and validate a ML-based prognostic model to predict survival in patients with cirrhosis after TIPS placement.
In this retrospective study in China, patients diagnosed with cirrhosis after TIPS placement from 2014 to 2020 in our cohort were included to develop a ML-based model. Patients from the other two tertiary hospitals between 2016 and 2022 were as external validation cohort. The random forest (RF) model was built using 7 selected features via the least absolute shrinkage and selection operator (LASSO) regression, and subsequent 10-fold cross-validation was performed.
A total of 400 patients in our cohort were included (median age and interquartile range, 59 (50, 66); 240 men). Two hundred and eighty patients made up the training set and 120 were in the testing set, and 346 patients were included in the external validation cohort. Seven attributes were selected: Na, ammonia (Amm), total bilirubin (Tb), albumin (Alb), age, creatinine (Cr), and ascites. These parameters were included in a new score named the RF model. The accuracy, precision, recall, and F1 Score of the RF model were 0.84 (95% CI: 0.76, 0.91), 0.84 (95% CI: 0.77, 0.91), 0.99 (95% CI: 0.95, 1.00), 0.91 (95% CI: 0.81, 0.10) in the testing set, and 0.88 (95% CI: 0.84, 0.91), 0.89 (95% CI: 0.85, 0.92), 0.99 (95% CI: 0.97, 1.00), 0.93 (95% CI: 0.85, 0.97) in the validation cohort, respectively. The calibration curve showed a slope of 0.875 in the testing set and a slope of 0.778 in the external validation cohort, suggesting well calibration performance. The RF model outperformed other scoring systems, such as the (Child-Turcotte-Pugh score) CTP, (model for end-stage liver disease) MELD, (sodium MELD) MELD-Na, (Freiburg index of post-TIPS survival) FIPS and (Albumin-Bilirubin) ALBI, showing the highest (area under the curve) AUC of 0.82 (95% CI: 0.72, 0.91) and 0.7 (95% CI: 0.60, 0.79) in predicting 1-year survival across the testing set and external validation cohort.
This study developed a RF model that better predicted 1-year survival for patients with cirrhosis after TIPS placement than the other scores.
National Natural Science Foundation of China (grant numbers 81900552 and 82370628).
多年来，尽管已经为经颈静脉肝内门体分流术（TIPS）置入术后的肝硬化患者开发了许多预后评分系统，但基于机器学习（ML）的准确模型仍然不可用。本研究的目的是开发并验证一种基于ML的预后模型，以预测TIPS置入术后肝硬化患者的生存率。
在这项中国的回顾性研究中，纳入了2014年至2020年在我们队列中诊断为TIPS置入术后肝硬化的患者，以开发基于ML的模型。将另外两家三级医院2016年至2022年的患者作为外部验证队列。通过最小绝对收缩和选择算子（LASSO）回归，使用7个选定特征构建随机森林（RF）模型，并随后进行10倍交叉验证。
我们队列中总共纳入了400例患者（中位年龄和四分位间距，59（50，66）；240例男性）。280例患者组成训练集，120例在测试集，346例患者纳入外部验证队列。选择了7个属性：钠、氨（Amm）、总胆红素（Tb）、白蛋白（Alb）、年龄、肌酐（Cr）和腹水。这些参数被纳入一个名为RF模型的新评分中。RF模型在测试集的准确率、精确率、召回率和F1分数分别为0.84（95%CI：0.76，0.91）、0.84（95%CI：0.77，0.91）、0.99（95%CI：0.95，1.00）、0.91（95%CI：0.81，0.10），在验证队列中分别为0.88（95%CI：0.84，0.91）、0.89（95%CI：0.85，0.92）、0.99（95%CI：0.97，1.00）、0.93（95%CI：0.85，0.97）。校准曲线在测试集的斜率为0.875，在外部验证队列的斜率为0.778，表明校准性能良好。RF模型优于其他评分系统，如（Child-Turcotte-Pugh评分）CTP、（终末期肝病模型）MELD、（钠MELD）MELD-Na、（TIPS术后生存的弗莱堡指数）FIPS和（白蛋白-胆红素）ALBI，在预测测试集和外部验证队列的1年生存率时，显示出最高的曲线下面积（AUC），分别为0.82（95%CI：0.72，0.91）和0.7（95%CI：0.60，0.79）。
本研究开发的RF模型在预测TIPS置入术后肝硬化患者的1年生存率方面比其他评分系统表现更好。
中国国家自然科学基金（批准号81900552和82370628）。
