Huo Zhenyu, Chong Feifei, Luo Siyu, Tong Ning, Lu Zongliang, Zhang Mengyuan, Liu Jie, Xu Hongxia, Li Na
Department of Clinical Nutrition, Daping Hospital, Army Medical University (Third Military Medical University), Chongqing, 400042, China.
Chongqing Municipal Health Commission Key Laboratory of Intelligent Clinical Nutrition and Transformation, Chongqing, 400042, China.
Support Care Cancer. 2024 Dec 16;33(1):31. doi: 10.1007/s00520-024-09060-7.
The purpose of this study is to investigate the relationship between preoperative cystatin C levels and the risk of serious postoperative complications in esophageal cancer (EC) patients, utilizing advanced machine learning (ML) methodologies.
We conducted an observational cohort study, involving 524 EC patients from December 2014 to July 2022. ML models, including logistic regression (LR) and multilayer perceptron (MLP), were applied to investigate the relationship between cystatin C and the serious postoperative complications. The predictive value of cystatin C was evaluated using receiver operating characteristic (ROC) analysis. Based on a restricted cubic spline (RCS) method, the potential nonlinear association was scrutinized.
The morbidity of serious postoperative complications was 8.78%. Bleeding volume, operating time, NRS2002 score, PONS score, and cystatin C were significantly associated with serious postoperative complications. The MLP model demonstrated superior predictive accuracy (AUC = 0.775, 95% CI: 0.701-0.849) compared to the LR model (AUC = 0.714, 95% CI: 0.630-0.798) and cystatin C alone (AUC = 0.612, 95% CI: 0.526-0.699). High cystatin C level independently predicted serious postoperative complications in EC patients. A positive and linear association was found between cystatin C and serious complications.
This research uncovers a notable correlation between cystatin C and the severe complications in EC patients after esophagectomy. Employing ML techniques offers a robust method for forecasting patient outcomes and emphasizes the potential of cystatin C as a predictive biomarker in medical practice.
本研究旨在利用先进的机器学习（ML）方法，探讨术前胱抑素C水平与食管癌（EC）患者术后严重并发症风险之间的关系。
我们进行了一项观察性队列研究，纳入了2014年12月至2022年7月期间的524例EC患者。应用包括逻辑回归（LR）和多层感知器（MLP）在内的ML模型，研究胱抑素C与术后严重并发症之间的关系。使用受试者工作特征（ROC）分析评估胱抑素C的预测价值。基于受限立方样条（RCS）方法，仔细研究了潜在的非线性关联。
术后严重并发症的发生率为8.78%。出血量、手术时间、NRS2002评分、PONS评分和胱抑素C与术后严重并发症显著相关。与LR模型（AUC = 0.714，95% CI：0.630 - 0.798）和单独的胱抑素C（AUC = 0.612，95% CI：0.526 - 0.699）相比，MLP模型显示出更高的预测准确性（AUC = 0.775，95% CI：0.701 - 0.849）。高胱抑素C水平独立预测EC患者术后严重并发症。胱抑素C与严重并发症之间存在正线性关联。
本研究揭示了胱抑素C与EC患者食管切除术后严重并发症之间的显著相关性。采用ML技术为预测患者预后提供了一种强大的方法，并强调了胱抑素C作为医学实践中预测生物标志物的潜力。
