人工神经网络预测腹腔镜袖状胃切除术后30天并发症与死亡
本研究基于2016年MBSAQIP全国数据库101,721例腹腔镜袖状胃切除术患者,筛选与30天复合终点(再入院、再手术、再干预或死亡)显著相关的术前因素,构建反向传播人工神经网络并与逻辑回归比较。神经网络在训练集和验证集中的AUROC分别为0.581和0.585,均优于逻辑回归的0.572。
要点
- 年龄、非白人种族、较高初始BMI、重度高血压、糖尿病、非独立功能状态及既往上消化道/减重手术与30天并发症相关。
- 神经网络模型预测30天复合终点的AUROC在验证集为0.585,略高于逻辑回归的0.572。
核心结论
- 简单的术前患者因素可提示腹腔镜袖状胃切除术后30天并发症风险增加。
- 人工神经网络模型对30天复合终点的预测优于标准逻辑回归模型。
研究设计与样本
- 回顾性队列研究,数据来自2016年MBSAQIP全国数据库,纳入101,721例腹腔镜袖状胃切除术患者。
- 通过双变量和多变量名义逻辑回归筛选与30天复合终点显著相关的术前因素。
- 将显著因素纳入反向传播人工神经网络,采用k折内部验证并与逻辑回归比较AUROC。
主要发现
- 双变量分析中,年龄较大(P=0.03)、非白人种族、较高初始BMI、重度高血压、糖尿病、非独立功能状态及既往上消化道/减重手术(均P<0.001)与30天并发症相关。
- 名义逻辑回归分析(n=100,791)中上述因素仍显著(P<0.001,r²=0.008,AUROC=0.572)。
- 神经网络训练集(n=80,633)AUROC=0.581(r²=0.011),验证集(n=20,158)AUROC=0.585(r²=0.012),均高于逻辑回归。
临床怎么用
- 可利用年龄、种族、BMI、高血压、糖尿病、功能状态及手术史等简单术前因素评估袖状胃切除术后30天并发症风险。
- 神经网络模型预测效能略优于逻辑回归,但区分度有限,不宜单独用于临床决策。
局限性
- 原文未明确报告局限性。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
BACKGROUND: Multiple patient factors may convey increased risk of 30-day morbidity and mortality after laparoscopic vertical sleeve gastrectomy (LVSG). Assessing the likelihood of short-term morbidity is useful for both the bariatric surgeon and patient. Artificial neural networks (ANN) are computational algorithms that use pattern recognition to predict outcomes, providing a potentially more accurate and dynamic model relative to traditional multiple regression. Using a comprehensive national database, this study aims to use an ANN to optimize the prediction of the composite endpoint of 30-day readmission, reoperation, reintervention, or mortality, after LVSG. METHODS: A cohort of 101,721 LVSG patients was considered for analysis from the 2016 Metabolic and Bariatric Surgery Accreditation and Quality Improvement Program national dataset. Select patient factors were chosen a priori as simple, pertinent and easily obtainable, and their association with the 30-day endpoint was assessed. Those factors with a significant association on both bivariate and multivariate nominal logistic regression analysis were incorporated into a back-propagation ANN with three nodes each assigned a training value of 0.333, with k-fold internal validation. Logistic regression and ANN models were compared using area under receiver-operating characteristic curves (AUROC). RESULTS: Upon bivariate analysis, factors associated with 30-day complications were older age (P = 0.03), non-white race, higher initial body mass index, severe hypertension, diabetes mellitus, non-independent functional status, and previous foregut/bariatric surgery (all P < 0.001). These factors remained significant upon nominal logistic regression analysis (n = 100,791, P < 0.001, r2= 0.008, AUROC = 0.572). Upon ANN analysis, the training set (80% of patients) was more accurate than logistic regression (n = 80,633, r2= 0.011, AUROC = 0.581), and it was confirmed by the validation set (n = 20,158, r2= 0.012, AUROC = 0.585). CONCLUSIONS: This study identifies a panel of simple and easily obtainable preoperative patient factors that may portend increased morbidity after LSG. Using an ANN model, prediction of these events can be optimized relative to standard logistic regression modeling.
引用
- Wise ES, Amateau SK, Ikramuddin S, et al. Prediction of thirty-day morbidity and mortality after laparoscopic sleeve gastrectomy: data from an artificial neural network. Surg Endosc. 2019. PMID: 31571034. doi:10.1007/s00464-019-07130-0
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。