ARTIFICIAL INTELLIGENCE AND DECISION-MAKING FOR VESTIBULAR SCHWANNOMA SURGERY.
ARTIFICIAL INTELLIGENCE AND DECISION-MAKING FOR VESTIBULAR SCHWANNOMA SURGERY.
复制标题
DOI：
10.1097/mao.0000000000003318
复制
发表时间：
2022-01-01
期刊：
影响因子：
--
通讯作者：
Abouzari M
中科院分区：
文献类型：
--
作者：
Risbud A;Tsutsumi K;Abouzari M
致编辑：我们怀着极大的兴趣阅读了Alkins等人题为“前庭神经鞘瘤手术后并发症的预测因素-一项基于人群的研究”的文章。在这项大型回顾性队列研究中，作者旨在确定可能与前庭神经鞘瘤（VS）手术后并发症相关的患者特征和合并症（1）。在多个机构治疗的1，456例患者中检查了人口统计学数据，术前合并症，手术方法以及短期和长期术后结局。他们的研究结果揭示了包括老年、糖尿病、痴呆和高血压在内的关键因素，这些因素可以预测再入院和心肌梗死等并发症。我们赞赏作者对这一人群数据的全面分析，并介绍了可能有助于VS患者术前计划和咨询的重要结果。鉴于我们小组在开发VS手术结局预测模型方面的经验，我们希望提供额外的见解，可能有助于解决作者确定的一些研究局限性。在最近的一项概念验证研究中，我们比较了逻辑回归模型与人工神经网络（ANN）在确定预测VS复发的患者报告因素方面的准确性（2）。在对698例VS患者的调查中，我们评估了患者的人口统计学资料、治疗后并发症和手术方法，以及肿瘤大小、症状、治疗中心和初始治疗后的年数。使用经过验证的分类算法，我们的数据集被分为训练，验证和测试子集，以评估ANN模型与逻辑回归相比的预测能力。总之，我们的ANN模型在正确分类病例和预测复发方面表现出上级性能，具有比标准回归模型更高的灵敏度（61 vs. 44%）和特异性（81 vs. 69%）ANN是机器学习的一种形式，是用于许多医学领域决策支持的新兴数学模型（3-5）。与传统统计方法相比，ANN的主要优点之一是它可以处理具有非线性分布的大型数据集。在VS患者的情况下，许多患者风险因素和术后事件本质上可能是多因素的，使得ANN特别适合作为治疗结果的预测工具（6）。利用国家数据库，我们小组最近应用了机器
To the Editor: We read with great interest the article entitled ‘‘Predictors of Postoperative Complications in Vestibular Schwannoma Surgery—A Population Based Study’’by Alkins et al. In this large retrospective cohort study, the authors aimed to identify the patient characteristics and comorbidities that may correlate with complications following surgery for vestibular schwannoma (VS)(1). Demographic data, preoperative comorbidities, surgical approach, and both short-and long-term postoperative outcomes were examined in 1,456 patients treated across multiple institutions. Their findings revealed key factors including older age, diabetes mellitus, dementia, and hypertension to be predictive of readmission and complications such as myocardial infarction. We applaud the authors for their comprehensive analysis of this population data and presentation of important results that may aid in preoperative planning and counseling for VS patients. Given our group’s experience in developing predictive models for VS surgical outcomes, we would like to offer additional insights that may help address some of the study limitations identified by the authors. In a recent proof-of-concept study, we compared the accuracy of logistic regression models to artificial neural networks (ANNs) in determining patient-reported factors that were predictive of VS recurrence (2). In a survey of 698 VS patients, we evaluated patients’ demographics, post-treatment complications, and surgical approach, in addition to tumor size, presenting symptoms, treatment centers, and years since initial treatment. Using validated classification algorithms, our dataset was divided into training, validation, and test subsets to assess the predictive power of ANN models compared to logistic regression. In summary, our ANN models demonstrated superior performance in correctly classifying cases and predicting recurrence, with a higher sensitivity (61 vs. 44%) and specificity (81 vs. 69%) than the standard regression model.ANN is a form of machine learning and emerging mathematic model used for decision support in many fields of medicine (3–5). One of the main advantages of ANN compared to traditional statistical methods is its handling of large datasets with nonlinear distributions. In the case of VS patients, many of the patient risk factors and postsurgical events are likely multifactorial in nature, making ANN particularly suitable as a predictive tool for treatment outcomes (6). Using a national database, our group recently applied machine