Department of Anesthesiology, Critical Care & Pain Management, Hospital for Special Surgery, New York, New York, USA.
Orthopaedics/Population Health Science & Policy, Icahn School of Medicine at Mount Sinai, New York, New York, USA.
Reg Anesth Pain Med. 2021 Sep;46(9):779-783. doi: 10.1136/rapm-2021-102715. Epub 2021 Jul 15.
With continuing financial and regulatory pressures, practice of ambulatory total hip arthroplasty is increasing. However, studies focusing on selection of optimal candidates are burdened by limitations related to traditional statistical approaches. Hereby we aimed to apply machine learning algorithm to identify characteristics associated with optimal candidates.
This retrospective cohort study included elective total hip arthroplasty (n=63 859) recorded in National Surgical Quality Improvement Program dataset from 2017 to 2018. The main outcome was length of stay. A total of 40 candidate variables were considered. We applied machine learning algorithms (multivariable logistic regression, artificial neural networks, and random forest models) to predict length of stay=0 day. Models' accuracies and area under the curve were calculated.
Applying machine learning models to compare length of stay=0 day to length of stay=1-3 days cases, we found area under the curve of 0.715, 0.762, and 0.804, accuracy of 0.65, 0.73, and 0.81 for logistic regression, artificial neural networks, and random forest model, respectively. Regarding the most important predictive features, anesthesia type, body mass index, age, ethnicity, white blood cell count, sodium level, and alkaline phosphatase were highlighted in machine learning models.
Machine learning algorithm exhibited acceptable model quality and accuracy. Machine learning algorithms highlighted the as yet unrecognized impact of laboratory testing on future patient ambulatory pathway assignment.
随着财务和监管压力的持续增加，门诊全髋关节置换术的实践正在增加。然而，专注于选择最佳候选者的研究受到与传统统计方法相关的限制。在此，我们旨在应用机器学习算法来确定与最佳候选者相关的特征。
本回顾性队列研究纳入了 2017 年至 2018 年国家手术质量改进计划数据库中记录的择期全髋关节置换术（n=63859）。主要结局是住院时间。共考虑了 40 个候选变量。我们应用机器学习算法（多变量逻辑回归、人工神经网络和随机森林模型）来预测住院时间=0 天。计算了模型的准确性和曲线下面积。
应用机器学习模型将住院时间=0 天与住院时间=1-3 天的病例进行比较，我们发现曲线下面积分别为 0.715、0.762 和 0.804，逻辑回归、人工神经网络和随机森林模型的准确性分别为 0.65、0.73 和 0.81。关于最重要的预测特征，麻醉类型、体重指数、年龄、族裔、白细胞计数、钠水平和碱性磷酸酶在机器学习模型中得到了强调。
机器学习算法表现出可接受的模型质量和准确性。机器学习算法突出了实验室检测对未来患者门诊路径分配的尚未被认识到的影响。
