Department of Systems Science and Industrial Engineering, State University of New York at Binghamton, Binghamton, New York, USA.
Department of Surgical Oncology and Colorectal Surgery, Medical Center Navicent Health, Macon, Georgia, USA.
Surg Infect (Larchmt). 2020 Nov;21(9):784-792. doi: 10.1089/sur.2019.138. Epub 2020 Mar 6.
Post-operative infections have many negative consequences for patients' health and the healthcare system. Among other things, they increase the recovery time and the risk of re-admission. Also, infection results in penalties for hospitals and decreases the quality performance measures. Surgeons can take preventive actions if they can identify high-risk patients. The purpose of this study was to develop a model to help predict those patients at risk for post-operative infection. A retrospective analysis was conducted on patients with colorectal post-operative infections. Univariable analysis was used to identify the features associated with post-operative infection. Then, a support vector classification-based method was employed to select the right features and build prediction models. Decision tree, support vector machine (SVM), logistic regression, naïve Bayes, neural network, and random forest algorithms were implemented and compared to determine the performance algorithm that best predicted high-risk patients. From 2016 to the first quarter of 2017, 208 patients who underwent colorectal resection were analyzed. The factors with a statistically significant association (p < 0.05) with post-operative infections were elective surgery, origin status, steroid or immunosuppressant use, >10% loss of body weight in the prior six months, serum creatinine concentration, length of stay, unplanned return to the operating room, administration of steroids or immunosuppressants for inflammatory bowel disease, use of a mechanical bowel preparation, various Current Procedural Terminology (CPT) codes, and discharge destination. However, accurate prediction models can be developed with seven factors: age, serum sodium concentration, blood urea nitrogen, hematocrit, platelet count, surgical procedure time, and length of stay. Logistic regression and SVM were stable models for predicting infections. The models developed using the pre-operative features along with the full list of features helped us interpret the results and determine the significant factors contributing to infections. These factors present opportunities for proper interventions to mitigate infection risks and their consequences.
术后感染会对患者的健康和医疗系统造成许多负面影响。除其他外，它们会增加康复时间和再次入院的风险。此外，感染会导致医院受到处罚，并降低质量绩效指标。如果外科医生能够识别出高风险患者，他们可以采取预防措施。本研究的目的是开发一种模型，以帮助预测那些有术后感染风险的患者。对接受结直肠手术后感染的患者进行回顾性分析。采用单变量分析来确定与术后感染相关的特征。然后，采用基于支持向量分类的方法选择合适的特征并构建预测模型。实现了决策树、支持向量机（SVM）、逻辑回归、朴素贝叶斯、神经网络和随机森林算法，并进行了比较，以确定预测高危患者的最佳性能算法。在 2016 年至 2017 年第一季度期间，对接受结直肠切除术的 208 例患者进行了分析。与术后感染有统计学显著关联（p<0.05）的因素包括择期手术、原籍状况、使用类固醇或免疫抑制剂、前六个月体重减轻>10%、血清肌酐浓度、住院时间、计划外返回手术室、为炎症性肠病使用类固醇或免疫抑制剂、使用机械肠道准备、各种当前程序术语（CPT）代码以及出院目的地。然而，可以使用七个因素开发准确的预测模型：年龄、血清钠浓度、血尿素氮、血细胞比容、血小板计数、手术时间和住院时间。逻辑回归和 SVM 是预测感染的稳定模型。使用术前特征和完整特征列表开发的模型有助于我们解释结果并确定导致感染的重要因素。这些因素为适当的干预措施提供了机会，可以降低感染风险及其后果。
