摘要
目的通过机器学习筛选出烧伤患者创面细菌感染的风险因素,建立烧伤患者入院后创面细菌感染的风险预测模型,为预防和治疗烧伤患者入院后创面细菌感染提供新思路。方法纳入南京医科大学附属苏州医院北区烧伤科2023年1月1日至8月1日因烧伤而收治入院的139例患者,按创面细菌培养结果分为Ⅰ组(培养结果阴性)77例和Ⅱ组(培养结果阳性)62例。收集2组患者年龄、性别、烧伤面积(BA)及入院时进行检测的血常规指标(17项)、血凝指标(7项)和生化指标(17项)等44项指标。通过最小绝对值选择和收缩算子(LASSO)回归技术来筛选变量数作为风险因素,使用逻辑回归logistic方法构建风险预测列线图模型,通过内部随机分组(训练集0.8∶验证集0.2),并分别绘制出其受试者工作特征曲线,计算曲线下面积评价预测效果。通过绘制校准曲线对模型进行最后的性能验证。结果通过LASSO回归技术分析筛选出性别(拦截值:0.30)、BA(拦截值:1.60)、白细胞计数(拦截值:0.34)、碱性磷酸酶(拦截值:0.28)、甘油三酯(拦截值:-0.71)、嗜碱性粒细胞百分率(拦截值:1.05)、白蛋白(拦截值:-1.62)和白蛋白/球蛋白(拦截值:0.05)8个可以评估烧伤患者创面细菌感染风险的因素,并且根据这些变量构建了列线图,表现出良好的一致性和准确性。结论烧伤患者性别、BA及血液检验中6个指标与伤口创面细菌感染之间存在非线性和正相关联,建立的预测模型对烧伤患者创面细菌感染的风险有一定预测价值。
Objective To screen risk factors for wound bacterial infection in burn patients using machine learning and establish a risk prediction model for wound bacterial infection in burn patients after admission,to provid new insights for the prevention and treatment of wound bacterial infection in burn patients after admission.Methods A total of 139 patients admitted to the Burn Department of the Northern District of Suzhou Hospital Affiliated to Nanjing Medical University due to burns from January 1,2023 to August 1,2023 were enrolled and divided into groupⅠ(77 patients with negative culture results)and groupⅡ(62 patients with positive culture results)based on wound bacterial culture results.Forty-four indicators including age,gender,burn area(BA),and 17 blood routine indicators,7 blood coagulation indicators,and 17 biochemical indicators tested upon admission were collected from both groups.The Least Absolute Shrinkage and Selection Operator(LASSO)regression technique was used to screen the number of variables as risk factors,and a risk prediction nomogram model was constructed using the logistic regression method.Internal random grouping(training set 0.8∶test set 0.2)was performed,and their receiver operating characteristic curves were plotted respectively to calculate the area under the curve for evaluating the prediction effect.The final performance validation of the model was conducted by plotting a calibration curve.Results Through LASSO regression technique analysis,eight factors were screened,including gender(intercept:0.30),BA(intercept:1.60),white blood cell count(intercept:0.34),alkaline phosphatase(intercept:0.28),triglycerides(intercept:-0.71),basophil percentage(intercept:1.05),albumin(intercept:-1.62),and albumin/globulin(intercept:0.05),which could assess the risk of wound bacterial infection in burn patients.A nomogram was constructed based on these variables,demonstrating good consistency and accuracy.Conclusion There is a nonlinear and positive correlation between gender,BA and six blood test i
作者
杨晓春
丁亚
顾秀玉
YANG Xiaochun;DING Ya;GU Xiuyu(Department of Laboratory Medicine,Suzhou Hospital Affiliated to Nanjing Medical University,Suzhou,Jiangsu 215008,China)
出处
《现代医药卫生》
2024年第24期4155-4161,共7页
Journal of Modern Medicine & Health
关键词
烧伤
创面细菌感染
风险预测
机器学习
列线图
Burn
Wound bacterial infection
Risk prediction
Machine learning
Alignment Diagram
