基于logistic回归模型的重症监护室转入率预测研究

Study on the prediction of intensive care unit transfer rate based on logistic regression model

目的探讨基于logistic回归的重症监护室(intensive care unit,ICU)转入率预测模型,评价模型的预测效能。方法选取2022年10月至2023年2月河北北方学院附属第一医院HIS系统患者1000例,采用随机抽样的方法将患者分为训练集和验证集,每集各500例。采用多因素logistic回归方程分析训练集患者转入ICU的影响因素,采用ROC曲线评价ICU转入率预测模型效能并内部验证。结果训练集500例患者中男358例,女142例,年龄37~79岁,平均(54.5±3.1)岁。多因素logistic回归分析结果显示,年龄≥60岁、有吸烟史、饮酒史、心脏疾病、血管疾病、凝血功能障碍、肝脏疾病、肾脏疾病、高血压、颈椎骨折、胸部骨折、下肢骨折、脑损伤、气血胸、肠道损伤、膀胱损伤、呼吸系统手术、胃肠系统手术、肝胆系统手术的患者,转入ICU的风险越高(P<0.05)。急重症患者转入ICU的预测模型为:logit(P)=1.050X_(1)+2.014X_(2)+1.024X_(3)+1.102X_(4)+1.079X_(5)+1.060X_(6)+1.906X_(7)+1.947X_(8)+1.059X_(9)+1.016X_(10)+1.903X_(11)+1.078X_(12)+2.017X_(13)+1.055X_(14)+1.078X_(15)+1.042X_(16)+3.051X_(17)+2.011X_(18)+1.026X_(19)-145.23。训练集和验证集模型的ROC曲线的AUC分别为0.925(95%CI:0.890~0.957,P<0.05)、0.896(95%CI:0.852~0.931,P<0.05),灵敏度分别为0.8461、0.8102,特异性分别为0.9072、0.8852,约登指数分别为0.7533,0.6954,一致性指数分别为0.901(95%CI:0.882~0.937)、0.862(95%CI:0.837~0.896)。训练集和验证集模型均未偏离完美拟合(χ^(2)=0.133,P=0.625;χ^(2)=0.255,P=0.198),校准曲线Brier score分别为0.145(95%CI:0.103~0.180)、0.183(95%CI:0.133~0.220),平均绝对误差分别为0.017、0.026。结论ICU转入率预测模型效能较为理想,能够为合理调配医疗资源提供一定的参考。

Objective To explore the intensive care unit(ICU)transfer rate prediction model based on logistic regression,and to evaluate the predictive performance of the model.Methods A total of 1000 patients with HIS system in the First Affiliated Hospital of Hebei North University from October 2022 to February 2023 were selected,and were randomly divided into the training set and the validation set,with 500 patients in each set.The influencing factors of patients transferred to ICU in the training set by multivariate logistic regression,and the effectiveness of the ICU transfer rate prediction model was evaluated by using ROC curve and the internal validation was conducted.Results Among the 500 patients in the training set,there were 358 males and 142 females,aged from 37 to 70 years,with an average age of(54.5±3.1)years.The results of multivariate logistic regression analysis showed that patients over 60 years old with smoking history,drinking history,heart disease,vascular disease,coagulation dysfunction,liver disease,kidney disease,hypertension,cervical fracture,chest fracture,lower limb fracture,brain injury,pneumohemothorax,intestinal injury,bladder injury,respiratory system surgery,gastrointestinal system surgery and hepatobiliary system surgery had higher risk of being transferred to ICU(P<0.05).The predictive model of acute and severe patients transferring to ICU was:logit(P)=1.050X_(1)+2.014X_(2)+1.024X_(3)+1.102X_(4)+1.079X_(5)+1.060X_(6)+1.906X_(7)+1.947X_(8)+1.059X_(9)+1.016X_(10)+1.903X_(11)+1.078X_(12)+2.017X_(13)+1.055X_(14)+1.078X_(15)+1.042X_(16)+3.051X_(17)+2.011X_(18)+1.026X_(19)-145.23.The AUC of ROC curve of the training set and the validation set were 0.925(95%CI:0.890-0.957,P<0.05)and 0.896(95%CI:0.852-0.931,P<0.05),with sensitivity of 0.8461 and 0.8102,specificity of 0.9072 and 0.8852,Youden index of 0.7533 and 0.6954,and consistency index of 0.901(95%CI:0.882-0.937)and 0.862(95%CI:0.837-0.896),respectively.Neither the training set or the validation set model deviated from the perfect fit(χ^(2)=0.133,P=0.625;χ^(2)=0.255,P=0.198).The calibration curve Brier scores were 0.145(95%CI:0.103-0.180)and 0.183(95%CI:0.133-0.220),respectively,with average absolute errors of 0.017 and 0.026.Conclusions The ICU transfer rate prediction model has ideal performance and can provide a certain reference for the rational allocation of medical resources.