老年髋部骨折患者围术期输血风险预测模型建立及验证的研究

Study of perioperative blood transfusion risk prediction model and verification for elderly patients with hip fracture

目的建立老年髋部骨折患者围术期输血风险预测模型。方法选取2017年5月至2020年12月就诊于北京积水潭医院急诊科的新发髋部骨折患者3813例,回顾性分析患者的性别、年龄、骨折类型、手术方式、既往病史及并发症。随机抽取70%的患者作为建模集,分析老年髋部骨折患者围术期输血的危险因素,并建立老年髋部骨折患者围术期输血风险预测模型;余30%的患者纳入验证集并代入预测模型,进行老年髋部骨折患者围术期输血风险预测模型的内部验证。结果建模集共纳入2668例患者,其中输血1107例(41.5%)、未输血1561例(58.5%)。多因素logistic回归分析显示,女性(OR=0.479,95%CI:0.389~0.591,P=0.000)、年龄越大(OR=1.038,95%CI:1.025~1.052,P=0.000)、股骨粗隆间骨折(OR=0.037,95%CI:0.024~0.058,P=0.000)、内固定术(OR=0.192,95%CI:0.125~0.296,P=0.000)、泌尿系统疾病史(OR=1.545,95%CI:1.125~2.122,P=0.007)、肿瘤病史(OR=1.593,95%CI:1.100~2.307,P=0.014)、血液病史(OR=7.385,95%CI:4.138~13.179,P=0.000)和低蛋白血症(OR=3.572,95%CI:2.388~5.342,P=0.000)是老年髋部骨折患者围术期输血的危险因素,骨质疏松史(OR=0.538,95%CI:0.322~0.897,P=0.017)和其他动脉粥样硬化疾病史(OR=0.368,95%CI:0.170~0.795,P=0.011)是老年髋部骨折患者围术期输血的保护因素。建立预测模型:logit(P)=-0.735×性别+0.037×年龄(岁)-3.288×骨折部位-1.648×手术方式-0.620×骨质疏松史+0.435×泌尿系统疾病史+0.466×肿瘤病史+1.999×血液病史-1.000×其他动脉粥样硬化疾病史+1.273×低蛋白血症-0.020。输血风险模型预测的AUC为0.805,截断值为0.441,灵敏度为0.711,特异性为0.763,约登指数为0.474。对模型进行内部验证,结果显示灵敏度、特异性、诊断准确率、约登指数和相对危险度分别为0.825、0.620、0.739、0.445和2.886。结论性别、年龄、骨折部位、手术方式、泌尿系统疾病史、肿瘤病史、血液病史和低蛋白血症为老年髋部骨折患者围术期输血的危险因素,骨质疏松和其他动脉粥样硬化疾病史是其保护因素。基于上述因素建立的预测模型,对于老年髋部骨折患者围术期是否输血具有良好的筛查能力。

Objective To establish a perioperative blood transfusion risk prediction model for elderly patients with hip fractures.Methods A total of 3813 elderly patients with hip fractures in the Department of Emergency Medicine,Beijing Jishuitan Hospital from May 2017 to December 2020 were selected.The data of patients including gender,age,fracture type,surgical method,past medical history and complications were retrospectively analyzed.Firstly,70%of all cases(modeling dataset)were randomly selected to analyze the risk factors of blood transfusion in elderly hip fracture patients and establish a perioperative blood transfusion risk prediction model for elderly hip fracture patients.Secondly,the other 30%cases(validation dataset)were substituted to verify the perioperative blood transfusion risk prediction model for elderly hip fracture patients.Results Among the 2668 patients in the modeling dataset,1107(41.5%)cases received blood transfusion during the perioperative period,while the other 1561(58.5%)cases didn′t.Multivariate logistic regression analysis showed that female(OR=0.479,95%CI:0.389-0.591,P=0.000),older age(OR=1.038,95%CI:1.025-1.052,P=0.000),intertrochanteric fracture of femur(OR=0.037,95%CI:0.024-0.058,P=0.000),internal fixation(OR=0.192,95%CI:0.125-0.296,P=0.000),history of urological disease(OR=1.545,95%CI:1.125-2.122,P=0.007),history of tumor disease(OR=1.593,95%CI:1.100-2.307,P=0.014),history of blood disease(OR=7.385,95%CI:4.138-13.179,P=0.000)and hypoproteinemia(OR=3.572,95%CI:2.388-5.342,P=0.000)were risk factors for perioperative blood transfusion in elderly patients with hip fracture.History of osteoporosis(OR=0.538,95%CI:0.322-0.897,P=0.017)and history of other atherosclerotic disease(OR=0.368,95%CI:0.170-0.795,P=0.011)were protective factors for perioperative blood transfusion in elderly patients with hip fracture.The prediction model was:logit(P)=-0.735×gender+0.037×age(years)-3.288×fracture type-1.648×operation type-0.620×osteoporosis+0.435×urinary system diseases+0.466×tumor history+1.999×blood disease-1.000×other arterial diseases+1.273×hypoproteinemia-0.020.The area under the curve predicted by the blood transfusion risk model was 0.805,the cutoff value was 0.441,the sensitivity was 0.711,the specificity was 0.763,and the Youden index was 0.474.The internal verification results showed that the sensitivity,specificity,diagnostic accuracy,Youden index and relative risk were 0.825,0.620,0.739,0.445 and 2.886,respectively.Conclusions Among the influencing factors of perioperative blood transfusion in elderly patients with hip fracture,gender,age,fracture site,surgical method,history of urinary diseases,history of tumor disease,history of blood disease and hypoproteinemia are risk factors,and the history of osteoporosis and other atherosclerotic diseases are protective factors.The prediction model based on the above factors has a good screening ability for perioperative blood transfusion in elderly patients with hip fracture.