髋关节置换术异体红细胞需求量的智能模型构建

发布时间：2018-03-29 08:55

本文选题：数据清洗　切入点：临床输血数据库　出处：《中国人民解放军医学院》2017年硕士论文

【摘要】：目的本研究旨在探索快速清洗临床输血数据库的方法,并利用清洗后的有效数据,进行数据挖掘研究,对髋关节手术病人的输血现状进行概述分析,建立髋关节置换术病人的异体红细胞需求量智能模型,探索血液保护策略,促进个体化、精细化的输血治疗方案的建立与发展。方法1、通过学习数据清洗理论,探索数据清洗的实现方式与科学方法,针对清洗过程的不同阶段,设计数据清洗流程,结合计算机技术及人工校正的手段,利用“筛选”、“诊断”、“编辑”的方法针对不同的“问题”数据进行分别处理,并通过大数据学习过程,清洗得到研究分析所需要的数据类型及格式,并获得全部数据内容,以便为数据挖掘及数据分析做基础。2、利用清洗临床输血数据库所得到的髋关节手术的病人病例,回顾性分析我院自1999年-2015年病例共4250例,包括病人完整的病历信息数据,利用SPSS软件进行统计分析,比较各年间异体红细胞输注概况的差异,对近年来临床用血概况进行描述。3、筛选2000年-2015年之间我院完整病人病例共1064例,将数据资料进行清洗、汇总,获取病人完整信息数据。通过查阅文献得到与髋关节置换术异体红细胞输注相关因素,利用多元回归模型及机器学习方法建立髋关节置换术异体红细胞需求量的预测模型。4、选取2013年1月-2015年6月我院由同一骨科团队实施手术的病人181例,分组为使用TXA组88例,对照组93例,比较手术前后Hb值、术中Hb下降值、异体血输注量、失血量、引流量等指标之间的差异。结果1、按照数据清洗流程图对临床输血数据库提取出的数据进行“筛选”、“诊断”、“编辑”,每例病例包括病人的基本信息、手术信息、生命体征、术前术后实验室检查信息、输血信息在内的99项信息列表,所涵盖的信息包括字符、数字等多种数据类型。取得全部有效数据之后,可以根据不同的数据挖掘的研究需求,调取所需病例及相关的数据信息。2、4250例进行髋关节手术的病例中,进行过输血治疗的病人比例为55.72%,异体红细胞平均输注量为3.17±1.27单位;平均住院日逐年下降;各年的输血比例约在40%-70%之间;手术中异体红细胞台均输注量在2006年之后均维持在4单位左右,而台均申请量在2009年之后,各年均低于3单位;单/双髋关节手术围术期异体红细胞平均输注量分别在2009/2010年达到最大,之后均逐年下降。3、自2000年-2015年在我院手术、经清洗后具有完整记录的1064例,选取异体红细胞输注的独立影响因素纳入模型,经多元线性回归、神经网络、GBDT等人工智能方法建模,可知GBDT最适用于输血量预测模型的建立。4、病人术前Hb(P=0.850)、失血量(P=0.072)、引流量(P=0.214)、异体红细胞平均使用量(P=0.091)等指标在两组间的差异不存在统计学意义;术后Hb、术中Hb下降程度在两组间的差异均有统计学意义。结论利用计算机技术以及数据学习过程对临床输血数据库进行清洗的流程及方法是有效的;对于输血量的预测模型,使用GBDT方法较为有效;低剂量使用TXA药物可以降低手术中Hb的损失,但并不能改善输血量,作为血液保护手段,可能需要更大剂量的TXA来降低病人对于异体红细胞的依赖。
[Abstract]:The purpose of this study was to explore the method of rapid cleaning of clinical blood transfusion database, and the use of effective data cleaning after the study of data mining, summarizes the status quo analysis of blood transfusion on patients with hip surgery, allogeneic red blood cells demand Chi established hip replacement patients can model, explore the blood conservation strategy, promote the establishment of individual. With the development of blood transfusion treatment fine. Methods 1, through the study of the theory of data cleansing, data exploration and scientific methods for cleaning, according to different stages of the cleaning process, the design of data cleaning process, the combination of computer technology and artificial means of correction, the use of "screening", "diagnosis", "Edit" method according to the different "problem" data are processed, and through big data analysis on the learning process, cleaning the data needed to pass the type type, and all In order to analyze the content of the data,.2 based data mining and data cleaning, using hip surgery by clinical blood transfusion database of patients were analyzed retrospectively in our hospital from 1999 -2015 in 4250 cases, including patients with complete medical information data, statistical analysis was performed using SPSS software. The comparison between allogeneic red blood cells the difference of the lost note, in recent years of blood for clinical use are described in.3, screening between 2000 -2015 cases in our hospital patients complete a total of 1064 cases, data cleaning, summary, complete access to patient information data. Through the literature note factors associated with hip replacement allogeneic red blood cell transfusion, using multiple regression model and machine learning method of.4 model of hip replacement allogeneic red blood cell demand, by the same team Department of orthopedics from January 2013 -2015 year in June in our hospital implementation of hand 181 cases of patients, divided into TXA group of 88 cases, 93 cases in control group were compared before and after the operation of Hb, with the decrease of Hb, allogenic blood transfusion volume, blood loss, the difference between the drainage index. Results 1, to "screen" in accordance with the data flow chart of clinical transfusion database cleaning the extracted data, "diagnosis", "Edit", each of the cases including basic information, patient information, vital signs, preoperative and postoperative laboratory examination information, a list of 99 items of information, information of blood transfusion, which includes information on characters, digital and other data types. After all the valid data. Can the demand according to different data mining research, the transfer of required cases and related data of.24250 cases were hip surgery cases, the proportion of patients had blood transfusion was 55.72%, the average amount of infusion of allogeneic red blood cells was 3.17 + 1.27 units; The average hospitalization days decreased year by year; the annual rate of blood transfusion is about 40%-70%; allogeneic red blood cell transfusion in surgery after 2006 was maintained at about 4 units, and the units are applications in 2009, the average of less than 3 units; single / double hip surgery perioperative allogeneic red blood cell average the infusion amount in 2009/2010 reached the maximum after.3 were decreased year by year, from 2000 -2015 years in our hospital surgery, 1064 cases after cleaning with a complete record, select the independent influencing factors of allogeneic red blood cell transfusion into the model, the multivariate linear regression, neural network, GBDT modeling method of artificial intelligence the GBDT is most suitable for blood transfusion in the prediction model of.4 patients before Hb (P=0.850), blood volume (P=0.072), drainage (P=0.214), the average amount of allogeneic red blood cells (P=0.091) and other indicators in the difference between the two groups was not statistically significant; after Hb Hb, intraoperative decreased degree of difference between the two groups were statistically significant. Conclusion the use of computer technology and data learning process and methods of cleaning of clinical blood transfusion database is effective for blood transfusion; prediction model, using the GBDT method is effective; low dose can reduce the operating losses of Hb use of TXA drugs but can not improve the amount of blood transfusion, blood as a means of protection, may require higher doses of TXA to reduce patient dependence on allogeneic red blood cells.

【学位授予单位】：中国人民解放军医学院
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R687.4;R446.1

【相似文献】