大数据与临床科研大数据大数据指不用随机分析法(抽样调查)这样的捷径，而采用所有数据进行分析处理维克托迈尔-舍恩伯格大数据时代2Velocity(高速)Variety(多样)Veracity(真实)Value(价值)Volume(大量)01040502035VRCT的困境BiologicalBiologicalBiological efficacyefficacyefficacyZhangZ.Big data and clinical research:focusing on the area of critical care medicine in mainland China.QuantImagingMedSurg.2014Oct;4(5):426-9.3困境严格纳入/排除标准20%患者的信息用于治疗80%的患者严格的执行方案EffectivenessEffectivenessEffectiveness4ProMISeTrialInvestigators.Trial of early,goal-directed resuscitation for septic shock.NEnglJMed.2015Apr2;372(14):1301-11.1260例56研究中心3.5年6.4例/年/中心临床大数据优点5大量整个EMR系统；非抽样统计多样各种并发症，非单病种高速云计算真实真实世界研究(effectiveness)价值Critical data66My experience with big dataZhangZ,ChenK,NiH.Calciumsupplementationimprovesclinicaloutcomeinintensivecareunitpatients:apropensityscorematchedanalysisofalargeclinicaldatabaseMIMIC-II.Springerplus.2015Oct13;4:594.ZhangZ.Amathematicalmodelforpredictingglucoselevelsincritically-illpatients:thePIGnOLImodel.PeerJ.2015Jun9;3:e1005.ZhangZ,ChenL,NiH.Antipyretictherapyincriticallyillpatientswithsepsis:aninteractionwithbodytemperature.PLoSOne.2015Mar30;10(3):e0121919.7Zhang Z.Accessing critical care big data:a step by step approach.J Thorac Dis.2015 Mar;7(3):238-42.8 8Zhang Z.Big data and clinical research:focusing on the area of critical care medicine in mainland China.Quant Imaging Med Surg.2014 Oct;4(5):426-9.9我的大数据专栏10ZhangZ.DatamanagementbyusingR:bigdataclinicalresearchseries.Ann Transl Med.2015Nov;3(20):303.ZhangZ.Missingvaluesinbigdataresearch:somebasicskills.Ann Transl Med.2015Dec;3(21):323.ZhangZ.Missingdataexploration:highlightinggraphicalpresentationofmissingpattern.Ann Transl Med.2015Dec;3(22):356.ZhangZ.MultipleimputationfortimeseriesdatawithAmeliapackage.Ann Transl Med.2016Feb;4(3):56.分析方法*logistic、coxregression多因素分析*单因素分析亚组分析分层分析倾向性评分1111倾向性评分近年来比较时髦的方法五大综合性医学杂志上用倾向分方法发表的文章数量JCompEffRes.2014Jan;3(1):63-78.1212羟考酮为例说明年资A经济B病情C医保D降维倾向性评分对所有条件进行逐一匹配，使两组具有可比性；但如果有100个因素呢？自然状态下羟考酮使用情况1.高年资医生2.家庭条件较好3.病情严重、不稳定4.医保情况1313HighprobabilityofparticipatinggivenXDensity01Propensity scoreRegionofcommonsupportDensity of scores for participantsDensity of scores for non-participants1414Zhang Z,Chen K,Ni H.Calcium supplementation improves clinical outcome in intensive care unit patients:a propensity score matched analysis of a large clinical database MIMIC-II.Springerplus.2015 Oct 13;4:594.15分数多项式模型16Zhang Z,Chen K,Ni H,Fan H.Predictive value of lactate in unselected critically ill patients:an analysis using fractional polynomials.J Thorac Dis.2014 Jul;6(7):995-1003.17适用于一些非线性关系的变量描述18ZhangZetal.Quadraticfunctionbetweenarterialpartialoxygenpressureandmortalityriskinsepsispatients:aninteractionwithsimplifiedacutephysiologyscore.SciRep.2016Oct13;6:35133.机器学习19ZhangZ.When doctors meet with AlphaGo:potential application of machine learning to clinical medicine.AnnTranslMed2016;4(6):125.Pubmed results20机器学习掠影近邻取样01贝叶斯分类法02决策树/随机森林03神经网络模型0421K-近邻取样22ZhangZ.Introductiontomachinelearning:k-nearestneighbors.AnnTranslMed.2016Jun;4(11):218.决策树23ZhangZ.DecisiontreemodelingusingR.AnnTranslMed.2016Aug;4(15):275.熵不断减少的过程，寻求有序化神经网络模型24ZhangZ.Agentleintroductiontoartificialneuralnetworks.AnnTranslMed.2016Oct;4(19):370.应用场景ChenL,DubrawskiA,WangD,FiterauM,Guillame-BertM,UsingSupervisedMachineLearningtoClassifyRealAlertsandArtifactinOnlineMultisignalVitalSignMonitoringData.CritCareMed.2016Mar17.DevelopmentandValidationofaDeepLearningAlgorithmforDetectionofDiabeticRetinopathyinRetinalFundusPhotographs.JAMA.2016Nov29.TempleMW,LehmannCU,FabbriD.PredictingDischargeDatesFromtheNICUUsingProgressNoteData.Pediatrics.2015Aug;136(2):e395-405.25123统计软件SPSS1STATA 11.022626R语言：优雅、卓越的统计分析及绘图环境R已成为人们走出学校后必修的第二门外语，并且在R中有太多令人惊奇的代码。-辉瑞公司（Pfizer）临床数据部的副总监 Max Kuhn永久地改变了人们分析、可视化、处理数据的方式；是一个优雅的、被广泛接受的、不朽的软件系统。-美国计算机协会（ACM）的评价还有很多人提到，R语言的学习曲线非常陡峭。但从我多年的使用经验上看，陡峭的学习曲线并不是因为R语言本身，而是隐藏在后面的统计知识很难在短时间内掌握的缘故。R27作者:刘思喆来源:程序员Success of R course from Spring 2008,judged by self-reported%usage of R among all statistical programs28Final Words of Warning“UsingRisabitakintosmoking.Thebeginningisdifficult,onemaygetheadachesandevengagthefirstfewtimes.Butinthelongrun,itbecomespleasurableandevenaddictive.Yet,deepdown,forthosewillingtobehonest,thereissomethingnotfullyhealthyinit.”-FrancoisPinardR29Take home message大数据时代传统RCT的困境临床医生与大数据研究传统统计学框架下的大数据使用机器学习3031