Machine learning in medicine 3

234 Pages·2014·3.801 MB·English

by  Cleophas, Ton J.;Zwinderman, Aeilko H

Ton J. Cleophas · Aeilko H. Zwinderman
Machine
Learning in
Medicine
Part Three
Machine Learning in Medicine

Ton J. Cleophas (cid:129) Aeilko H. Zwinderman
Machine Learning
in Medicine
Part Three
TonJ.Cleophas AeilkoH.Zwinderman
DepartmentMedicine DepartmentBiostatisticsandEpidemiology
AlbertSchweitzerHospital AcademicMedicalCenter
Sliedrecht,TheNetherlands Amsterdam,TheNetherlands
Additionalmaterial tothis bookcan bedownloaded from extras.springer.com
ISBN978-94-007-7868-9 ISBN978-94-007-7869-6(eBook)
DOI10.1007/978-94-007-7869-6
SpringerDordrechtHeidelbergNewYorkLondon
LibraryofCongressControlNumber:2013955741
©SpringerScience+BusinessMediaDordrecht2013
Thisworkissubjecttocopyright.AllrightsarereservedbythePublisher,whetherthewholeorpart
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation,broadcasting,reproductiononmicrofilmsorinanyotherphysicalway,andtransmissionor
informationstorageandretrieval,electronicadaptation,computersoftware,orbysimilarordissimilar
methodologynowknownorhereafterdeveloped.Exemptedfromthislegalreservationarebriefexcerpts
inconnectionwithreviewsorscholarlyanalysisormaterialsuppliedspecificallyforthepurposeofbeing
enteredandexecutedonacomputersystem,forexclusiveusebythepurchaserofthework.Duplication
ofthispublicationorpartsthereofispermittedonlyundertheprovisionsoftheCopyrightLawofthe
Publisher’s location, in its current version, and permission for use must always be obtained from
Springer.PermissionsforusemaybeobtainedthroughRightsLinkattheCopyrightClearanceCenter.
ViolationsareliabletoprosecutionundertherespectiveCopyrightLaw.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publicationdoesnotimply,evenintheabsenceofaspecificstatement,thatsuchnamesareexempt
fromtherelevantprotectivelawsandregulationsandthereforefreeforgeneraluse.
While the advice and information in this book are believed to be true and accurate at the date of
publication,neithertheauthorsnortheeditorsnorthepublishercanacceptanylegalresponsibilityfor
anyerrorsoromissionsthatmaybemade.Thepublishermakesnowarranty,expressorimplied,with
respecttothematerialcontainedherein.
Printedonacid-freepaper
SpringerispartofSpringerScience+BusinessMedia(www.springer.com)
Machine Learning in Medicine Part Three
by
TONJ.CLEOPHAS,MD,PhD,Professor,
Past-PresidentAmericanCollegeofAngiology,
Co-ChairModuleStatisticsAppliedtoClinicalTrials,
EuropeanInteruniversityCollegeofPharmaceuticalMedicine,Lyon,France,
DepartmentMedicine,AlbertSchweitzerHospital,Dordrecht,Netherlands
AEILKOH.ZWINDERMAN,MathD,PhD,Professor,
PresidentInternationalSocietyofBiostatistics,
Co-ChairModuleStatisticsAppliedtoClinicalTrials,
EuropeanInteruniversityCollegeofPharmaceuticalMedicine,Lyon,France,
Department Biostatistics and Epidemiology, Academic Medical Center, Amsterdam,
Netherlands
Withthehelpfrom
HENNYI.CLEOPHAS-ALLERS,BChem
v

Preface
Machinelearningisanoveldisciplinefordataanalysisthatstartedsome40years
ago with the advent of the computer. It is already widely implemented in socio-/
econometry, but not so in medicine, probably due to the traditional belief of
cliniciansinclinicaltrialswheretheeffectsofmultiplevariablestendtoevenout
by the randomization process and are not further taken into account. Machine
learning is different from traditional data analysis, because, unlike means and
standarddeviations,itusesproximitiesbetweendata,datapatterns,patternrecog-
nition, data thresholding and data trafficking. It is more flexible than traditional
statisticalmethods,anditcanprocessbigdataandhundredsofvariables,anddata,
forwhichanalysiswithoutthehelpofacomputerwouldbeimpossible.
This does not mean that the mathematics of machine learning is simple. Often
integrals,derivatives,complexmatricesandothercalculusmethods,asdeveloped
in the seventeenth century by great western mathematicians like Newton and
Leibnitz,are applied.It isgratifying toobserve, that, three centurieslater,mathe-
maticians from the same schools are responsible for the founding of important
machinelearningmethodologies.
Thefirsttwopartsofthisvolumereviewedbasicmachinelearningmethodslike,
cluster analysis,neural networks, factor analysis, Bayesian networksand support
vector machines. The current Part Three of this volume assesses more recent and
moreadvancedmethodssuchastheonesdescribedunderneath:
Newton’smethods,asthemostadvancedmethodsforfittingexperimentaldatatoa
mathematicalpredictionfunction.
Stochasticmethods,includingadvancedMarkovmodeling,adjustedforirreversible
complicationsanddeath.
Complexsampling,forobtainingunbiasedsamplesfrombigdemographicdata.
Optimal binning, for advanced classification learning of health and health-
educationparameters.
Evolutionary operations, for advanced search of optimal solutions to many
scientificquestions,includingevolutionaryalgorithms,neuroevolution,genetic
programming,andgeneticalgorithms.
vii
viii Preface
Each chapter of the current book was written in a way much similar to that of
thefirsttwovolumes,and
1. Will describe a novel method that has already been successfully applied in the
authors’ownresearch
2. Willprovidetheanalysisofmedicaldataexamples
3. Willprovidestep-by-stepanalysesforthebenefitofthereaders
4. Will provide the commands to be given to the software programs applied
(mostlySPSS)
5. Canbestudiedwithouttheneedtoconsultotherchapters
6. Willbewritteninaexplanatorywayforareadershipofmainlynon-mathematicians
7. Willbeprovidedtogetherwithadatafileontheinternetthroughextras.springer.
comforthebenefitofinvestigatorswhowishtoperformtheirownanalyses
We should add that the authors are well-qualified in their field. Professor
Zwinderman is president of the International Society of Biostatistics
(2012–2015), and Prof. Cleophas is past-president of the American College of
Angiology(2000–2002).Fromtheirexpertisetheyshouldbeabletomakeadequate
selectionsofmodernmethodsforclinicaldataanalysisforthebenefitofphysicians,
students,andinvestigators.Theauthorshavebeenworkingandpublishingtogether
for 15 years, and their research can be characterized as a continued effort to
demonstrate that clinical data analysis is not mathematics but rather a discipline
attheinterfaceofbiologyandmathematics.
The authors as professors and teachers in statistics for 20 years are convinced
that the current part three of this three-volume book contains information
unpublished so far and vital to the analysis of the complex medical data as
commonlywitnessedtoday.
Machinelearningisanoveldisciplineconcernedwiththeanalysisoflargedata
and multiple variables. It involves computationally intensive methods, and is
currently mainly the domain of computer scientists, and is already commonly
used in social sciences, marketing research, operational research and applied
sciences.
Machinelearningisvirtuallyunusedinclinicalresearch.Thisisprobablydueto
the traditional belief of clinicians in clinical trials where multiple variables are
equally balanced by the randomization process and are not further taken into
account. In contrast, modern computer data files often involve hundreds of vari-
ables like genes and other laboratory values, and computationally intensive
methodsarerequiredfortheiranalysis.
Lyon,France TonJ.Cleophas
02-09-2013 AeilkoH.Zwinderman
Contents
1 IntroductiontoMachineLearninginMedicinePartThree. . . . . . 1
1 Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Objective. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 ResultsandConclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
3 ContentsofPartOneofThis3VolumeBook. . . . . . . . . . . . . . . 2
4 ContentsofPartTwoofThis3VolumeBook. . . . . . . . . . . . . . . 3
5 ContentsofPartThreeofThis3VolumeBook. . . . . . . . . . . . . . 5
6 Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
References. . . . . . . . . . .. . . . . . . . . . . . . . .. . . . . . . . . . . . . .. . . . 9
2 EvolutionaryOperations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1 Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.1 Background. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2 Objective. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 Methods. .. . . .. . . .. . . .. . . .. . . .. . . .. . . .. . . .. . . .. 11
1.4 Results.. . . .. . . .. . . .. . . . .. . . .. . . .. . . .. . . .. . . .. . 11
1.5 Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3 ExampleData. . . .. . . . .. . . .. . . .. . . .. . . .. . . .. . . .. . . .. . 13
4 MethodologicalBackgroundofGeneticAlgorithms. . . . . . . . . . . 13
5 ResultsofaGeneticAlgorithmtoVariableSelection
inOurExampleData. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
6 Discussion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
7 Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
References. . . . . . . . . . .. . . . . . . . . . . . . . .. . . . . . . . . . . . . .. . . . 17
3 MultipleTreatments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1 Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.1 Background. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.2 Objective. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.3 Methods. .. . . .. . . .. . . .. . . .. . . .. . . .. . . .. . . .. . . .. 19
ix