chapter 3 risk assessment models for patients with ... 3.pdf · in primary care research, routine...
TRANSCRIPT
VU Research Portal
Medically unexplained physical symptoms in primary care
den Boeft, M.
2016
document versionPublisher's PDF, also known as Version of record
Link to publication in VU Research Portal
citation for published version (APA)den Boeft, M. (2016). Medically unexplained physical symptoms in primary care: Identification, management andsocietal aspects.
General rightsCopyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright ownersand it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
• Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal ?
Take down policyIf you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediatelyand investigate your claim.
E-mail address:[email protected]
Download date: 24. Feb. 2021
1
CHAPTER3 Riskassessmentmodelsforpatientswithpersistentmedicallyunexplainedphysical
symptomsinprimarycareusingelectronicmedicalrecords
MadelondenBoeft
MarkHoogendoorn
JosWR.Twisk
SjanneNap
TimvanderNeut
JohannesC.vanderWouden
HenrietteE.vanderHorst
MattijsE.Numans
Submitted
2
ABSTRACT
Background
Patients with persistent medically unexplained physical symptoms (MUPS) often experience
limitationsduetotheirsymptoms,haveanimpairedqualityoflifeandincursubstantialhealth
carecosts. EarlyidentificationofpatientswithMUPSwhohaveahighriskofpersistentMUPS
could lead to early intervention by general practitioners (GPs), which could prevent the
transitiontopersistentMUPS.However, therearenoeffectiveandefficientscreeningmethods
to identifypatientswithahighriskofpersistentMUPS.Therefore, toenhanceidentificationof
theseriskpatientsandthussupportGPs,weaimedtodevelopariskassessmentmodelusinga
largedatasetextractedfromprimarycareelectronicmedicalrecords(EMRs).
Methods
WeusedanonymisedEMRsfrom22Dutchgeneralpractices(n=156.176)overafive-yearperiod
(2007-2011).WeoperationalizedpersistentMUPSby combining international classificationof
primarycare(ICPC)codesforirritablebowelsyndrome,fibromyalgia,chronicfatiguesyndrome
and lowbackpainwithout radiationas anendpoint.Webalanced thedatasetby includingall
patientswith persistentMUPS (n=7840) and a randomly selected sample of patientswithout
persistentMUPS(n=7988).Weappliedlogisticregressionanalysisanddecisiontreeanalysisto
assess thepersistentMUPS riskon80%of thedataset.Theperformancesof themodelswere
evaluatedwiththeareaunderthecurve(AUC)ontheremaining20%ofthedata,thevalidation
set.
Results
The variable best discriminating for persistent MUPS in both models was the number of
episodes.RiskassessmentofpersistentMUPSwasperformedgoodwiththedecisiontree(AUC
= 0.81) and moderate to good with logistic regression (AUC = 0.70). The validation showed
acceptablestability(0.78and0.70,respectively).
Conclusion
We explored the possibilities of assessment of patients’ individual risks for persistent MUPS
based on routine EMRs. We developed two moderate to good performing risk assessment
models, partly by using relatively new datamining techniques.While our algorithms require
further validation and fine-tuning, they provide a starting point fromwhichGPs could evolve
towardsamoreproactive,structuredMUPSmanagementtogetherwiththeirpatients.
3
BACKGROUND
Twenty to thirtyper centof all physical symptomspresented to thegeneralpractitioner (GP)
remains unsatisfactorily explained after adequate evaluation (1,2). Most of the so called
medicallyunexplainedphysicalsymptoms(MUPS)resolvespontaneously.However,aminority
of patients in whom MUPS become persistent suffer from major impairments in daily
functioning anda reducedqualityof life.Also, persistentMUPS lead tohigh costsdue tohigh
healthcareutilizationanddecreasedworkfunctioning(3,4).
Early assessment of the risk of patients developing persistent MUPS by the GP could create
possibilities for offering structured, proactive care and the prevention from the transition of
MUPS to persistentMUPS in subgroups of patientswith a high risk. This combination of risk
assessment followed by structured, specified proactive care interventions is called panel
management (5). Depending on the nature of the symptoms, functional limitations and the
preferences of the patient, different treatment options, for example cognitive behavioural
therapyorphysicaltherapy,canbediscussedandofferedatanearlystage(6,7).
TheidentificationofpatientsatriskforpersistentMUPSthatisneededforpanelmanagementis
not a trivial task as there are no generally accepted determinants to guide any selection
procedure. Other challenges include the heterogeneity amongMUPS patients (8), the varying
definitionsandclassificationsusedinliterature(9)andthelackofagenericMUPScodeinthe
international classification of primary care (ICPC) system, a commonly used classification
systeminprimarycare(10).OnlyafewsyndromescloselyrelatedtotheconceptofMUPSwitha
persistentcoursehaveaspecificICPCcode:irritablebowelsyndrome,chronicfatiguesyndrome,
fibromyalgiaand(persistentorrecurrent)lowbackpainwithoutradiation.
Inprimarycareresearch,routineelectronicmedicalrecords(EMRs)canbeusedforidentifying
subgroupsofpatientsatrisk.Thereisevidencethatsuggeststhatidentifyingsubgroupsatrisk
followed by panel management can improve quality of life, for example in the frail elderly
population (11,12). Only a few previous studies developed risk assessment or identification
modelsforMUPSusingEMRs,buteitherthemethoddidnotprovetobeeffectiveoritwasnot
directlysuitableforusageforprimarycare(13,14).
4
In the light of the need to identify potentially new risk variables for MUPS in large
heterogeneousEMRdatasets,more traditionalhypothesisdrivenapproachesmightnotbe the
mostsuitableones.Nowadays,relativelynewdataminingtechniquesareavailableforanalysing
large heterogeneous databases, such as EMR databases, to unravel relationships between
potentialriskvariablesandoutcomesandtogaindeeperunderstandingofthedata(15–17).In
thepresentobservationalstudyweaimedtoexplorealternativeapproachesindevelopingrisk
assessment models based on the analysis of primary care EMRs to enhance identification of
patientsatriskforpersistentMUPSusingtwodifferentstatisticaltechniques.
METHODS
Designandstudypopulation
We used an observational study design. We were able to use a database consisting of fully
anonymisedextractedroutineEMRsfrom22generalpracticecenters(156176patients)inthe
areaofUtrecht, TheNetherlands, covering a five-yearperiod (January2007-December2011).
All general practice centers represented in this dataset participate in the Julius General
Practitioners’Network(JGPN)andsharetheiranonymisedroutinehealthcaredatathroughthe
JGPNdatabase,connectedwiththeJuliusCenter,UniversityMedicalCenterUtrecht(UMCU)and
approvedbythemedicalethicalcommitteeoftheUMCU(file#99-240).
Alldataare codedusing the ICPC for registrationof symptomsanddiagnoses, theAnatomical
Therapeutic Chemical (ATC) classification system for drug prescription and various
measurement codes for biometry, interventions and referrals. All practice centers involved in
theJGPNroutinelyinformtheirenlistedpatientsthatEMRsareanonymisedandsharedthrough
theJGPNdatabaseandthatresearchiscarriedoutwiththiscodeddatabase.Individualpatients
whodonotwanttheirdatatobeusedforthispurposeareenabledtoopt-out.Sincepatientdata
arecompletelyanonymised,nofurtherinformedconsentornewapprovalofthemedicalethics
committeewasneeded.
5
Definitionofoutcome
SinceMUPSisnotaclearhomogeneousconditionordiagnosisandweaimedatdiscriminatinga
subgroupofpatientsat risk forpersistentMUPS,weoperationalized theconceptofpersistent
MUPS by choosing any combination of ICPC codes for irritable bowel syndrome (D93),
fibromyalgia(L18.01),chronicfatiguesyndrome(A04.01)andlowbackpainwithoutradiation
(L03) as thepotential outcome.We consideredpatients registeredwith anyof these codes to
closely represent the concept of persistent MUPS, since all four tend to develop into chronic
symptomatologywithoutstraightforwardpathophysiologicalexplanation.
Thedataset
ThedatabaseincludedindividualpatientrecordseachwiththefollowingroutineEMRvariables:
patient characteristics (gender, age), consultation characteristics (date of consultation, ICPC
code assigned per consultation, to which episode the consultation belongs), medication
(prescription date, type of medication using ATC code), referrals (date of referral to any
specialtytype)andlaboratoryresults(datesoftests,typeoftests,testresults,frequencyofeach
test per patient). The first occurrence of any of the four mentioned MUPS ICPC codes was
consideredtobethefirstregistrationofpersistentMUPSinourdefinition.Foreachpatientwith
persistentMUPS inourdefinition,alldataregisteredduring theyearprior to the firstdefined
ICPC codewere selected. For the patients in the dataset without one of the persistentMUPS
codes, a one-year period was randomly selected. As the number of patients with persistent
MUPSinthedatasetwassmall(n=7840),webalancedthedatasetbytakingarandomsampleof
patients without persistent MUPS (n=7988). From this balanced dataset, only patients aged
between18and65wereincludedintheanalyses(6398patientswithpersistentMUPSand5988
patientswithoutpersistentMUPS).
Constructingvariables
The one-year history per patient is event-based, while the construction of a risk assessment
modelrequiresasinglerecordperpatientthatsummarizestheeventsovertheentireyear. In
order tocreatesucha summarization,wederivedvariables fromtheeventsbycounting their
frequencywithinthegivenperiod.Theincludedvariablesaredemographicdata,summationsof
ICPC codes belonging to a chapter during the given period (both the number of unique ICPC
codes assigned from the chapter and the total number of consultations associated with each
chapter),numberofepisodes,numberofprescriptionsofmedicationsperATCgroup,aswellas
variables about laboratory results and referrals which are represented as the number of
referralsperspecialism.
6
Modelling
Weperformed a pre-selectionprocedure to reduce the number of variables to be used in the
modelling. In thispre-selection first thevariableswith less than10caseswereexcluded.This
wasdoneinthefulldataset.Second,becausethenumberofuniqueICPCcodesbelongingtoan
ICPC chapter and the number of consultations are obviously correlated, we only used the
number of consultations or contacts per chapter in the modelling procedure. In exploring
possibilities to generate an algorithm, we used two statistical methods to generate a risk
assessmentmodel for persistentMUPS. Firstwe used a relatively classical logistic regression
analysisapproachwithaforwardselectionprocedure.Second,weappliedarelativelyadvanced
decisiontreeanalysismorecommonlyusedindatamining.Thedecisiontreeanalysis isbased
onthemethodofrecursive-partitioninganalysis(18,19).Inthisanalysis,thetotalgroupissplit
into subgroupsbasedon the variable,whichdistinguishes thepersistentMUPS from thenon-
MUPS patients best. These subgroups are again partitioned, until no further significant
partitioningispossible.Incaseofcontinuouscandidateriskvariables,groupsweresplitatthe
cut-offpoint(s)withthehighestdiscriminativevalueforpersistentMUPS.Theterminalnodesin
thedecision treewereused toclassify thepatient’s risk forpersistentMUPS.For thedecision
treeanalysis,thechi-squaredautomaticinteractiondetector(CHAID)algorithmwasused(20),
with amaximum tree depth of 10 splits.Wedeveloped themodel in 80% of the dataset and
validatedthemodelintheremaining20%.Weassessedtheperformanceoftheriskassessment
models by computing the area under the curve (AUC) of the receiver operating characteristic
curve (ROC), which is also known as the C-statistic. All analyses were performed in SPSS
ModelerVersion16(IBM).
7
RESULTS
Thedataset
Table1showsdescriptiveinformationforthevariablesinthecomposeddataset,whichweused
todevelopourriskassessmentmodels.InthepersistentMUPSgroup62%werefemaleandthe
meanageofthegroupwas41years.Fortypercenthadoneepisode,26%hadtwoepisodesand
32%hadmorethantwoepisodes.MedicationfromATCchapterS(sensoryorgans)was
prescribedmostfrequentlyfollowedbymedicationfromchapterL(antineoplasticandimmune-
modulatingagents),R(respiratorysystem)andA(alimentarytractandmetabolism).Inthenon-
persistentMUPSgroups,57%werefemaleandthemeanageofthegroupwas40years.Fifty-
ninepercenthadzeroepisodes,24%hadoneepisodeand17%hadmorethanoneepisode.
MedicationfromATCChapterA(Alimentarytractandmetabolism)wasprescribedmost
frequently,followedbymedicationfromChapterSandLandR.Bothgroupswerereferredin
threepercentforadditionalexaminationortosecondarycarespecialisms.Mostreferralswere
notincludedinthepredictionmodellingbecausethenumberofpatientsreferredtosecondary
carespecialismswaslessthan10,toolowtoallowmeaningfulanalyses.
Table1.Descriptiveinformationregardingthevariablesusedtobuildpredictionmodels
forpersistentMUPS
Variabledescription PersistentMUPS
(n=6398)Non-persistent
MUPS(n=5988)Patient
Female,% 62 57Age,mean(SD) 41(13) 40(13)
Consultations
Totalnumberofepisodes,%
0 42 591 26 24
>1 32 17ICPCChapters,oneormorecontactsin%
A(GeneralandUnspecified) 14 23B(BloodandBloodformingorgans) 1 2D(Digestive) 9 10
F(Eye) 3 5H(Ear) 4 6K(Circulatory) 5 5
8
L(Musculoskeletal) 15 21
N(Neurological) 4 4P(Psychological) 10 8R(Respiratory) 14 18
S(Skin) 17 21T(Endocrine,metabolicandnutritional) 4 4U(Urology) 4 4
W(Pregnancy,childbirth,familyplanning) 6 8X(Femalegenitalsystemandbreasts) 8 10Y(Malegenitalsystem) 2 3
Z(Socialproblems) 3 3Referrals,oneormorein%
Total 3 3Imagingdiagnostics 1 1
Physiotherapy 1 0Laboratory 2 0ATCMedicationprescriptions,oneormorein%
A(Alimentarytractandmetabolism) 24 15B(Bloodandbloodformingorgans) 4 4
C(Cardiovascularsystem) 11 11D(Dermatologicals) 14 15
G(Genito-urinarysystemandsex-hormones) 11 13
J(Anti-infectivesforsystemicuse) 11 12H(Systemichormonalpreparations) 3 3L(Anti-neoplasticandimmune-modulating) 1 1
M(Musculoskeletalsystem) 24 10N(Nervoussystem) 21 16
P(Anti-parasiticproducts) 1 1R(Respiratorysystem) 16 16S(Sensoryorgans) 6 6
Productpp(numberofprescriptionproductswith
regardstomedication)58 57
Producttp(numberoftradeproductswithregardsto
medication)15 15
Productmp(numberofmedicalproductsfreely
available)16 9
Productcpp(numberofproductscustompreparedby
pharmacist)2 2
9
Totalmedication 66 64
Laboratory,oneormorein%
QuestionD(diagnosticquestions) 3 2
QuestionY(yes/noquestions) 2 2QuestionMC(multiplechoicequestions) 1 1QuestionM(measurements) 30 29
QuestiontypeFT(freetextquestions) 27 26Totallabresults 44 42
ICPCinternationalclassificationofprimarycare;ATCanatomicaltherapeuticclassification
Modelling
Table2showsthefinalriskassessmentmodelbasedonthelogisticregressionanalysis.Based
ontheWaldstatisticitcanbeseenthatthetotalnumberofepisodeswasthemost
discriminativevariableofthelogisticmodel,followedbymedicationforATCchapterM
(musculoskeletalsystem).Theoddsratioforthenumberofepisodes(i.e.1.486)indicatesthat
whenapatienthasonemoreepisode,theoddsofhavingpersistentMUPSbecomes1.486higher.
TheoddsratioformedicationforATCchapterMindicatesthatonemoremedicationinATC
chapterMisassociatedwitha1.935timeshigheroddsofhavingpersistentMUPS.Figure1(A
tillD)showsthemostoptimaldecisiontreebasedontheCHAIDalgorithm.Thefirstsplitinthe
decisiontreewasalsobasedthetotalnumberofepisodes.Thehighestprobabilityofhaving
persistentMUPS(figure1D),withaprobabilityofpersistentMUPSof96%wasobtainedvia
numberofepisodesmorethan3,numberofcontactsinICPCChapterA=zero,numberof
contactsinICPCChapterL=zeroandnumberofprescriptionproductsw.r.t.medication=zero.
ThelowestprobabilityofhavingpersistentMUPS(6%)wasobtainedbythefollowingsteps:
numberofepisodes=one,numberofcontactsinICPCChapterA=zero,numberofcontactsin
ICPCChapterLmorethanzeroandgender=male(figure1A).TheAUCcomputedforthefinal
logisticregressionpredictionmodelwas0.70andforthedecisiontree,theAUCwas0.81.The
validationprocedureofbothmodelsprovidedAUCvaluesof0.70forthelogisticregression
modeland0.78forthedecisiontree.Figure2showstheROCcurvesforthetworiskassessment
modelsinthevalidationset.
10
Table2.FinalpredictionmodelforpersistentMUPSbasedonalogisticregression
analysiswithaforwardselectionprocedure
Variabledescription Oddsratio Wald P-value
Female 1.253 21.649 <0.001
Age 1.014 51.334 <0.001
Numberofepisodes 1.486 311.645 <0.001
Numberofcontactsperchapter
D(Digestive) 0.974 6.228 0.013
F(Eye) 0.898 8.473 0.004
H(Ear) 0.957 3.202 0.074
L(Musculoskeletal) 0.912 63.265 <0.001
R(Respiratory) 0.981 4.516 0.034
S(Skin) 0.952 15.871 <0.001
W(Pregnancy,childbirth,familyplanning) 0.942 11.119 0.001
X(Femalegenitalsystemandbreasts) 0.912 23.169 <0.001
Y(Malegenitalsystem) 0.952 3.582 0.058
Z(Socialproblems) 0.889 7.833 0.005
Physiotherapy 9.747 24.085 <0.001
Numberofreferrals 0.754 5.667 0.017
Medication
A(Alimentarytractandmetabolism) 1.243 51.479 <0.001
J(Anti-infectivesforsystemicuse) 0.859 8.238 0.004
L(Anti-neoplasticand
Immune-modulatingagents) 0.578 8.438 0.004
M(Musculoskeletalsystem) 1.935 173.626 <0.001
N(Nervoussystem) 1.056 9.862 0.002
R(Respiratorysystem) 1.069 6.756 0.009
Productpp(numberofprescriptionproductswith
regardstomedication) 0.934 29.336 <0.001
Producttp(numberoftradeproductswith
regardstomedication) 0.912 21.054 <0.001
QuestiontypeFT(freetextquestions) 1.037 11.905 0.001
Numberoflaboratoryresults 0.988 48.002 <0.001
ICPCinternationalclassificationofprimarycare;ATCanatomicaltherapeuticclassification
Figure1.(AtoD)MostoptimalriskassessmentdecisiontreeforpersistentMUPS
11
1A
1B
12
1C
1D
13
MUPSmedicallyunexplainedphysicalsymptoms;ICPCinternationalclassificationofprimary
care;ATCanatomicaltherapeuticclassification;#episodes=numberofepisodes;*_contacts=
numberofcontactsinICPCchapter*;ATC_*;numberofmedicationsinATCchapter*;Product
pp=numberofprescriptionproductswithregardstomedication;#labresults=numberof
laboratoryresults
Figure2.Receiveroperatingcharacteristicsforthelogisticregressionmodelanddecision
treeonvalidationset
DISCUSSION
Mainfindings
Theobjectiveofourstudywastodevelopriskassessmentmodelsthatpotentiallycouldbeused
toidentifypatientsatriskforpersistentMUPSfromroutineprimarycareEMRs.Wecompared
the results froma relatively classical logistic regressionprocedurewith those fromadecision
tree model. We were able to assess the risk of persistent MUPS by using the final logistic
regressionmodelwithamoderateAUC,whilethedecisiontreeanalysesperformedsomewhat
betterwith amoderate to goodAUC.Thevariablemost clearlydiscriminating inbothmodels
wasthetotalnumberofepisodesamongpatientswithpersistentMUPS.
14
Comparisonwithliterature
Only fewstudiesaimed todevelopa riskassessmentor identificationmodel forpatientswith
MUPS from EMRs. They showed that there is no simple or straightforward method. Smith’s
model, containing gender, total number of consultations andpercentage of consultationswith
MUPS potential, initially predicted MUPS with an AUC of 0.90. Upon external validation, the
valuedecreasedto0.78(13).MorrissaimedtoestimatetheprevalenceofMUPSfromprimary
care EMRs (14). Theirmodels, related toMUPS and severeMUPS, hadAUCs of 0.70 and 076,
respectively.However,themodelsalsorequiredfurthervalidationandtheyconcludedthatthe
modelsdidnotyetprove tobeeffective forclinical screeningpurposesdue to lowsensitivity.
Rosmalenetalconductedalatentclassanalysistoidentifypatientswithsomatization,aconcept
closely related toMUPS, and found that a simple symptom count could support the diagnosis
somatization(21).Ourstudyisessentiallymoreorlessaconfirmationofthesepreviousstudies
because we also found that the number of episodes was the most important discriminative
variableforpersistentMUPSinbothourmethods.
Althoughnotentirelycomparablewithourstudy,otherstudiesidentifyingsubgroupsofMUPS
showed varying results. Tiandeveloped a risk assessmentmodel for chronic painwherepain
scores,painmedicationandspecificinternationalstatisticalclassificationofdiseasesandrelated
health problems (ICD-9) scores were combined. They found a high AUC of 0.98 (22). Viniol
performedk-meansclusteranalysisandstudiedpatientswithchroniclowbackpain(23).They
found three clusters specifically characterized by age and psychological variables such as
distress.Harkness et al usedREAD codes, standard clinical terminology system codesused in
generalpracticeintheUnitedKingdom,toidentifypatientsatriskforirritablebowelsyndrome.
Theyconcludedthat theresultswerea largeunderestimationof thecommunityprevalenceof
irritablebowelsyndromeflawingtheiranalysis(24).
Strengthsandlimitations
Themainstrengthofourstudy is thatweuseda largedatabasewithroutineEMRs, reflecting
what is registered indailyclinicalpractice.Tobe included in theregistration,apatienthas to
consulttheGPandtheGPhastocodethecomplaintasasymptomoradiagnosis.Byusingdata
frommanypracticesandfrommanypatients,theriskofselectionbiasandinformationbiasis
reduced and the influence of loss of detail is reduced. This study potentially provides new
insights inresearchmethodologybecauseofapplyingadvancedbigdataanalysisonrelatively
rawdatathataregatheredfromroutineprimarycare.
15
Ourresultsshouldbeinterpretedinthelightofsomelimitations.Aswithallroutinelycollected
EMRs, we could not verify the correctness and completeness of the coded information and
different GPs have different coding patterns. Also, by using four ICPC codes to conceptualize
persistentMUPS,wemighthavemissedsomepotentialothercharacteristicsofMUPSpatients.
But because GPs apply these codes only after medical evaluation thereby ruling out other
diseases,we are confident thatwe really identified patients resembling thosewith persistent
MUPSasthereference,thusenablingustoassesstheriskofotherstoevolvetothatcondition.
Anotherlimitationisthat,bybalancingthedatasetinordertoorganizeouranalysisstatistically
efficient, theprevalenceofMUPSchangedtomoreor less50%,whichpotentiallyhampersthe
generalizability of the results in a way that we look at relative risks and not absolute risks.
Finally, inthepresentstudy,weusedtheAUCtoquantifythequalityofthepredictionmodels.
The AUC is an indicator for the internal validity of the model. External validity should be
assessedby testing thealgorithms inanotherEMRdataset. In thepresentstudywemimicked
external validation by performing a cross-validation of the risk assessment models in an
independentpartoftheoriginaldataset.ForthelogisticregressionmodeltheAUCremainedthe
same, while for the decision tree, the AUC became slightly lower. This is not a big surprise,
becausethebuildingofthetreeishighlydatadriven.Nevertheless,thevalidationAUCremained
moderatetogood.
Implicationsforclinicalpracticeandresearch
Manueletalwhostatedthattheuseofriskassessmentmodelsinclinicalmedicinehasincreased
overthepast20yearsandthattheyhavethepotentialtosupporttheprocessofdecisionmaking
andmanagement of population health (16). Christensen et al confirmed that recognition and
managementbyGPsof commonprimary carementalhealthdisorders includingMUPScanbe
positivelyinfluencedbyusingscreeningquestionnaires(25).Riskassessmentamongsubgroups
ofpatientspotentiallydevelopingpersistentMUPScouldbeausefulfirststepforimplementing
panel management. By deriving a software algorithm from the risk assessment model to be
applied in routine primary care, the tendency to move towards persistent MUPS could be
increased.PatientsatriskcouldbeapproachedproactivelybytheirGPs.Andwhenapatientat
risk for persistent MUPS presents, GPs are potentially enabled to adjust their consultation
strategies by paying attention to empathic communication, exploring the patients’ functional
limitations and providing rational explanations before evolving targeted interventions in
consultationwiththepatient(26,27).Infutureresearch,firstthemodelshavetobeexternally
validated and a sensitivity analyses should be performedwhere the ratio between persistent
MUPSandnon-MUPSisvariedtoincreasethegeneralizability.
16
CONCLUSIONS
We explored the possibilities of assessment of patients’ individual risks for persistent MUPS
based on routineEMRs andwedeveloped twomoderate to goodperforming risk assessment
models, partly by using relatively new datamining techniques.While our algorithms require
further validation and fine-tuning, they provide a starting point fromwhichGPs could evolve
towardsamoreproactiveandstructuredMUPSmanagementtogetherwiththeirpatients.
ACKNOWLEDGEMENTS
TheauthorswouldliketothankthegeneralpractitionersfromtheJGPNUMCUandtheir
patientsforsharingtheiranonymouselectronicmedicalrecords.
17
REFERENCES
1. JacksonJL,PassamontiM.Theoutcomesamongpatientspresentinginprimarycarewitha
physicalsymptomat5Years.JGenInternMed.2005;20:1032–7.
2. BarskyAJBorusJF.Somatizationandmedicalizationintheeraofmanagedcare.JAMA.
1995;274:1931–4.
3. VerhaakPFM,MeijerSA,VisserAP,etal.Persistentpresentationofmedicallyunexplained
symptomsingeneralpractice.FamPract.2006;23:414–20.
4. BarskyAJOravE.Somatizationincreasesmedicalutilizationandcostsindependentof
psychiatricandmedicalcomorbidity.ArchGenPsychiatry.2005;62:903–10.
5. NeuwirthE,B,SchmittdielJA,TallmanK,etal.UnderstandingPanelManagement:A
comparativestudyofanemergingapproachtopopulationcare.PermJ.2007;11:12–20.
6. BurtonC.Beyondsomatisation:areviewoftheunderstandingandtreatmentofmedically
unexplainedphysicalsymptoms.BrJGenPract.2003;53:231–9.
7. VanDesselN,denBoeftM,vanderWoudenJC,etal.Non-pharmacologicalinterventionsfor
somatoformdisordersandmedicallyunexplainedphysicalsymptoms(MUPS)inadults.
CochraneDatabaseSystRev.2014;11:CD011142
8. AamlandA,MalterudK,WernerEL.Patientswithpersistentmedicallyunexplainedphysical
symptoms:adescriptivestudyfromNorwegiangeneralpractice.BMCFamPract.
2014;15:107.
9. FinkP,RosendalM,OlesenF.Classificationofsomatizationandfunctionalsomatic
symptomsinprimarycare.AustNZJPsychiatry.2005;39:772–81.
10. LambertsH,WoodM,Hofmans-OkkesIM.Internationalprimarycareclassifications:the
effectoffifteenyearsofevolution.FamPract.1992;9:330–9.
11. Silow-CarrollS,EdwardsJN,RodinD.Usingelectronichealthrecordstoimprovequalityand
efficiency:theexperiencesofleadinghospitals.IssueBriefCommonwFund.2012;17:1–40.
12. LooTS,DavisRB,LipsitzLA,etal.Electronicmedicalrecordremindersandpanel
managementtoimproveprimarycareofelderlypatients.ArchInternMed.2011;171:1552–
8.
13. SmithRC,GardinerJC,ArmattiS,etal.Screeningforhighutilizingsomatizingpatientsusing
apredictionrulederivedfromthemanagementinformationsystemofanHMO:a
preliminarystudy.MedCare.2001;39:968–78.
14. MorrissR,LindsonN,CouplandC,etal.Estimatingtheprevalenceofmedicallyunexplained
symptomsfromprimarycarerecords.PublicHealth.2012;126:846–54.
15. CiosKJ,WilliamMooreG.Uniquenessofmedicaldatamining.ArtifIntellMed.2002;26:1–24.
16. ManuelDG,RosellaLC,HennessyD,etal.Predictiveriskalgorithmsinapopulationsetting:
18
anoverview.JEpidemiolCommunityHealth.2012;66:859–65.
17. ParvathiI,SiddharthR.Surveyondataminingtechniquesforthediagnosisofdiseasesinthe
medicaldomain.IJCSIT.2014;5:838-846.
18. BreimanL,FriedmanJH,OlshenRA,etal.Classificationandregressiontrees.BocaRaton.
ChapmanHall.1984.
19. LemonSC,RoyJ,ClarkMA,etal.Classificationandregressiontreeanalysisinpublichealth:
methodologicalreviewandcomparisonwithlogisticregression.AnnBehavMed.
2003;26:172–81.
20. BhartiyaS,MehrotraD.ApplyingCHAIDalgorithmtoinvestigatecriticalattributesof
securedinteroperablehealthdataexchange.IntJElectronHeal.2015;8:25–50.
21. RosmalenJGM,TakLM,deJongeP.Empiricalfoundationsforthediagnosisofsomatization:
implicationsforDSM-5.PsycholMed.2011;41:1133–42.
22. TianTY,ZlatevaI,AndersonDR.Usingelectronichealthrecordsdatatoidentifypatients
withchronicpaininaprimarycaresetting.JAmMedInformAssocJAMIA.2013;20:e275–
280.
23. ViniolA,JeganN,HirschO,etal.Chroniclowbackpainpatientgroupsinprimarycare-A
crosssectionalclusteranalysis.BMCMusculoskeletDisord.2013;14:294
24. HarknessEF,GrantL,O’BrienSJ,etal.Usingreadcodestoidentifypatientswithirritable
bowelsyndromeingeneralpractice:adatabasestudy.BMCFamPract.2013;14:183.
25. ChristensenKS,ToftT,FrostholmL,etal.Screeningforcommonmentaldisorders:whowill
benefit?Resultsfromarandomisedclinicaltrial.FamPract.2005;22:428–34.
26. SalmonP,RingA,DowrickCF,etal.Whatdogeneralpracticepatientswantwhenthey
presentmedicallyunexplainedsymptoms,andwhydotheirdoctorsfeelpressurized?J
PsychosomRes.2005;59:255–60.
27. BurtonC,LucassenP,AamlandA,etalExplainingsymptomsafternegativetests:towardsa
rationalexplanation.JRSocMed.2015;108:84–8.