Page 1
Funded by the Robert Wood Johnson Foundation
Strategies to Achieve Alignment, Collaboration, and Synergy across Delivery and Financing Systems
Using Predictive Modeling to Identify Patients Who Need Social Services
Research In Progress WebinarWednesday, January 10, 2017
12:00-1:00 pm ET/ 9:00 am-10:00 pm PT
Page 2
AgendaWelcome: CBMamaril,PhD
ResearchFaculty,RWJFSystemsforAction NationalCoordinatingCenter,UniversityofKentuckyCollegeofPublicHealth
Presenter: JoshuaR.Vest,PhD,MPHDirector,CenterforHealthPolicyAssociateProfessor,HealthPolicy&ManagementIndianaUniversityRichardM.FairbanksSchoolofPublicHealthatIUPUIjoshvest@iu.edu
CommentarySpeaker:MichaelShafer,PhDDirectorCenterforAppliedBehavioralHealthPolicyProfessorArizonaStateUniversitymichael.shafer@asu.edu
QuestionsandDiscussion:ModeratedbyDr.Mamaril
2
Page 3
Presenter
JoshuaR.Vest,PhD,MPHDirectorCenterforHealthPolicyAssociateProfessorHealthPolicy&ManagementIndianaUniversityRichardM.FairbanksSchoolofPublicHealthatIUPUIjoshvest@iu.edu
3
Page 4
Commentary Speaker
4
MichaelShafer,PhDDirectorCenterforAppliedBehavioralHealthPolicyProfessorArizonaStateUniversitymichael.shafer@asu.edu
Page 5
Webinars
5
Archives http://systemsforaction.org/research-progress-webinars
UpcomingWednesday, January 24, 12-1pm ET/ 9-10am PTOPTIMIZING GOVERNMENTAL HEALTH AND SOCIAL SPENDING INTERACTIONSJohns Hopkins University Bloomberg School of Public HealthPrincipal Investigators: Beth Resnick, DrPH, MPH, and David Bishai, MD, MPH, PhD
Wednesday, February 7, 12-1pm ET/ 9-10am PTSTRENGTHENING THE CARRYING CAPACITY OF LOCAL HEALTH AND SOCIAL SERVICE NETWORKSTrailhead Institute in ColoradoPrincipal Investigators: Danielle Varda, PhD, and Katie Edwards, MPA
Wednesday, February 21, 12-1pm ET/ 9-10am PTLINKING MEDICAL HOMES TO SOCIAL SERVICE SYSTEMS FOR MEDICAID POPULATIONSNational Committee for Quality AssurancePrincipal Investigators: Sarah Scholle, DrPH, and Keri Christensen, MS
Page 6
Thank you for participating in today’s webinar!
For more information about the webinars, contact:[email protected]
111 Washington Avenue #201, Lexington, KY 40536859.218.2289
www.systemsforaction.org
6
Page 7
Speaker Bios
7
Joshua R.Vest,PhD,MPH,isahealthservicesresearcherwithinterestsinorganizationaldeterminantsandeffectivenessofhealthinformationtechnologyandsystems,specificallytheadoption,utilization,andpolicyissuesoftechnologiesthatfacilitatethesharingofpatientinformationbetweendifferentorganizations.Heiswidelypublishedandhisworkhasemployedavarietyofresearchtechniquesfromlargescaledatabaseanalyses,togeographicalinformationsystemmapping,tosurveyresearch,toqualitativefocusgroupsandinterviews.Asaformerlocalpublichealthpractitioner,Dr.Vesthasaparticularinterestineffectivepublichealthinformationsystemsincludingtheroleofinformationtechnologygovernancestructuresonlocalpublichealthdepartments'adoptionofinformationtechnologyandsystems,thestructureofstateandlocalpublichealthinformationsystems,aswellasanevaluationofemailinterventiontoimprovediseasenotificationefforts.
MichaelS.Shafer,Ph.D.,isaprofessorintheSchoolofSocialWorkatArizonaStateUniversity’sCollegeofPublicServiceandCommunitySolutionswherehealsoholdsaffiliateappointmentsintheCenterforHealthInformationResearchandtheSchoolofCriminologyandCriminalJustice. Dr.ShaferisthefoundingdirectoroftheCenterforAppliedBehavioralHealthPolicywhichhas,forthepast25years,conductedcuttingedgeresearchontheadoptionandimplementationofinnovativepracticesinbehavioralhealthcare. Dr.Shaferhasauthoredmorethan40peer-reviewedresearcharticlesandgeneratedmorethan$45millioningrantsandcontractsthattargetcapacitybuildingandinnovationinbehavioralhealthservices. Dr.ShaferearnedhisPh.D.inEducationin1988fromVirginiaCommonwealthUniversity. Hehasreceivednumerousawardsandcitations,includingrecognitionfromtheU.S.DepartmentofJusticeforthedevelopmentofcrisisinterventiontrainingforlawenforcementpersonnel. Dr.Shaferisafrequentcontributortoprofessionalliteratureandheconsultswithbehavioralhealthagenciesthroughoutthecountry.
Page 8
Usingpredictivemodelingtoidentifypatientswhoneedsocialservices
JoshuaRVest,PhD,MPHDirector,CenterforHealthPolicy
AssociateProfessor,HealthPolicy&ManagementIndianaUniversityRichardMFairbanksSchoolofPublicHealthatIUPUI
AffiliatedScientist,RegenstriefInstitute,Inc.
ThisworkwassupportedbytheRobertWoodJohnsonFoundationthroughtheSystemsforActionNationalCoordinatingCenter,ID73485.
Page 9
Acknowledgements
IndianaUniversity• Suranga N.Kasthurirathne• Nir Menachemi• PaulKHalverson• ShaunJ.Grannis• JenniferWilliams• KarenComer
EskenaziHealth• DawnHaut• JenniferFerrell
• DonnaBurke• AlishaJessup
Kasthurirathne etal2017doi:10.1093/jamia/ocx130
Page 10
Predictivemodelinginhealthcare:statisticalapproachestoidentifyingpatientsathighrisk(morelikely)fornegativeoutcomes
alliconsfromflaticon.com
Diagnosis&Utilization
Page 11
Predictivemodelingiswidelyapplied...
Page 12
Limitationsofcurrentpredictivemodeling
• LimitedtoEHRorclaimsdata• Socialdeterminantsoftenabsent• Oftensingle-sitedata
• Focuson“toolate”outcomes(reactivenotproactive)
• Don’tprovideinsightsintowhatservicespatientsshouldget
Page 13
Objective1:Evaluatepredictivemodelsthatusecombinationsofclinical,socioeconomic,andpublichealthdata
Diagnosis&UtilizationEnvironmental&socialcontext
Neighborhoodhealthcontext
HealthBehaviors&System-widehealthdata
Page 14
Diagnosis&UtilizationEnvironmental&socialcontext
Neighborhoodhealthcontext
HealthBehaviors&System-widehealthdata
Dataonhealthstatus
Page 15
Diagnosis&UtilizationEnvironmental&socialcontext
Neighborhoodhealthcontext
HealthBehaviors&System-widehealthdata
Dataondriversofhealth
Page 16
Diagnosis&UtilizationEnvironmental&socialcontext
Neighborhoodhealthcontext
HealthBehaviors&System-widehealthdata
Acomprehensivepatientview
Page 17
s
HEALTH
Economic&socialopportunities&resources
Living&workingconditionsinhomes&communities
Medicalcare
Personalbehavior
Frameworkfororganizingthefactorsincludedinriskidentificationtool
“Social Determinants of Health Model” by Braveman et al (2011) Annu. Rev. Public Health, 32:381-398
Theonlydataincludedinmostpredictionmodels
Whatweareadding
Page 18
Fitday.com
Objective2:Contributionofthesedataonthenoveloutcomeofreferralstosocialservices
http://cchci.org/_services/behavioral-health/
www.chelseajew
ish.org/celebrating-social-w
ork-month-senior-care/
Pplay.google.com
guide.berkeley.edu/undergraduate
dmkelite.com
usf.edu
Page 19
Fitday.com
Toberesponsivetonewpaymentstrategies,healthcareorganizationsintheUSarebeginningtoofferthesenon-medicalservices.
http://cchci.org/_services/behavioral-health/
www.chelseajew
ish.org/celebrating-social-w
ork-month-senior-care/
Pplay.google.com
guide.berkeley.edu/undergraduatedmkelite.com
usf.edu
Page 20
Objective1:Evaluatepredictivemodelsthatusecombinationsofclinical,socioeconomic,andpublichealthdata.
Objective2:Contributionofthesedataonthenoveloutcomeofreferralstosocialservices.
Page 21
Approach
Comparetheperformanceofriskpredictionmodelswith:
1)clinicaldataonly
2)clinicaldatawith community-levelsocioeconomic&publichealthindicators
Page 22
Setting&sample
• EskenaziHealthoutpatientclinics– Indianapolis
safety-netprovider(formedicalindigent)
– urbanpopulation– allsocialservicesoffered
onaco-locatedbasis(noreferralstootherorganizations)
• 84,317adultpatients– atleast1outpatientvisit
between2011-2016
Demographics
Age (mean, sd) 43.9(15.6)
Male gender 35.1Race / ethnicityWhite, non-Hispanic 25.2African American, non-Hispanic 37.2Hispanic 19.5DiagnosesHypertension 38.7Asthma 7.9Cancer 7.6COPD 9.5Depression 19.0Diabetes 20.3Substance abuse 15.1Tobacco use 21.3
Sampledemographics
Page 23
Data&measures(outcome)
Referraltosocialservices– Socialwork– Dietitian– Mentalhealth– Allotherservices(duetolowfrequency)
Datasources– EskenaziEHRbillingandencounterdata– schedulingsystemdata(includingkept,missed,&cancelledappointments)– unstructuredEHRordersandnotes
Page 24
s
HEALTH
Economic&socialopportunities&resources
Living&workingconditionsinhomes&communities
Medicalcare
Personalbehavior
Data&measures(predictors)
“Social Determinants of Health Model” by Braveman et al (2011) Annu. Rev. Public Health, 32:381-398
Page 25
Living&workingconditionsinhomes&communities
Medicalcare
Personalbehavior
Data&measures(predictors)
“Social Determinants of Health Model” by Braveman et al (2011) Annu. Rev. Public Health, 32:381-398
• Diagnoses• Asthma• Coronaryarterydisease• Chronickidneydisease• Congestiveheartfailure• COPD• Stroke/cerebrovascularaccident• Depression• Diabetes• Hypertension• Ischemicvasculardisease• Obesity• Pregnancy• ….
• EDvisits(number)• Inpatientadmissions• PCPvisits• Mentalillness
• Smoking• Substanceabuse
EHR
Page 26
• IndianaNetworkforPatientCare
• US’oldestHIE• StartedatRegenstriefInstitutein1995
• Oneofthenation’slargest• >80hospitals’medicalrecords• 17.2millionindividualpatients• 4.6billionclinicalobservations• 165milliontextreports• Over68%ofIndianapopulationcapturedin2014
• Datainclude:• admissionanddischarge• labreports• Microbiology• Pathology• Radiology• Cardiology• EKGdata
Page 27
s
Economic&socialopportunities&resources
Living&workingconditionsinhomes&communities
Medicalcare
Personalbehavior
Data&measures(predictors)
“Social Determinants of Health Model” by Braveman et al (2011) Annu. Rev. Public Health, 32:381-398
• Employmentrates• Taxdelinquentproperties• Crimeindices• Educationrates• Voterparticipation• Income
• Smokingprevalence• Perceivedsafety• Mortalityrates• Infantmortalityrates• Maternalsmoking• Overweight/obesity
prevalence• Walkability
Fromsurveyorcensusdataandlinkedbygeolocation.
Page 28
Frameworkfororganizingthefactors
kff.org/disparities-policy/issue-brief/beyond-health-care-the-role-of-social-determinants-in-promoting-health-and-health-equity/
Page 29
Analyticapproach:performanceofpredictionmodelswithnoveldata
1) Clinicaldataonly(41variables)2) Clinicalplussocioeconomic&
publichealth(48variables)Splitsamplesfortraining&testing
Randomforestclassificationalgorithm
Modeleachoutcometwice.
Page 30
Prevalenceofsocialservicereferralneed
Type of service %
Any service 53.0
Mental health 18.5
Social work 8.7
Dietitian 32.6
Other services 20.0
Page 31
Predictionforsocialservicesreferralswasinthe“useful”range.
Clinicaldata Clinical +socioeconomic&publichealth
Anyreferral 0.745 0.741Mentalhealth 0.785 0.778Socialwork 0.731 0.714Dietitian 0.743 0.730Otherreferral 0.711 0.708
AreaundertheROCcurvevaluesforeachdecisionmodel
Consistentwithperformanceofmodelson:• Mortality• Readmissions• Diseasedevelopment• Carecoordinationneed
Page 32
Socioeconomic&publichealthdatadidnotcontributesignificantly.
Clinicaldata Clinical +socioeconomic&publichealth
Anyreferral 0.745 0.741Mentalhealth 0.785 0.778Socialwork 0.731 0.714Dietitian 0.743 0.730Otherreferral 0.711 0.708
AreaundertheROCcurvevaluesforeachdecisionmodel
Page 33
Socioeconomic&publichealthdatadidnotcontributesignificantly.
Page 34
Limitations
• Socioeconomicmeasuresataggregatelevel– smallgeographicarea,butstillaggregate– limitedgeographicvariationbecauseonlywithinasingleurbanarea– individuallevelmeasuresgenerallyunavailablefromEHRs
• Highneed,vulnerablepopulation– limitedgeneralizability– probablylotsofunmetneed
• Allserviceswereco-locatedwithprimarycare– Maynotapplytoreferralstooutsideservices/otherorganizations
• Noassessmentwhetherornotthereferralwasappropriateorappointmentwaskept
Page 35
Predictivemodelsforreferralstosocialservicesarecurrentlylive.
5:00AM
Beforeclinicsopen
Page 36
Impactofpredictedmodelsonreferralratescurrentlybeingevaluated.
August September October November December JanuaryJulyJune
Next3cliniclocationsliveBaseline
3cliniclocationsliveBaseline
Last3cliniclocationsliveBaseline
Page 37
Usingpredictivemodelingtoidentifypatientswhoneedsocialservices.
• Indicationsthatpredictivemodelingforsocialservicesmaybeuseful– modelsleveragedEHR andHIE data– performancecouldbeimproved,but
consistentwithliterature• Socioeconomic&publichealthmeasures
(attheaggregatelevel)didnotimprovemodelperformance
JoshuaRVest,PhD,[email protected]