john billings: towards evidence based policy making
TRANSCRIPT
January 14, 2010
New York UniversityRobert F. Wagner Graduate School of Public Service
TOWARDS EVIDENCE BASEDPOLICY MAKINGEXPERIENCE OF THE
CHRONIC ILLNESS DEMONSTRATION PROGRAMFOR NEW YORK MEDICAID PATIENTS
WHAT I’M GOING TO TALK ABOUT
• The application of predictive modeling in an incredibly challenging environment– The U.S. health care “system”– A subpopulation of the New York Medicaid program
• An evidence based approach to policy making/program design (that almost/sort of got it right)
• What we hope to learn along the way
AN UNUSUAL CONFLUENCEOF CIRCUMSTANCES
• A policy wonk frenzy about the small number of high cost patients responsible for substantial portion of costs
50%
19%
25%
12%
15%
43%
10%26%
0%
20%
40%
60%
80%
100%
Enrollees Expenditures
MEDICAID ENROLLEES AND EXPENDITURES
BY ENROLLMENT GROUP – U.S. 2006
Elderly
Blind andDisabled
Adults
Children
Source: Kaiser Commission on Medicaid and the Uninsured – 2009.
59 Million $268 Million
NEW YORK MEDICAIDAdult Disabled – Non-Mandatory Managed Care
3.0%
30.0%
7.0%
25.9%
10.0%
17.0%80.0%
27.1%
0%
20%
40%
60%
80%
100%Pe
rcen
t of T
otal
Patients Expenditures
Source: NYU Center for Health and Public Service Research, UHF, NYSDOH, 2006.
72.9%
AN UNUSUAL CONFLUENCEOF CIRCUMSTANCES
• A policy wonk frenzy about the small number of high cost patients responsible for substantial portion of costs
• An emerging body of literature about predictive modeling for high cost patients
AN UNUSUAL CONFLUENCEOF CIRCUMSTANCES
• A policy wonk frenzy about the small number of high cost patients responsible for substantial portion of costs
• An emerging body of literature about predictive modeling for high cost patients
• A change in political administration at the state level, with infusion of some pretty smart people
AN UNUSUAL CONFLUENCEOF CIRCUMSTANCES
• A policy wonk frenzy about the small number of high cost patients responsible for substantial portion of costs
• An emerging body of literature about predictive modeling for high cost patients
• A change in political administration at the state level, with infusion of some pretty smart people
• Pre-economic crisis/panic/kerfuffle
AN UNUSUAL CONFLUENCEOF CIRCUMSTANCES
• A policy wonk frenzy about the small number of high cost patients responsible for substantial portion of costs
• An emerging body of literature about predictive modeling for high cost patients
• A change in political administration at the state level, with infusion of some pretty smart people
• Pre-economic crisis/panic/kerfuffle
• State legislature authorization for a demonstration [Chronic Illness Demonstration Project – CIDP]
AN UNUSUAL CONFLUENCEOF CIRCUMSTANCES
• A policy wonk frenzy about the small number of high cost patients responsible for substantial portion of costs
• An emerging body of literature about predictive modeling for high cost patients
• A change in political administration at the state level, with infusion of some pretty smart people
• Pre-economic crisis/panic/kerfuffle
• State legislature authorization for a demonstration [Chronic Illness Demonstration Project – CIDP]
• The federal authorities go along
A SOMEWHAT IDEALIZEDDESCRIPTION OF THE APPROACH
TO THE PROBLEM
Step 1: See if you can develop a predictive model to identify patientsfor whom you think you can do something
A SOMEWHAT IDEALIZEDDESCRIPTION OF THE APPROACH
TO THE PROBLEM
Step 1: See if you can develop a predictive model to identify patientsfor whom you think you can do something
Step 2: Learn as much as you can about these patients to helpin designing the intervention(s)- Use available administrative data- Apply algorithm to real patients – interview a sample of these
patients (and their providers, families, caregivers, etc.)
A SOMEWHAT IDEALIZEDDESCRIPTION OF THE APPROACH
TO THE PROBLEM
Step 1: See if you can develop a predictive model to identify patientsfor whom you think you can do something
Step 2: Learn as much as you can about these patients to helpin designing the intervention(s)- Use available administrative data- Apply algorithm to real patients – interview a sample of these
patients (and their providers, families, caregivers, etc.)
Step 3: Implement/evaluate demonstration projects based on thisinformation- Pilot with a quasi-experimental design (intervention/control)- Conduct “formative” evaluation during early phases of
implementation- Assess impact of intervention on outcomes/utilization
A SOMEWHAT IDEALIZEDDESCRIPTION OF THE APPROACH
TO THE PROBLEM
Step 1: See if you can develop a predictive model to identify patientsfor whom you think you can do something
Step 2: Learn as much as you can about these patients to helpin designing the intervention(s)- Use available administrative data- Apply algorithm to real patients – interview a sample of these
patients (and their providers, families, caregivers, etc.)
Step 3: Implement/evaluate demonstration projects based on thisinformation- Pilot with a quasi-experimental design (intervention/control)- Conduct “formative” evaluation during early phases of
implementation- Assess impact of intervention on outcomes/utilization
Step 4: Disseminate results/Scale up if it works
A SOMEWHAT IDEALIZEDDESCRIPTION OF THE APPROACH
TO THE PROBLEM
Step 1: See if you can develop a predictive model to identify patientsfor whom you think you can do something
Step 2: Learn as much as you can about these patients to helpin designing the intervention(s)- Use available administrative data- Apply algorithm to real patients – interview a sample of these
patients (and their providers, families, caregivers, etc.)
Step 3: Implement/evaluate demonstration projects based on thisinformation- Pilot with a quasi-experimental design (intervention/control)- Conduct “formative” evaluation during early phases of
implementation- Assess impact of intervention on outcomes/utilization
Step 4: Disseminate results/Scale up if it works
Evid
ence
d-ba
sed
man
agem
ent/p
olic
y m
akin
g
THE PREDICTIVE MODELINGALGORITHM DEVELOPMENT
• Take 5 years of claims data or hospital/ED records
• Look back from the 4th year of the data at prior utilization and diagnostic history
• Apply logistic regression techniques utilizing these data to predict patients at high risk for re-hospitalization
• Learn as much as possible about the characteristics of these patients from the data
BASIC APPROACH FOR DEVELOPMENTOF RISK PREDICTION ALGORITHM
IndexQuarters
Year 4 Year 5Year 3Year 2Year 1Q1 Q2 Q3 Q4
• Prior hospital utilization– Number of admissions– Intervals/recentness
• Prior emergency department utilization• Prior outpatient utilization/claims
– By type of visit (primary care, specialty care, substance abuse, etc)– By service type (transportation, home care, personal care, etc)
• Diagnostic information from prior hospital utilization– Chronic conditions (type/number)– Hierarchical grouping (HCCs)
• Prior costs– Pharmacy– DME– Total
• Patient characteristics: Age, gender, race/ethnicity• Predominant hospital/primary care provider characteristics
BASIC APPROACHTYPES OF VARIABLES IN ALGORITHM
0
5,000
10,000
15,000
20,000
25,000
30,000
35,000
40,000
45,000
50,000
40 45 50 55 60 65 70 75 80 85 90 95
CASE FINDING ALGORITHMNUMBER OF PATIENTS IDENTIFIED CIDP ELIGIBLE - MODEL DEVELOPMENT RUN
TOTAL FLAGGED
CORRECTLY FLAGGED
Risk Score Threshold
Patie
nts
Iden
tifie
d
FalsePositives
33%
FalsePositives
15%False
Positives7%
CHARACTERISTICS OF PATIENTS FLAGGEDBY CASE FINDING ALGORITHMCIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Demographic Characteristics
Risk Score> 50
Risk Score> 75
Risk Score> 90
AllPatients
N 33,363 8,713 2,176 64,446
Age 45.1 44.8 44.3 47.6Female 43.9% 38.5% 34.7% 49.7%
NYC Fiscal County 72.2% 80.0% 84.4% 69.1%
White 28.2% 23.6% 22.9% 32.7%Black 40.7% 48.1% 49.4% 33.1%Hispanic 15.0% 14.2% 12.2% 14.6%Other/Unknown 16.1% 14.2% 15.4% 19.5%
CHARACTERISTICS OF PATIENTS FLAGGEDBY CASE FINDING ALGORITHMCIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Diagnoses Reported in Claims Records
Risk Score> 50
Risk Score> 75
Risk Score> 90
AllPatients
Cereb Vasc Dis 4.9% 6.3% 8.1% 4.7%AMI 6.2% 9.5% 12.9% 5.2%Ischemic Heart Dis 22.5% 28.8% 35.5% 20.3%Congestive Heart Failure 16.4% 22.6% 26.9% 12.2%Hypertension 50.1% 58.3% 64.1% 48.2%Asthma 34.8% 45.7% 50.5% 26.2%COPD 23.5% 33.8% 42.3% 17.4%Diabetes 28.8% 33.7% 38.3% 26.0%Renal Disease 6.1% 9.3% 10.3% 4.1%Sickle Cell Dis 2.6% 5.2% 9.4% 1.6%
Any Chronic Disease 75.9% 86.2% 91.4% 70.9%Multiple Chronic Disease 52.2% 64.3% 73.3% 46.1%
Cancer 14.0% 13.7% 14.7% 15.1%
HIV/AIDS 23.0% 28.0% 26.1% 16.4%
Alcohol/Substance Abuse 73.0% 86.7% 90.8% 52.1%
Any Mental Illness 68.6% 78.4% 84.8% 57.2%Schizophrenia 26.7% 32.7% 36.9% 19.5%Pyschosis 19.6% 28.1% 36.6% 13.7%BiPoloar Disorder 39.0% 48.6% 54.3% 30.2%
MH or Substance Abuse 87.9% 94.4% 97.0% 73.8%MH and Substance Abuse 53.7% 70.8% 78.6% 35.6%
CHARACTERISTICS OF PATIENTS FLAGGEDBY CASE FINDING ALGORITHMCIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Diagnoses Reported in Claims Records
Risk Score> 50
Risk Score> 75
Risk Score> 90
AllPatients
Cereb Vasc Dis 4.9% 6.3% 8.1% 4.7%AMI 6.2% 9.5% 12.9% 5.2%Ischemic Heart Dis 22.5% 28.8% 35.5% 20.3%Congestive Heart Failure 16.4% 22.6% 26.9% 12.2%Hypertension 50.1% 58.3% 64.1% 48.2%Asthma 34.8% 45.7% 50.5% 26.2%COPD 23.5% 33.8% 42.3% 17.4%Diabetes 28.8% 33.7% 38.3% 26.0%Renal Disease 6.1% 9.3% 10.3% 4.1%Sickle Cell Dis 2.6% 5.2% 9.4% 1.6%
Any Chronic Disease 75.9% 86.2% 91.4% 70.9%Multiple Chronic Disease 52.2% 64.3% 73.3% 46.1%
Cancer 14.0% 13.7% 14.7% 15.1%
HIV/AIDS 23.0% 28.0% 26.1% 16.4%
Alcohol/Substance Abuse 73.0% 86.7% 90.8% 52.1%
Any Mental Illness 68.6% 78.4% 84.8% 57.2%Schizophrenia 26.7% 32.7% 36.9% 19.5%Pyschosis 19.6% 28.1% 36.6% 13.7%BiPoloar Disorder 39.0% 48.6% 54.3% 30.2%
MH or Substance Abuse 87.9% 94.4% 97.0% 73.8%MH and Substance Abuse 53.7% 70.8% 78.6% 35.6%
CHARACTERISTICS OF PATIENTS FLAGGEDBY CASE FINDING ALGORITHMCIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Diagnoses Reported in Claims Records
Risk Score> 50
Risk Score> 75
Risk Score> 90
AllPatients
Cereb Vasc Dis 4.9% 6.3% 8.1% 4.7%AMI 6.2% 9.5% 12.9% 5.2%Ischemic Heart Dis 22.5% 28.8% 35.5% 20.3%Congestive Heart Failure 16.4% 22.6% 26.9% 12.2%Hypertension 50.1% 58.3% 64.1% 48.2%Asthma 34.8% 45.7% 50.5% 26.2%COPD 23.5% 33.8% 42.3% 17.4%Diabetes 28.8% 33.7% 38.3% 26.0%Renal Disease 6.1% 9.3% 10.3% 4.1%Sickle Cell Dis 2.6% 5.2% 9.4% 1.6%
Any Chronic Disease 75.9% 86.2% 91.4% 70.9%Multiple Chronic Disease 52.2% 64.3% 73.3% 46.1%
Cancer 14.0% 13.7% 14.7% 15.1%
HIV/AIDS 23.0% 28.0% 26.1% 16.4%
Alcohol/Substance Abuse 73.0% 86.7% 90.8% 52.1%
Any Mental Illness 68.6% 78.4% 84.8% 57.2%Schizophrenia 26.7% 32.7% 36.9% 19.5%Pyschosis 19.6% 28.1% 36.6% 13.7%BiPoloar Disorder 39.0% 48.6% 54.3% 30.2%
MH or Substance Abuse 87.9% 94.4% 97.0% 73.8%MH and Substance Abuse 53.7% 70.8% 78.6% 35.6%
CHARACTERISTICS OF PATIENTS FLAGGEDBY CASE FINDING ALGORITHMCIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Diagnoses Reported in Claims Records
Risk Score> 50
Risk Score> 75
Risk Score> 90
AllPatients
Cereb Vasc Dis 4.9% 6.3% 8.1% 4.7%AMI 6.2% 9.5% 12.9% 5.2%Ischemic Heart Dis 22.5% 28.8% 35.5% 20.3%Congestive Heart Failure 16.4% 22.6% 26.9% 12.2%Hypertension 50.1% 58.3% 64.1% 48.2%Asthma 34.8% 45.7% 50.5% 26.2%COPD 23.5% 33.8% 42.3% 17.4%Diabetes 28.8% 33.7% 38.3% 26.0%Renal Disease 6.1% 9.3% 10.3% 4.1%Sickle Cell Dis 2.6% 5.2% 9.4% 1.6%
Any Chronic Disease 75.9% 86.2% 91.4% 70.9%Multiple Chronic Disease 52.2% 64.3% 73.3% 46.1%
Cancer 14.0% 13.7% 14.7% 15.1%
HIV/AIDS 23.0% 28.0% 26.1% 16.4%
Alcohol/Substance Abuse 73.0% 86.7% 90.8% 52.1%
Any Mental Illness 68.6% 78.4% 84.8% 57.2%Schizophrenia 26.7% 32.7% 36.9% 19.5%Pyschosis 19.6% 28.1% 36.6% 13.7%BiPoloar Disorder 39.0% 48.6% 54.3% 30.2%
MH or Substance Abuse 87.9% 94.4% 97.0% 73.8%MH and Substance Abuse 53.7% 70.8% 78.6% 35.6%
CHARACTERISTICS OF PATIENTS FLAGGEDBY CASE FINDING ALGORITHMCIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Diagnoses Reported in Claims Records
Risk Score> 50
Risk Score> 75
Risk Score> 90
AllPatients
Cereb Vasc Dis 4.9% 6.3% 8.1% 4.7%AMI 6.2% 9.5% 12.9% 5.2%Ischemic Heart Dis 22.5% 28.8% 35.5% 20.3%Congestive Heart Failure 16.4% 22.6% 26.9% 12.2%Hypertension 50.1% 58.3% 64.1% 48.2%Asthma 34.8% 45.7% 50.5% 26.2%COPD 23.5% 33.8% 42.3% 17.4%Diabetes 28.8% 33.7% 38.3% 26.0%Renal Disease 6.1% 9.3% 10.3% 4.1%Sickle Cell Dis 2.6% 5.2% 9.4% 1.6%
Any Chronic Disease 75.9% 86.2% 91.4% 70.9%Multiple Chronic Disease 52.2% 64.3% 73.3% 46.1%
Cancer 14.0% 13.7% 14.7% 15.1%
HIV/AIDS 23.0% 28.0% 26.1% 16.4%
Alcohol/Substance Abuse 73.0% 86.7% 90.8% 52.1%
Any Mental Illness 68.6% 78.4% 84.8% 57.2%Schizophrenia 26.7% 32.7% 36.9% 19.5%Pyschosis 19.6% 28.1% 36.6% 13.7%BiPoloar Disorder 39.0% 48.6% 54.3% 30.2%
MH or Substance Abuse 87.9% 94.4% 97.0% 73.8%MH and Substance Abuse 53.7% 70.8% 78.6% 35.6%
CHARACTERISTICS OF PATIENTS FLAGGEDBY CASE FINDING ALGORITHMCIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Selected Ambulatory Care Use Prior 12 Months
Risk Score> 50
Risk Score> 75
Risk Score> 90
AllPatients
Any primary care visit 71.7% 72.9% 68.3% 64.8%Any speciatly care visit 39.2% 40.8% 39.9% 35.6% No primary care visit 28.3% 27.1% 31.7% 35.2% No PC/spec care visit 24.2% 22.6% 26.7% 31.3% No PC/spec/OBGYN visit 23.7% 22.1% 26.1% 30.7%
Any psych visit 35.3% 35.8% 36.9% 29.6%Any alcohol/drug visit 29.5% 38.8% 38.8% 19.5%
Any dental visit 37.3% 39.6% 37.5% 32.4%Any home care 12.8% 17.2% 18.6% 8.5%Any transportation 45.9% 61.1% 70.2% 32.2%Any pharmacy 88.0% 89.5% 85.6% 78.3%Any DME 18.7% 20.9% 20.5% 15.2%
Any comp case mgt 7.6% 10.8% 10.3% 5.2%Any community rehab 1.1% 1.3% 0.8% 0.8%
CHARACTERISTICS OF PATIENTS FLAGGEDBY CASE FINDING ALGORITHMCIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Selected Ambulatory Care Use Prior 12 Months
Risk Score> 50
Risk Score> 75
Risk Score> 90
AllPatients
Any primary care visit 71.7% 72.9% 68.3% 64.8%Any speciatly care visit 39.2% 40.8% 39.9% 35.6% No primary care visit 28.3% 27.1% 31.7% 35.2% No PC/spec care visit 24.2% 22.6% 26.7% 31.3% No PC/spec/OBGYN visit 23.7% 22.1% 26.1% 30.7%
Any psych visit 35.3% 35.8% 36.9% 29.6%Any alcohol/drug visit 29.5% 38.8% 38.8% 19.5%
Any dental visit 37.3% 39.6% 37.5% 32.4%Any home care 12.8% 17.2% 18.6% 8.5%Any transportation 45.9% 61.1% 70.2% 32.2%Any pharmacy 88.0% 89.5% 85.6% 78.3%Any DME 18.7% 20.9% 20.5% 15.2%
Any comp case mgt 7.6% 10.8% 10.3% 5.2%Any community rehab 1.1% 1.3% 0.8% 0.8%
“MEDICAL HOME”OUTPATIENT CARE[PRIMARY/SPECIALTY/OB]
• “Loyal” patients: 3+ visits with one provider having ≥ 50%of visits during the 2-year period
• “Shopper” patients: 3+ visits with no provider having ≥ 50%of visits during the 2-year period
• “Occasional users”: Less than 3 visits during the 2-yearperiod
• “No PC/Spec/OB” patients: No primary care, specialty care,or OB visits during the 2-year period
Source: NYU Center for Health and Public Service Research, UHF, NYSDOH, 2006.
“Medical Home” for Patients with Risk Score ≥50Based on Prior 2-Years of Ambulatory Use
CHARACTERISTICS OF PATIENTS FLAGGEDBY CASE FINDING ALGORITHMCIDP ELIGIBLE - MODEL DEVELOPMENT RUN
"Medical Home" Status AllNYS
Number ofPC/Spec/OB
ProvidersTouched
Loyal 48.9% 2.80 OPD/Satellite 25.1% 2.97 D&TC 15.0% 2.55 MD 8.8% 2.71Shopper 18.8% 5.39Occasional User 13.3% 1.18No PC/Spec/OB 19.0% 0.00
Total 100.0% 2.54
Source: NYU Center for Health and Public Service Research, UHF, NYSDOH, 2006.
“Medical Home” for Patients with Risk Score ≥50Based on Prior 2-Years of Ambulatory Use
CHARACTERISTICS OF PATIENTS FLAGGEDBY CASE FINDING ALGORITHMCIDP ELIGIBLE - MODEL DEVELOPMENT RUN
"Medical Home" Status AllNYS
Number ofPC/Spec/OB
ProvidersTouched
Loyal 48.9% 2.80 OPD/Satellite 25.1% 2.97 D&TC 15.0% 2.55 MD 8.8% 2.71Shopper 18.8% 5.39Occasional User 13.3% 1.18No PC/Spec/OB 19.0% 0.00
Total 100.0% 2.54
51%
Source: NYU Center for Health and Public Service Research, UHF, NYSDOH, 2006.
“Medical Home” for Patients with Risk Score ≥50Based on Prior 2-Years of Ambulatory Use
CHARACTERISTICS OF PATIENTS FLAGGEDBY CASE FINDING ALGORITHMCIDP ELIGIBLE - MODEL DEVELOPMENT RUN
"Medical Home" Status AllNYS
Number ofPC/Spec/OB
ProvidersTouched
Loyal 48.9% 2.80 OPD/Satellite 25.1% 2.97 D&TC 15.0% 2.55 MD 8.8% 2.71Shopper 18.8% 5.39Occasional User 13.3% 1.18No PC/Spec/OB 19.0% 0.00
Total 100.0% 2.54
Source: NYU Center for Health and Public Service Research, UHF, NYSDOH, 2006.
“Medical Home” for Patients with Risk Score ≥50Based on Prior 2-Years of Ambulatory Use
CHARACTERISTICS OF PATIENTS FLAGGEDBY CASE FINDING ALGORITHMCIDP ELIGIBLE - MODEL DEVELOPMENT RUN
"Medical Home" Status AllNYS
Number ofPC/Spec/OB
ProvidersTouched
Loyal 48.9% 2.80 OPD/Satellite 25.1% 2.97 D&TC 15.0% 2.55 MD 8.8% 2.71Shopper 18.8% 5.39Occasional User 13.3% 1.18No PC/Spec/OB 19.0% 0.00
Total 100.0% 2.54
Number ofPC/Spec/OB
ProvidersTouched
% ofPatients
AllNYS
1 Provider 0.0%2 Providers 4.9%3 Providers 22.7%4-5 Providers 35.7%5-9 Providers 28.8%10+ Providers 8.0% Total 100.0%
CHARACTERISTICS OF PATIENTS FLAGGEDBY CASE FINDING ALGORITHMCIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Risk Score> 50
Risk Score> 75
Risk Score> 90
AllPatients
Costs Prior 12 MonthsInpatient 20,973 42,357 75,221 12,442Emergency Department 306 576 1,040 199Primary Care Visit 489 535 495 416Specialty Care Visit 80 83 75 71Psychiatric Care Visit 1,045 862 693 899Substance Abuse Visit 1,129 1,342 1,070 748Other Ambulatory 1,989 2,746 3,223 1,494Pharmacy 6,470 7,711 7,545 4,905Transportation 427 658 810 289Community Rehab 109 112 57 73Case Management 349 544 554 230Personal Care 853 914 755 754Home Care 875 1,201 1,357 601LTHHC 49 116 214 29All Other 2,388 3,500 3,738 1,738
Total Cost 37,530 63,259 96,848 24,885
Costs Next 12 MonthsInpatient 26,777 45,513 70,491 16,791Emergency Department 299 527 921 198Primary Care Visit 415 394 360 375Specialty Care Visit 52 44 34 55Psychiatric Care Visit 1,041 786 582 964Substance Abuse Visit 1,155 1,320 1,061 796Other Ambulatory 2,183 2,831 2,987 1,678Pharmacy 7,246 7,726 7,194 5,834Transportation 548 752 794 389Community Rehab 170 184 59 173Case Management 392 547 533 267Personal Care 1,017 1,023 795 918Home Care 1,229 1,327 1,392 986LTHHC 117 117 63 110All Other 3,895 5,071 5,409 3,089
Total Cost 46,537 68,162 92,674 32,622
CHARACTERISTICS OF PATIENTS FLAGGEDBY CASE FINDING ALGORITHMCIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Risk Score> 50
Risk Score> 75
Risk Score> 90
AllPatients
Costs Prior 12 MonthsInpatient 20,973 42,357 75,221 12,442Emergency Department 306 576 1,040 199Primary Care Visit 489 535 495 416Specialty Care Visit 80 83 75 71Psychiatric Care Visit 1,045 862 693 899Substance Abuse Visit 1,129 1,342 1,070 748Other Ambulatory 1,989 2,746 3,223 1,494Pharmacy 6,470 7,711 7,545 4,905Transportation 427 658 810 289Community Rehab 109 112 57 73Case Management 349 544 554 230Personal Care 853 914 755 754Home Care 875 1,201 1,357 601LTHHC 49 116 214 29All Other 2,388 3,500 3,738 1,738
Total Cost 37,530 63,259 96,848 24,885
Costs Next 12 MonthsInpatient 26,777 45,513 70,491 16,791Emergency Department 299 527 921 198Primary Care Visit 415 394 360 375Specialty Care Visit 52 44 34 55Psychiatric Care Visit 1,041 786 582 964Substance Abuse Visit 1,155 1,320 1,061 796Other Ambulatory 2,183 2,831 2,987 1,678Pharmacy 7,246 7,726 7,194 5,834Transportation 548 752 794 389Community Rehab 170 184 59 173Case Management 392 547 533 267Personal Care 1,017 1,023 795 918Home Care 1,229 1,327 1,392 986LTHHC 117 117 63 110All Other 3,895 5,071 5,409 3,089
Total Cost 46,537 68,162 92,674 32,622
CHARACTERISTICS OF PATIENTS FLAGGEDBY CASE FINDING ALGORITHMCIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Risk Score> 50
Risk Score> 75
Risk Score> 90
AllPatients
Costs Prior 12 MonthsInpatient 20,973 42,357 75,221 12,442Emergency Department 306 576 1,040 199Primary Care Visit 489 535 495 416Specialty Care Visit 80 83 75 71Psychiatric Care Visit 1,045 862 693 899Substance Abuse Visit 1,129 1,342 1,070 748Other Ambulatory 1,989 2,746 3,223 1,494Pharmacy 6,470 7,711 7,545 4,905Transportation 427 658 810 289Community Rehab 109 112 57 73Case Management 349 544 554 230Personal Care 853 914 755 754Home Care 875 1,201 1,357 601LTHHC 49 116 214 29All Other 2,388 3,500 3,738 1,738
Total Cost 37,530 63,259 96,848 24,885
Costs Next 12 MonthsInpatient 26,777 45,513 70,491 16,791Emergency Department 299 527 921 198Primary Care Visit 415 394 360 375Specialty Care Visit 52 44 34 55Psychiatric Care Visit 1,041 786 582 964Substance Abuse Visit 1,155 1,320 1,061 796Other Ambulatory 2,183 2,831 2,987 1,678Pharmacy 7,246 7,726 7,194 5,834Transportation 548 752 794 389Community Rehab 170 184 59 173Case Management 392 547 533 267Personal Care 1,017 1,023 795 918Home Care 1,229 1,327 1,392 986LTHHC 117 117 63 110All Other 3,895 5,071 5,409 3,089
Total Cost 46,537 68,162 92,674 32,622
$0
$2,000
$4,000
$6,000
$8,000
$10,000
$12,000
$14,000
$16,000
$18,000
0% 1% 2% 3% 4% 5% 6% 7% 8% 9% 10%
11%
12%
13%
14%
15%
16%
17%
18%
19%
20%
21%
22%
23%
24%
25%
`
CASE FINDING ALGORITHMMAXIMUM EXPENDITURE/PATIENT FOR BREAK EVEN
% Reduction in Admissions
Inte
rven
tion
Cos
t/Pat
ient
Risk Score50+
Risk Score90+
Risk Score75+
$2,799
$4,521
$6,630
$4,199
$6,783
$9,990
$5,599
$9,044
$13,320
CHARACTERISTICS OF PATIENTS FLAGGEDBY CASE FINDING ALGORITHMCIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Top 25 Principal Diagnosis of “Future Admissions”
ICD-9 ICD-9 Description Numberof Adms
% ofTotal
Cumula-tive %
30391 ALCOH DEP NEC/NOS-CONTIN 7,493 8.7% 8.7%29181 ALCOHOL WITHDRAWAL 4,518 5.2% 13.9%30401 OPIOID DEPENDENCE-CONTIN 4,198 4.8% 18.7%042 HUMAN IMMUNO VIRUS DIS 3,563 4.1% 22.8%30421 COCAINE DEPEND-CONTIN 3,283 3.8% 26.6%2920 DRUG WITHDRAWAL 3,048 3.5% 30.1%30390 ALCOH DEP NEC/NOS-UNSPEC 2,099 2.4% 32.6%4280 CHF NOS 1,983 2.3% 34.9%29570 SCHIZOAFFECTIVE DIS NOS 1,807 2.1% 36.9%28262 HB-SS DISEASE W CRISIS 1,515 1.7% 38.7%486 PNEUMONIA, ORGANISM NOS 1,478 1.7% 40.4%78659 CHEST PAIN NEC 1,469 1.7% 42.1%49392 ASTHMA NOS W (AC) EXAC 1,443 1.7% 43.8%30471 OPIOID/OTHER DEP-CONTIN 1,428 1.6% 45.4%78039 CONVULSIONS NEC 998 1.2% 46.6%29284 DRUG-INDUCED MOOD DISORD 980 1.1% 47.7%49121 OBS CHR BRONC W(AC) EXAC 917 1.1% 48.8%29574 SCHIZOAFFTV DIS-CHR/EXAC 914 1.1% 49.8%49322 CH OBST ASTH W (AC) EXAC 900 1.0% 50.9%311 DEPRESSIVE DISORDER NEC 832 1.0% 51.8%6826 CELLULITIS OF LEG 816 0.9% 52.8%29534 PARAN SCHIZO-CHR/EXACERB 765 0.9% 53.6%29530 PARANOID SCHIZO-UNSPEC 726 0.8% 54.5%41401 CRNRY ATHRSCL NATVE VSSL 714 0.8% 55.3%2989 PSYCHOSIS NOS 637 0.7% 56.0%
A SOMEWHAT IDEALIZEDDESCRIPTION OF THE APPROACH
TO THE PROBLEM
Step 1: See if you can develop a predictive model to identify patientsfor whom you think you can do something
Step 2: Learn as much as you can about these patients to helpin designing the intervention(s)- Use available administrative data- Apply algorithm to real patients – interview a sample of these
patients (and their providers, families, caregivers, etc.)
Step 3: Implement/evaluate demonstration projects based on thisbased on this information- Pilot with a quasi-experimental design (intervention/control)- Conduct “formative” evaluation during early phases of
implementation- Assess impact of intervention on outcomes/utilization
Step 4: Disseminate results/Scale up if it works
CHARACTERISTICS OFINTERVIEWED BELLEVUE PATIENTS
Source: High Cost Medicaid Project – Bellevue Hospital Center, NYU Center for Health and Public Service Research, 2006.
Characteristic % ofTotal
Marrital statusMarried/living with partner 14%Separated 16%Divorced 10%Widowed 4%Never married 56%
Curently living alone 52%
No "close" frriends/relatives 16%Two or fewer "close" friends/relatives 48%
Low "Perceived Availablity of Support" 42%
Bellevue Hospital Center
Characteristic % ofTotal
Usual source of careNone 16%Emergency department 42%OPD/Clinic 20%Community based clinic 8%Private/Group MD/other 14%
Source: High Cost Medicaid Project – Bellevue Hospital Center, NYU Center for Health and Public Service Research, 2006.
58%
CHARACTERISTICS OFINTERVIEWED BELLEVUE PATIENTS
Bellevue Hospital Center
Characteristic % ofTotal
Current housing statusApartment/home rental 34%Public housing 2%Residential facility 2%Staying with family/friends 24%Shelter 8%Homeless 28%
Homeless anytime previous 2 years 50%
Source: High Cost Medicaid Project – Bellevue Hospital Center, NYU Center for Health and Public Service Research, 2006.
60%
CHARACTERISTICS OFINTERVIEWED BELLEVUE PATIENTS
Bellevue Hospital Center
SO WHAT’S IT GOING TO TAKE?
• Multi-disciplinary approach for individualized needs assessment and care planning for participating patients
SO WHAT’S IT GOING TO TAKE?
• Multi-disciplinary approach for individualized needs assessment and care planning for participating patients
• Integrated/organized/coordinated health and social service delivery system– “Enhanced” primary care– Specialty care– Substance abuse/mental health services– Inpatient care– Community based social support– Supportive housing for many– Etc, etc, etc
SO WHAT’S IT GOING TO TAKE?
• Multi-disciplinary approach for individualized needs assessment and care planning for participating patients
• Integrated/organized/coordinated health and social service delivery system
• Some sort of care/service-coordinator/arranger– With a reasonable caseload size– With a clear mission (to improve health and to reduce costs)
SO WHAT’S IT GOING TO TAKE?
• Multi-disciplinary approach for individualized needs assessment and care planning for participating patients
• Integrated/organized/coordinated health and social service delivery system
• Some sort of care/service-coordinator/arranger
• Core IT and care coordination support capacity to…– Track patient utilization in close to real time– Mine administrative data and target interventions/outreach– Provide analysis of utilization patterns
• Identify trends/problems to continuously re-design intervention strategies• Provide feed-back to providers on performance
– Hospital admission rates– ED visit rates– Adherence to evidence based practice standards
– Support effective use of electronic medical records where available
SO WHAT’S IT GOING TO TAKE?
• Multi-disciplinary approach for individualized needs assessment and care planning for participating patients
• Integrated/organized/coordinated health and social service delivery system
• Some sort of care/service-coordinator/arranger
• Core IT and care coordination support capacity
• Ability to provide real time support at critical junctures– ED visit - prevention of “social admissions”– Hospital discharge - effective community support/management planning– Patient initiated - help for an emerging crisis
SO WHAT’S IT GOING TO TAKE?
• Multi-disciplinary approach for individualized needs assessment and care planning for participating patients
• Integrated/organized/coordinated health and social service delivery system
• Some sort of care/service-coordinator/arranger
• Core IT and care coordination support capacity
• Ability to provide real time support at critical junctures
• Incentives/reimbursement policies to encourage and reward “effective and cost efficient care”– Hospitals must have a shared interest in avoiding admissions– Reimbursement rates for OP services need to be related to their costs– Costs of social support need to be recognized– [No new money – new/augmented services offset by IP savings]
A SOMEWHAT IDEALIZEDDESCRIPTION OF THE APPROACH
TO THE PROBLEM
Step 1: See if you can develop a predictive model to identify patientsfor whom you think you can do something
Step 2: Learn as much as you can about these patients to helpin designing the intervention(s)- Use available administrative data- Apply algorithm to real patients – interview a sample of these
patients (and their providers, families, caregivers, etc.)
Step 3: Implement/evaluate demonstration projects based on thisbased on this information- Pilot with a quasi-experimental design (intervention/control)- Conduct “formative” evaluation during early phases of
implementation- Assess impact of intervention on outcomes/utilization
Step 4: Disseminate results/Scale up if it works
SO WHERE ARE WE NOW?
• After a competitive procurement process that took 13 months to implement, awards for 7 pilots March, 2009– 2 pilots with moderately integrated health care delivery “systems”– 2 from community based primary care providers– 3 largely involving managed care organizations as key players
SO WHERE ARE WE NOW?
• After a competitive procurement process that took 13 months to implement, awards for 7 pilots March, 2009– 2 pilots with moderately integrated health care delivery “systems”– 2 from community based primary care providers– 3 largely involving managed care organizations as key players
• July, 2009: One pilot dropped out
SO WHERE ARE WE NOW?
• After a competitive procurement process that took 13 months to implement, awards for 7 pilots March, 2009– 2 pilots with moderately integrated health care delivery “systems”– 2 from community based primary care providers– 3 largely involving managed care organizations as key players
• July, 2009: One pilot dropped out
• August, 2009: Enrollment begins 6 remaining pilots
SO WHERE ARE WE NOW?
• After a competitive procurement process that took 13 months to implement, awards for 7 pilots March, 2009– 2 pilots with moderately integrated health care delivery “systems”– 2 from community based primary care providers– 3 largely involving managed care organizations as key players
• July, 2009: One pilot dropped out
• August, 2009: Enrollment begins 6 remaining pilots
• January, 2010:– Two learning collaborative meetings have been held– Sites have received 2 enrollment refreshments– Most sites experiencing problems locating patients– Way too early to assess impact (first formative evaluation site visits under way)
A SOMEWHAT IDEALIZEDDESCRIPTION OF THE APPROACH
TO THE PROBLEM
Step 1: See if you can develop a predictive model to identify patientsfor whom you think you can do something
Step 2: Learn as much as you can about these patients to helpin designing the intervention(s)- Use available administrative data- Apply algorithm to real patients – interview a sample of these
patients (and their providers, families, caregivers, etc.)
Step 3: Implement/evaluate demonstration projects based on thisinformation- Pilot with a quasi-experimental design (intervention/control)- Conduct “formative” evaluation during early phases of
implementation- Assess impact of intervention on outcomes/utilization
Step 4: Disseminate results/Scale up if it works