1. TOWARDS EVIDENCE BASED
POLICY MAKING
EXPERIENCE OF THE
CHRONIC ILLNESS DEMONSTRATION PROGRAM
FOR NEW YORK MEDICAID PATIENTS
January 14, 2010
New York University
Robert F. Wagner Graduate School of Public Service

2. 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

3. AN UNUSUAL CONFLUENCE
OF CIRCUMSTANCES
• A policy wonk frenzy about the small number of high cost
patients responsible for substantial portion of costs

4. MEDICAID ENROLLEES AND
EXPENDITURES
BY ENROLLMENT GROUP – U.S. 2006
100%
10% Elderly
26%
15%
80%
Blind and
25% Disabled
60%
43%
Adults
40%
50% 12%
20% Children
19%
0%
Enrollees Expenditures
59 Million $268 Million
Source: Kaiser Commission on Medicaid and the Uninsured – 2009.

5. NEW YORK MEDICAID
Adult Disabled – Non-Mandatory Managed Care
100%
27.1%
80%
Percent of Total
17.0%
60% 80.0%
25.9%
40%
72.9%
20%
10.0% 30.0%
7.0%
3.0%
0%
Patients Expenditures
Source: NYU Center for Health and Public Service Research, UHF, NYSDOH, 2006.

6. AN UNUSUAL CONFLUENCE
OF 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

7. AN UNUSUAL CONFLUENCE
OF 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

8. AN UNUSUAL CONFLUENCE
OF 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

9. AN UNUSUAL CONFLUENCE
OF 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]

10. AN UNUSUAL CONFLUENCE
OF 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

11. A SOMEWHAT IDEALIZED
DESCRIPTION OF THE APPROACH
TO THE PROBLEM
Step 1: See if you can develop a predictive model to identify patients
for whom you think you can do something

12. A SOMEWHAT IDEALIZED
DESCRIPTION OF THE APPROACH
TO THE PROBLEM
Step 1: See if you can develop a predictive model to identify patients
for whom you think you can do something
Step 2: Learn as much as you can about these patients to help
in 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.)

13. A SOMEWHAT IDEALIZED
DESCRIPTION OF THE APPROACH
TO THE PROBLEM
Step 1: See if you can develop a predictive model to identify patients
for whom you think you can do something
Step 2: Learn as much as you can about these patients to help
in 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 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

14. A SOMEWHAT IDEALIZED
DESCRIPTION OF THE APPROACH
TO THE PROBLEM
Step 1: See if you can develop a predictive model to identify patients
for whom you think you can do something
Step 2: Learn as much as you can about these patients to help
in 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 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

15. A SOMEWHAT IDEALIZED
DESCRIPTION OF THE APPROACH
TO THE PROBLEM
Evidenced-based management/policy making
Step 1: See if you can develop a predictive model to identify patients
for whom you think you can do something
Step 2: Learn as much as you can about these patients to help
in 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 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

16. THE PREDICTIVE MODELING
ALGORITHM 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

17. BASIC APPROACH FOR DEVELOPMENT
OF RISK PREDICTION ALGORITHM
Index
Quarters
Q1 Q2 Q3 Q4
Year 1 Year 2 Year 3 Year 4 Year 5

18. BASIC APPROACH
TYPES OF VARIABLES IN ALGORITHM
• 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

19. CASE FINDING ALGORITHM
NUMBER OF PATIENTS IDENTIFIED
CIDP ELIGIBLE - MODEL DEVELOPMENT RUN
50,000
45,000
40,000
Patients Identified
35,000
TOTAL FLAGGED
30,000 False
Positives
25,000 33%
20,000
15,000 False
Positives
10,000 CORRECTLY FLAGGED 15%
False
Positives
5,000 7%
0
40 45 50 55 60 65 70 75 80 85 90 95
Risk Score Threshold

20. CHARACTERISTICS OF PATIENTS FLAGGED
BY CASE FINDING ALGORITHM
CIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Demographic Characteristics
Risk Score Risk Score Risk Score All
> 50 > 75 > 90 Patients
N 33,363 8,713 2,176 64,446
Age 45.1 44.8 44.3 47.6
Female 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%

21. CHARACTERISTICS OF PATIENTS FLAGGED
BY CASE FINDING ALGORITHM
CIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Diagnoses Reported in Claims Records
Risk Score Risk Score Risk Score All
> 50 > 75 > 90 Patients
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%

22. CHARACTERISTICS OF PATIENTS FLAGGED
BY CASE FINDING ALGORITHM
CIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Diagnoses Reported in Claims Records
Risk Score Risk Score Risk Score All
> 50 > 75 > 90 Patients
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%

23. CHARACTERISTICS OF PATIENTS FLAGGED
BY CASE FINDING ALGORITHM
CIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Diagnoses Reported in Claims Records
Risk Score Risk Score Risk Score All
> 50 > 75 > 90 Patients
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%

24. CHARACTERISTICS OF PATIENTS FLAGGED
BY CASE FINDING ALGORITHM
CIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Diagnoses Reported in Claims Records
Risk Score Risk Score Risk Score All
> 50 > 75 > 90 Patients
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%

25. CHARACTERISTICS OF PATIENTS FLAGGED
BY CASE FINDING ALGORITHM
CIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Diagnoses Reported in Claims Records
Risk Score Risk Score Risk Score All
> 50 > 75 > 90 Patients
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%

26. CHARACTERISTICS OF PATIENTS FLAGGED
BY CASE FINDING ALGORITHM
CIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Selected Ambulatory Care Use Prior 12 Months
Risk Score Risk Score Risk Score All
> 50 > 75 > 90 Patients
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%

27. CHARACTERISTICS OF PATIENTS FLAGGED
BY CASE FINDING ALGORITHM
CIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Selected Ambulatory Care Use Prior 12 Months
Risk Score Risk Score Risk Score All
> 50 > 75 > 90 Patients
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%

28. “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-year
period
• “No PC/Spec/OB” patients: No primary care, specialty care,
or OB visits during the 2-year period

29. CHARACTERISTICS OF PATIENTS FLAGGED
BY CASE FINDING ALGORITHM
CIDP ELIGIBLE - MODEL DEVELOPMENT RUN
“Medical Home” for Patients with Risk Score ≥50
Based on Prior 2-Years of Ambulatory Use
Number of
All PC/Spec/OB
"Medical Home" Status
NYS Providers
Touched
Loyal 48.9% 2.80
OPD/Satellite 25.1% 2.97
D&TC 15.0% 2.55
MD 8.8% 2.71
Shopper 18.8% 5.39
Occasional User 13.3% 1.18
No PC/Spec/OB 19.0% 0.00
Total 100.0% 2.54
Source: NYU Center for Health and Public Service Research, UHF, NYSDOH, 2006.

30. CHARACTERISTICS OF PATIENTS FLAGGED
BY CASE FINDING ALGORITHM
CIDP ELIGIBLE - MODEL DEVELOPMENT RUN
“Medical Home” for Patients with Risk Score ≥50
Based on Prior 2-Years of Ambulatory Use
Number of
All PC/Spec/OB
"Medical Home" Status
NYS Providers
Touched
Loyal 48.9% 2.80
OPD/Satellite 25.1% 2.97
D&TC 15.0% 2.55
MD 8.8% 2.71
Shopper 18.8% 5.39
51% Occasional User 13.3% 1.18
No PC/Spec/OB 19.0% 0.00
Total 100.0% 2.54
Source: NYU Center for Health and Public Service Research, UHF, NYSDOH, 2006.

31. CHARACTERISTICS OF PATIENTS FLAGGED
BY CASE FINDING ALGORITHM
CIDP ELIGIBLE - MODEL DEVELOPMENT RUN
“Medical Home” for Patients with Risk Score ≥50
Based on Prior 2-Years of Ambulatory Use
Number of
All PC/Spec/OB
"Medical Home" Status
NYS Providers
Touched
Loyal 48.9% 2.80
OPD/Satellite 25.1% 2.97
D&TC 15.0% 2.55
MD 8.8% 2.71
Shopper 18.8% 5.39
Occasional User 13.3% 1.18
No PC/Spec/OB 19.0% 0.00
Total 100.0% 2.54
Source: NYU Center for Health and Public Service Research, UHF, NYSDOH, 2006.

32. CHARACTERISTICS OF PATIENTS FLAGGED
BY CASE FINDING ALGORITHM
CIDP ELIGIBLE - MODEL DEVELOPMENT RUN
“Medical Home” for Patients with Risk Score ≥50
Based on Prior 2-Years of Ambulatory Use
Number of % of
Number of PC/Spec/OB Patients
All PC/Spec/OB Providers All
"Medical Home" Status
NYS Providers Touched NYS
Touched
1 Provider 0.0%
Loyal 48.9% 2.80 2 Providers 4.9%
OPD/Satellite 25.1% 2.97 3 Providers 22.7%
D&TC 15.0% 2.55 4-5 Providers 35.7%
MD 8.8% 2.71 5-9 Providers 28.8%
Shopper 18.8% 5.39 10+ Providers 8.0%
Occasional User 13.3% 1.18 Total 100.0%
No PC/Spec/OB 19.0% 0.00
Total 100.0% 2.54
Source: NYU Center for Health and Public Service Research, UHF, NYSDOH, 2006.

33. CHARACTERISTICS OF PATIENTS FLAGGED
BY CASE FINDING ALGORITHM
CIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Risk Score Risk Score Risk Score All
> 50 > 75 > 90 Patients
Costs Prior 12 Months
Inpatient 20,973 42,357 75,221 12,442
Emergency Department 306 576 1,040 199
Primary Care Visit 489 535 495 416
Specialty Care Visit 80 83 75 71
Psychiatric Care Visit 1,045 862 693 899
Substance Abuse Visit 1,129 1,342 1,070 748
Other Ambulatory 1,989 2,746 3,223 1,494
Pharmacy 6,470 7,711 7,545 4,905
Transportation 427 658 810 289
Community Rehab 109 112 57 73
Case Management 349 544 554 230
Personal Care 853 914 755 754
Home Care 875 1,201 1,357 601
LTHHC 49 116 214 29
All Other 2,388 3,500 3,738 1,738
Total Cost 37,530 63,259 96,848 24,885
Costs Next 12 Months
Inpatient 26,777 45,513 70,491 16,791
Emergency Department 299 527 921 198
Primary Care Visit 415 394 360 375
Specialty Care Visit 52 44 34 55
Psychiatric Care Visit 1,041 786 582 964
Substance Abuse Visit 1,155 1,320 1,061 796
Other Ambulatory 2,183 2,831 2,987 1,678
Pharmacy 7,246 7,726 7,194 5,834
Transportation 548 752 794 389
Community Rehab 170 184 59 173
Case Management 392 547 533 267
Personal Care 1,017 1,023 795 918
Home Care 1,229 1,327 1,392 986
LTHHC 117 117 63 110
All Other 3,895 5,071 5,409 3,089
Total Cost 46,537 68,162 92,674 32,622

34. CHARACTERISTICS OF PATIENTS FLAGGED
BY CASE FINDING ALGORITHM
CIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Risk Score Risk Score Risk Score All
> 50 > 75 > 90 Patients
Costs Prior 12 Months
Inpatient 20,973 42,357 75,221 12,442
Emergency Department 306 576 1,040 199
Primary Care Visit 489 535 495 416
Specialty Care Visit 80 83 75 71
Psychiatric Care Visit 1,045 862 693 899
Substance Abuse Visit 1,129 1,342 1,070 748
Other Ambulatory 1,989 2,746 3,223 1,494
Pharmacy 6,470 7,711 7,545 4,905
Transportation 427 658 810 289
Community Rehab 109 112 57 73
Case Management 349 544 554 230
Personal Care 853 914 755 754
Home Care 875 1,201 1,357 601
LTHHC 49 116 214 29
All Other 2,388 3,500 3,738 1,738
Total Cost 37,530 63,259 96,848 24,885
Costs Next 12 Months
Inpatient 26,777 45,513 70,491 16,791
Emergency Department 299 527 921 198
Primary Care Visit 415 394 360 375
Specialty Care Visit 52 44 34 55
Psychiatric Care Visit 1,041 786 582 964
Substance Abuse Visit 1,155 1,320 1,061 796
Other Ambulatory 2,183 2,831 2,987 1,678
Pharmacy 7,246 7,726 7,194 5,834
Transportation 548 752 794 389
Community Rehab 170 184 59 173
Case Management 392 547 533 267
Personal Care 1,017 1,023 795 918
Home Care 1,229 1,327 1,392 986
LTHHC 117 117 63 110
All Other 3,895 5,071 5,409 3,089
Total Cost 46,537 68,162 92,674 32,622

35. CHARACTERISTICS OF PATIENTS FLAGGED
BY CASE FINDING ALGORITHM
CIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Risk Score Risk Score Risk Score All
> 50 > 75 > 90 Patients
Costs Prior 12 Months
Inpatient 20,973 42,357 75,221 12,442
Emergency Department 306 576 1,040 199
Primary Care Visit 489 535 495 416
Specialty Care Visit 80 83 75 71
Psychiatric Care Visit 1,045 862 693 899
Substance Abuse Visit 1,129 1,342 1,070 748
Other Ambulatory 1,989 2,746 3,223 1,494
Pharmacy 6,470 7,711 7,545 4,905
Transportation 427 658 810 289
Community Rehab 109 112 57 73
Case Management 349 544 554 230
Personal Care 853 914 755 754
Home Care 875 1,201 1,357 601
LTHHC 49 116 214 29
All Other 2,388 3,500 3,738 1,738
Total Cost 37,530 63,259 96,848 24,885
Costs Next 12 Months
Inpatient 26,777 45,513 70,491 16,791
Emergency Department 299 527 921 198
Primary Care Visit 415 394 360 375
Specialty Care Visit 52 44 34 55
Psychiatric Care Visit 1,041 786 582 964
Substance Abuse Visit 1,155 1,320 1,061 796
Other Ambulatory 2,183 2,831 2,987 1,678
Pharmacy 7,246 7,726 7,194 5,834
Transportation 548 752 794 389
Community Rehab 170 184 59 173
Case Management 392 547 533 267
Personal Care 1,017 1,023 795 918
Home Care 1,229 1,327 1,392 986
LTHHC 117 117 63 110
All Other 3,895 5,071 5,409 3,089
Total Cost 46,537 68,162 92,674 32,622

36. CASE FINDING ALGORITHM
MAXIMUM EXPENDITURE/PATIENT FOR BREAK EVEN
$18,000
Risk Score
$16,000 90+
Intervention Cost/Patient
$14,000
$13,320
$12,000
Risk Score
$10,000 75+
$9,990
$9,044
$8,000
`
Risk Score
$6,783
$6,000 $6,630 50+
$5,599
$4,000 $4,521
$4,199
$2,000 $2,799
$0
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
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
% Reduction in Admissions

37. CHARACTERISTICS OF PATIENTS FLAGGED
BY CASE FINDING ALGORITHM
CIDP ELIGIBLE - MODEL DEVELOPMENT RUN
Top 25 Principal Diagnosis of “Future Admissions”
Number % of Cumula-
ICD-9 ICD-9 Description
of Adms Total 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%

38. A SOMEWHAT IDEALIZED
DESCRIPTION OF THE APPROACH
TO THE PROBLEM
Step 1: See if you can develop a predictive model to identify patients
for whom you think you can do something
Step 2: Learn as much as you can about these patients to help
in 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 this
based 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

39. CHARACTERISTICS OF
INTERVIEWED BELLEVUE PATIENTS
% of
Characteristic
Total
Marrital status
Married/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
Source: High Cost Medicaid Project – Bellevue Hospital Center, NYU Center for Health and Public Service Research, 2006.

40. CHARACTERISTICS OF
INTERVIEWED BELLEVUE PATIENTS
% of
Characteristic
Total
Usual source of care
None 16%
58%
Emergency department 42%
OPD/Clinic 20%
Community based clinic 8%
Private/Group MD/other 14%
Bellevue Hospital Center
Source: High Cost Medicaid Project – Bellevue Hospital Center, NYU Center for Health and Public Service Research, 2006.

41. CHARACTERISTICS OF
INTERVIEWED BELLEVUE PATIENTS
% of
Characteristic
Total
Current housing status
Apartment/home rental 34%
Public housing 2%
Residential facility 2%
Staying with family/friends 24%
Shelter 60% 8%
Homeless 28%
Homeless anytime previous 2 years 50%
Bellevue Hospital Center
Source: High Cost Medicaid Project – Bellevue Hospital Center, NYU Center for Health and Public Service Research, 2006.

42. SO WHAT’S IT GOING TO TAKE?
• Multi-disciplinary approach for individualized needs
assessment and care planning for participating patients

43. 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

44. 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)

45. 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

46. 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

47. 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]

48. A SOMEWHAT IDEALIZED
DESCRIPTION OF THE APPROACH
TO THE PROBLEM
Step 1: See if you can develop a predictive model to identify patients
for whom you think you can do something
Step 2: Learn as much as you can about these patients to help
in 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 this
based 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

49. 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

50. 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

51. 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

52. 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)

53. A SOMEWHAT IDEALIZED
DESCRIPTION OF THE APPROACH
TO THE PROBLEM
Step 1: See if you can develop a predictive model to identify patients
for whom you think you can do something
Step 2: Learn as much as you can about these patients to help
in 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 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