john billings: towards evidence based policy making

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 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]






• 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


TO THE PROBLEM


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.)


TO THE PROBLEM




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


TO THE PROBLEM






Step 4: Disseminate results/Scale up if it works


TO THE PROBLEM







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%


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%


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


“Medical Home” for Patients with Risk Score ≥50Based on Prior 2-Years of Ambulatory Use


"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


Total 100.0% 2.54

51%





Number ofPC/Spec/OB

ProvidersTouched


Total 100.0% 2.54





Number ofPC/Spec/OB

ProvidersTouched


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%


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


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%


TO THE PROBLEM




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



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


Usual source of careNone 16%Emergency department 42%OPD/Clinic 20%Community based clinic 8%Private/Group MD/other 14%


58%




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%


60%



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



• 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)




• 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





• 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





• 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]


TO THE PROBLEM




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



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




• 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)


TO THE PROBLEM







john billings: towards evidence based policy making

Health & Medicine

real patients

patients expendituressource

policy wonk frenzy

substantial portion

emerging body of literature

state level

political administration

policy makingprogramdesign