Identifying and Intervening with Patients at High Risk of Hospital Admission Academy Health Annual Research Meeting, June 5 th 2007 Maria C. Raven, MD,

Identifying and Intervening with Patients at High Risk of Hospital

AdmissionAcademy Health Annual Research Meeting,

June 5th 2007

Maria C. Raven, MD, MPH, MScJohn C. Billings, JD

Mark N. Gourevitch, MD, MPHEric Manheimer, MD

NYUMedicalCenter

Bellevue Hospital Center

High Cost Care Initiative (HCCI): Research Initiative at Bellevue Hospital Center, NYC

Supported by United Hospital Fund Goals:

Characterize high-cost patients with frequent hospital admissions

Use data to inform intervention to reduce admissions/costs and improve care

What we’re going to cover

Why focus on high cost Medicaid patients?

How can we target high cost patients to identify them for interventions?

What we have learned from identifying patients?

What are the next steps?

High Cost Medicaid Patients: the 80-20 rule

NYC MEDICAID SSI DISABLED ADULTSMedicaid Managed Care “MMC” Mandatory [Non-Dual, Non-HIV/AIDS, Non-SPMI] 2003- 2004

3.0%

30.0%

7.0%

25.9%

10.0%

17.0%80.0%

27.1%

0%

20%

40%

60%

80%

100%

Pe

rce

nt

of

To

tal

Patients Expenditures

Source: NYU Center for Health and Public Service Research, UHF, NYSDOH, 2006.

72.9%

Not only is it where the money is… These are some of the patients with the greatest

need Many moving into managed care

What used to be “revenue” is now “expense” Improved care offers potential for cost savings

Why Focus on High Cost Cases?

Predictive algorithm can identify high-risk patients

Predictive algorithm created by John C. Billings identifies Medicaid patients at high-risk for hospital admission in next 12 months

Algorithm generates risk score from 0-100 for every patient in a dataset Based on prior utilization Higher risk scores (>50) predictive of higher risk of

admission in next 12 months

(Reference)Admission

Year 4 Year 5Year 3Year 2Year 1

General Approach for Development of Risk Prediction Algorithm

Examine utilization for prior 3+ years

(Reference)Admission

Year 4 Year 5Year 3Year 2Year 1

General Approach for Development of Risk Prediction Algorithm

(Reference)Admission

Examine utilization for prior 3+ years

Predict admission next 12 months

Year 4 Year 5Year 3Year 2Year 1

General Approach for Development of Risk Prediction Algorithm

Bellevue-specific predictive algorithm

Pulled last five years of Bellevue’s Medicaid billing data Inpatient, ED, outpatient department

Logistic regression created Bellevue-specific case-finding algorithm

Created risk scores (0-100) applicable for any patient with a visit in the past 5 years

Cohort with risk scores>50 = high risk for admission in next 12 months

Subject Enrollment

Cross-checked all admitted patients against our high-risk cohort every 24 hrs

Identified and interviewed 50 such patients and their providers during hospital admission

Determined medical/social contributors to frequent admissions Qualitative/quantitative measures

Inclusion/Exclusion criteria

Ages 18-64 Medicaid fee-for-service visit to Bellevue from

2001-2005 Excluded:HIV, dual eligible, institutionalized

when not hospitalized, unable to communicate

Patients enrolled when algorithm-predicted admission occurred

Interview instruments

Quantitative data from 50 patients Demographics SF-12 (health and well-being) Usual Source of Care BSI-18 (anxiety/depression/somatization) Perceived Availability of Support Scale (social support) Patient Activation Measure WHO-ASSIST (substance use) Medications (adapted from Brief Medication Questionnaire)

Qualitative data from 47 patients, 40 physicians and 16 social workers

36,457 adult fee-for service Medicaid patients with visit to

Bellevue, 2001-2006

2,618 with algorithm-based risk score>50

139 admitted during 2-month study period

50 patients consented and interviewed

•89 ineligible or discharged prior to approach

•11 refusals

Recruitment scheme for Bellevue pilot project

Billings’ algorithm

Daily computer query checked past 24 hours’ admissions against 2,618 high-risk patients

Strength of algorithm

PPV=0.67 Of all admitted high risk patients, over 20 bounce-

backs among 16 patients Of these 16 patients, 9 eligible, 8 interviewed 5 patients had >1 bounce-back during study period

Source: NYU Center for Health and Public Service Research, UHF, NYSDOH, 2006.

Some representative patients…

Mr. O

58 y/o man with COPD and CHF Lives with daughter Feels hospital admission is unavoidable when

he has difficulty breathing Does not seek intervention at symptom onset

from primary doctor Multiple admissions for COPD and CHF

Mr. R

History of over 30 detox admissions One rehab Homeless on street Depression No other medical problems

Ms. C

Severe lupus Severe pain Outpatient doctors won’t prescribe her the

narcotics she wants/needs Repeated admissions for lupus flare and pain

control Often with 24-48 hour stays and no changes to

outpatient regimen

Demographic characteristics

Characteristic % of total

Male

Age in years 18-34 35-49 50-64Mean age=44.3

Ethnicity African American Hispanic White Other

72%

20%42%38%

24%54%14%8%

Education and work history

Characteristic % of total

Education Less than high school High school/GED or greater Unknown

Income source None Public Assistance Social security Work Friends/family

60%36%4%

8%34%38%4%12%

Diagnoses

Characteristic% ofTotal

Any chronic disease 68%Multiple chronic disease 44%

Stroke 6%Cancer 36%

Any mental illness 62%Schizoprhenia 10%Psychoses 20%Bi-polar/major depression 28%

Alcohol/substance abuse 66%

Mental illness or Alc/substance abuse 82%

Self-rated health

Characteristic % of total

General Health Status Excellent/Very good Good Fair/Poor

Health Limits Moderate Physical Activity A Lot A Little Not at all

6%24%70%

45%35%20%

Housing

Characteristic % of total

Current Housing Status Apartment/home rental Public Housing Residential Facility Staying with family/friends Shelter Homeless

34%4%2%24%8%28%

60%

Housing

Disproportionate admissions for substance use, mental illness, and substance use-related medical problems among homeless subjects

Similar differential in claims data

% of Total

CharacteristicPermanent

Housing

Staying WithFriends or

Family

Homelessor

In Shelter

Any chronic disease 85% 83% 39%Multiple chronic disease 65% 50% 17%

Stroke 10% 8% 0%Cancer 70% 17% 11%

Any mental illness 55% 75% 61%Schizoprhenia 5% 0% 22%Psychoses 15% 25% 22%Bi-polar/major depression 15% 50% 28%

Alcohol/substance abuse 45% 58% 94%

Mental illness or Alc/substance abuse 65% 83% 100%

Substance use: ASSIST data

74% had mid-high substance use risk scores (37/50) Risk for harmful use/dependence with related social,

legal, health problems 14% tobacco only (7/50) 60% multiple substances (30/50)

Majority tobacco and alcohol, followed by cocaine and opioids

7 pts had used IV drugs

Mental Health

SF-12 Mental Composite Score Lower scores = higher levels of anxiety and

depression Compared to the general US population:

38% (19/50) scored below the 25%ile 38% scored below the median

BSI-18 “cases” at high risk for psychopathology based on anxiety, depression, and somatization summary score 68% (34/50) cases

Characteristic % of total

Usual Source of Care

None 22%

ED 40%

Hospital outpatient 30%

Other 8%

Source: High Cost Medicaid Project – Bellevue Hospital Center, NYU Center for Health and Public Service Research, 2006

Usual Source of care

Access to care

Source: High Cost Medicaid Project – Bellevue Hospital Center, NYU Center for Health and Public Service Research, 2006.

% of Total

CharacteristicPermanent

Housing

Staying WithFriends or

Family

Homelessor

In Shelter

Usual source of care

None 15% 17% 17%

Emergency department 15% 50% 67%

OPD/Clinic 25% 25% 11%

Community based clinic 20% 0% 0%

Private/Group MD/other 25% 8% 6%

Social isolation

Characteristic % of total

Marital Status Married/living with partner Separated/divorced Widowed Never married

Lives aloneNo close friends/relativesTwo or fewer friends/relativesLow perceived availability of support

14%26%4%56%

52%16%48%38%

Medicaid expenditures prior year

CharacteristicMeanCosts

Bellevue costs prior yearInpatient $37,418Emergency department $174Primary care $168Specialty care $150Outpatient substance abuse treatment $343Outpatient mental health treatment $299Other costs $636

Total costs prior year $39,188

How much can we pay for an intervention, and still expect to save? (or break even) Depends on:

Risk score level Projected reduction in inpatient admissions in the

following year Based on annual Medicaid expenditures in our

cohort: 25% reduction in future admissions over 1 year allows

intervention spending of $9350 per patient

Limitations

Observational study-no control group Limited to English and Spanish speaking, non-

HIV, Medicaid fee-for service Bellevue Hospital population

Urban, underserved

Conclusions and Implications

• Patients with frequent hospital admissions comprise small percentage of all patients, but account for disproportionate share of visits and costs.

• Social isolation, substance use, mental health, and housing issues were prevalent in our study population• Cited by patients/providers as contributing substantially to their

hospital admissions.

• Interventions focused on more effective management of their complex issues could result in cost-savings via decreased utilization and improved health.

Next Steps

Intervention project planning

Intervention being informed by: Pilot data Partnership with providers of homeless services Successful components of similar programs in

other safety net settings around country* Meetings with community providers (CBOs) of

other services (e.g. substance use, mental health, HIV)

*Chicago Housing for Health Partnership, California Frequent Users of Health Services Initiative www.chcf.org

Bellevue intervention project model

Begin at patient’s bedside in hospital, continue after his/her discharge into the community Housing component

Flexible, intensive care management model, multi-disciplinary team approach, tailored to needs of each patient Bellevue-based team will partner with CBOs

Thank You

John C. Billings, JD Marc N. Gourevitch, MD, MPH Lewis R. Goldfrank, MD Mark D. Schwartz, MD Eric Manheimer, MD United Hospital Fund Supported in part by CDC T01 CD000146

Bellevue Hospital Intervention Project

Hospitalized high-risk patients identified using predictive algorithm

Small comprehensive multi-disciplinary team Intensive assessment, arrange and follow to ensure

and assist with provision of post-discharge support Housing, residential substance abuse treatment,

community based mental health treatment, specialized medical outpatient care

Provision of temporary housing while awaiting supportive housing placement/prompt placement into permanent housing

Bellevue Intervention Randomized Controlled Trial

12-month follow-up measures collected

Intervention team intensive care managementfor 12 months

In addition, health services use/costs,and intervention costs tracked

Baseline measuresIntervention team assigned, needs assessment

If homeless, Common Ground to bedside: Housing application begins: patient d/c to stabilizaition housing

150 subjects: Intervention

Usual care for 12 monthsIntervention team to track health services use

and related costs

Baseline measuresFollow-up information

150 subjects: Usual Care

Consent obtained25 subjects enrolled/month for 12 months

Randomization

Medicaid/UninsuredAlgorithm-based risk score>50Admitted to Bellevue Hospital

Bellevue intervention project baseline measures (RCT)

Baseline assessments: Self-report generated Charlson Comorbidity Index:

patient-reported disease severity measure predictive of 1-year mortality

Socio-demographic measures (e.g. age, gender, income, education)

Diagnoses obtained from subject’s electronic medical record

Bellevue intervention project baseline measures (RCT)

Baseline assessments (validated tools): Health and daily functioning Substance use Mental Health Support Scale Usual Source of Care Housing status/living situation Common Ground in-depth assessment

Bellevue intervention outcome measures

Primary outcome Hospital admissions and associated expenditures

Secondary outcomes Other health services (ED, outpatient clinics) utilization Other health services expenditures Intervention costs Housing status Change in psychosocial variables Appt adherence Benefits enrollment Entry into substance use services

The intervention must pay for itself Central goal: intervention that generates more savings to the

delivery system that it costs to implement and sustain. Eliminate even small % admissions and substantial cost savings

can be had. Comprehensive economic analysis planned that considers

Changes in the numbers of inpatient admissions, ED visits, and outpatient visits during the intervention period in addition to their related expenditures

All costs related to the intervention. Ability of intervention to succeed in this goal will help determine

whether it is Sustainable Exportable to other sites.

Admission diagnoses: 30/50 (60%) homeless/precariously housed

23/30 (82%) : Substance use, psychiatric, medical condition related to substance use 9 detoxification services 3 alcohol/drug withdrawal or intoxication 4 psychiatric 7 drug/alcohol-related medical diagnoses

CHF, trauma, chronic septic joint, cellulitis

5/30: infected ulcer, chest pain, catheter infection, GI bleed, COPD All with past or current substance use

Admission diagnoses, 22/50 (44%) housed 1 Diabetes/coagulopathy 3 Lupus 5 Cancer 1 Dialysis/pain medication related 3 non-compliance resulting in disease exacerbation

anemia, adrenal crisis, gastroparesis 2 Alcohol complications

Hepatitis and ESLD 3 infections (2 PNA, 1 cellulitis) 2 COPD/asthma 1 ortho 1 psych

Admission Diagnosis # of total

Diagnoses Cancer

Lupus erythematosos Infection Pneumonia Cellulitis/foot ulcer Dialysis catheter Septic joint (IVDU) Diabetes mellitus Ulcer COPD/asthma CHF Epilepsy Fracture non-union Adrenal Insufficiency Anemia Chest pain (ACS) End-stage liver disease Psychiatric Detoxification services Alcohol withdrawal/intoxication Trauma Alcoholic hepatitis

5 3 8 2 4 1 1 2 1 4 1 1 1 1 1 1 1 5 9 3 2 1

Medication

43% on medication at admission had missed at least one dose in prior week

Most common reasons inability to pay for prescriptions (4) forgetting to take a dose (3) being unable to get to clinic or hospital for refills or

medication administration (3) side effects (3) substance abuse (3)