Assessing MHICM: Program Effects on Mental Health Care Utilization and Costs*, ** *Funding from HSR&D grant IIR 06-115 and the VISN5 MIRECC **Data provided.

Assessing MHICM: Program Effects on Assessing MHICM: Program Effects on Mental Health Care Utilization and Mental Health Care Utilization and

Costs*Costs*,,****

*Funding from HSR&D grant IIR 06-115 and the VISN5 *Funding from HSR&D grant IIR 06-115 and the VISN5 MIRECCMIRECC

**Data provided by the following VA research centers: **Data provided by the following VA research centers: SMITREC, NEPEC, and HERCSMITREC, NEPEC, and HERC

Investigators: Eric Slade1,2

Lisa Dixon1,2

Marcia Valenstein3,4

John McCarthy3,4

Analysts: Stephanie Visnic3, Rose Ignacio3, Deborah Welsh3, Lan Li1,2

1VISN 5 Mental Illness Research and Education Clinical Center, Baltimore2University of Maryland School of Medicine3VA Serious Mental Illness Treatment Research and Evaluation Center, Ann

Arbor4University of Michigan School of Medicine

The MHICM ProgramThe MHICM Program• A psychiatric “hospital without walls”A psychiatric “hospital without walls”

• Uses the Assertive Community Treatment (ACT) model. Uses the Assertive Community Treatment (ACT) model. • Similar in staffing intensity to inpatient psychiatric careSimilar in staffing intensity to inpatient psychiatric care• Team-based mobile careTeam-based mobile care• Small caseloads per team memberSmall caseloads per team member• Team available 24/7Team available 24/7• Care is comprehensive Care is comprehensive

• Improves patients’ quality of life and Improves patients’ quality of life and satisfaction with care, and reduces their satisfaction with care, and reduces their inpatient utilizationinpatient utilization

• 1980’s VA randomized trials of Intensive Psychiatric 1980’s VA randomized trials of Intensive Psychiatric Community Care (IPCC) (Rosenheck et al., 1995 & 1998)Community Care (IPCC) (Rosenheck et al., 1995 & 1998)

MHICM

The MHICM ProgramThe MHICM Program

• Formally implemented as MHICM in Formally implemented as MHICM in FY2000FY2000– Implemented “high hospital use” entry criterionImplemented “high hospital use” entry criterion

• Rapid growthRapid growthFY00FY00 FY07FY07

TeamsTeams 4646 100100ClientsClients 2,6552,655 7,6097,609CostCost $14.5 mil.$14.5 mil. $46 mil.$46 mil.

• <10% of eligible veterans have <10% of eligible veterans have enrolledenrolled

• MHICM programs require a max. client-staff MHICM programs require a max. client-staff ratio of 12 to 1 ratio of 12 to 1

ObjectivesObjectives

• Estimate the effects of MHICM on Estimate the effects of MHICM on mental health services utilization mental health services utilization during the first 12 months following during the first 12 months following clients’ first enrollment in MHICMclients’ first enrollment in MHICM• psychiatric inpatient days of staypsychiatric inpatient days of stay

• partial hospital program days partial hospital program days

• other outpatient mental health days other outpatient mental health days

• Assess the VA cost consequences of Assess the VA cost consequences of the MHICM programthe MHICM program

Inpatient Psychiatric Utilization in the Prior Year

0

LowerHigherr

> 0

< 0

MHICM’s Net Impact on Psych CostsMHICM’s Net Impact on Psych Costs

Hypothesized

A

B

C

Net

Imp

act

($)

Declining VA psychiatric inpatient use Declining VA psychiatric inpatient use may reduce savings achieved with may reduce savings achieved with MHICMMHICM

FY00FY00 FY07FY07 %%ΔΔLOS (Days)LOS (Days) 15.0 15.0 11.411.4 -24-24Bed Census Bed Census 4,106 4,106 2,9582,958 -28-28Bed Days/Veteran Bed Days/Veteran 22.022.0 16.816.8 -24-24

Trend in VA Psychiatric Trend in VA Psychiatric Inpatient UseInpatient Use

$

0

Time

VA Trend in Inpatient Psych Days

Per Patient

Net Savings from MHICM ($)

Net Savings

Net Costs

Expected Cost Consequences of MHICM

Study DesignStudy DesignRetrospective observational designRetrospective observational design

Sample: MHICM-eligible VA patients in Sample: MHICM-eligible VA patients in FY01 to FY04FY01 to FY04

““Intervention”: Enrollment in MHICM Intervention”: Enrollment in MHICM

““Comparison”: Usual care Comparison”: Usual care

Follow-up period: The 12-month period Follow-up period: The 12-month period following following eithereither MHICM initiation MHICM initiation oror becoming MHICM-eligiblebecoming MHICM-eligible

Study TimelineStudy Timeline

MHICM Enrollees

MHICM Eligible Non-Enrollees

-12

-12

0 +12

+120

Months

Months

Enrolled

Eligible

“High Hospital Use”

“High Hospital Use”

SampleSampleData sourcesData sources

• VA National Psychosis Registry (SMITREC)VA National Psychosis Registry (SMITREC)

• VA MHICM enrollment data archive (NEPEC)VA MHICM enrollment data archive (NEPEC)

• VA HERC Average Costs data archiveVA HERC Average Costs data archive

Inclusion criteriaInclusion criteria• Schizophrenia or bipolar disorder diagnosisSchizophrenia or bipolar disorder diagnosis

• Residence within 60 miles of a VA hospitalResidence within 60 miles of a VA hospital

• Recent history of “high hospital use” Recent history of “high hospital use” • Inpatient psychiatric utilization of >30 days or ≥3 Inpatient psychiatric utilization of >30 days or ≥3

stays in the past 12 monthsstays in the past 12 months

SampleSample• 2,102 new MHICM clients 2,102 new MHICM clients

• 25,630 MHICM-eligible non-enrollees25,630 MHICM-eligible non-enrollees

EstimationEstimation

Potential selection biasPotential selection bias

Enrollment into MHICM could be Enrollment into MHICM could be related to severity of illness or need related to severity of illness or need for MHICMfor MHICM

EstimationEstimation• Want to estimate E(y|x,M), where: Want to estimate E(y|x,M), where:

yyii = = αα00 + + αα11’x’xii + + δδMMii + u + uii..– δδ is is the average effect of MHICM on study the average effect of MHICM on study

outcome y. outcome y. – E(u)=0, Cov(M,u)=0 are key assumptions of E(u)=0, Cov(M,u)=0 are key assumptions of

model.model.

• Initiation into MHICM services: Initiation into MHICM services:

Pr(MPr(Mii =1) = F( =1) = F(ββ00 + + ββ11’z’zii + v + vii))

• If Cov(v,u) If Cov(v,u) ≠ 0: Cov(M,u)≠ 0: Cov(M,u) ≠ 0, and ≠ 0, and regression estimates of regression estimates of δδ will be biased will be biased and inconsistent. and inconsistent.

EstimationEstimationPropensity score one-to-one matching Propensity score one-to-one matching was used to “balance” the sample on was used to “balance” the sample on observable characteristics observable characteristics zzIf selection into MHICM is correlated with If selection into MHICM is correlated with unmeasured confounders, unmeasured confounders,

i.e., E(y|z,M,v) i.e., E(y|z,M,v) ≠ E(y|z,M)≠ E(y|z,M)

propensity score matching will not alleviate biaspropensity score matching will not alleviate bias

→→ Method of “instrumental variables” was used to Method of “instrumental variables” was used to further minimize selection biasfurther minimize selection bias

But, there is another complication…But, there is another complication…

02

.0e

-05

4.0

e-0

56

.0e

-05

8.0

e-0

5D

en

sity

0 50000 100000 150000 200000 250000Mental Health Costs 12 Months Post Index Date

EstimationEstimation• IV methods with non-linear outcomes IV methods with non-linear outcomes

required modification of the modelrequired modification of the model• Terza, Basu, Rathouz, J. Health Econ, Terza, Basu, Rathouz, J. Health Econ,

20082008

a)a) Estimate: Estimate:

b)b) Obtain: Obtain:

c)c) Estimate: Estimate: 0 1 ˆi i i i iy x M v 0 1ˆ ˆ

i i iv M z 0 1i i iM z v

EstimationEstimation– IV model requires that z include at IV model requires that z include at

least one variable that is not in xleast one variable that is not in x

– These “instruments” must be These “instruments” must be correlated with M but not with y correlated with M but not with y conditional on Mconditional on M• For tests of these assumptions, see Baum For tests of these assumptions, see Baum

et. al., Stata Journal, 7(4), 2007. et. al., Stata Journal, 7(4), 2007.

– Instruments:Instruments:• distance to the nearest MHICM team and distance to the nearest MHICM team and • whether a MHICM team was onsite at the whether a MHICM team was onsite at the

VA hospital where client had last psych VA hospital where client had last psych inpatient stay inpatient stay

EstimationEstimation• Used two-part generalized linear model Used two-part generalized linear model

(GLM)(GLM)P(y) > 0 (vs. 0) modeled as a normally P(y) > 0 (vs. 0) modeled as a normally

distributed binary random variable distributed binary random variable

E(y | y > 0) modeled as a gamma distributed E(y | y > 0) modeled as a gamma distributed random variable with a “log link”random variable with a “log link”

ln{ E(y) } = ln{ E(y) } = αα00 + + αα11’x’xii + + δδMMii

where y is gamma distributed. where y is gamma distributed. • To calculate averages, used: To calculate averages, used:

E(y) = P(y>0)×E(y|y>0)E(y) = P(y>0)×E(y|y>0)

One-to-One MatchingOne-to-One Matching

Pre-matchingPre-matching• 2,102 new MHICM clients 2,102 new MHICM clients

• 25,630 MHICM-eligible non-enrollees25,630 MHICM-eligible non-enrollees

Post-matchingPost-matching• 2,102 new MHICM clients 2,102 new MHICM clients

• 2,102 MHICM-eligible non-enrollees2,102 MHICM-eligible non-enrollees

Descriptive characteristics pre-/post- matchingUnmatched Sample

(N=28,032)Matched Sample

(N=4,204)  No MHICM MHICM No MHICM MHICMMale % 92.7 90.6 * 90.2 90.6Age in years 51.9 51.3 * 51.2 51.3Any inpatient psych care past 30 days % 45.3 67.3 * 66.0 67.3

Inpatient psych days past 12 months 51.5 68.5 * 68.1 68.5Inpatient psych stays past 12 months 2.7 2.7 2.7 2.7Partial hospital days past 12 months 3.3 9.8 * 10.6 9.8Schizophrenia dx % 62.2 80.3 * 81.4 80.3Homeless past 12 months 36.2 26.5 * 27.6 26.5% with a service connected disability rating >50% 48.3 54.3 * 54.8 54.3Substance use dx % 57.5 47.3 * 48.2 47.3Charlson comorbidity index 0.75 0.70 * 0.70 0.70

Died during 12 month follow-up period % 10.7 4.4 * 4.4 4.4 * Differs from “No MHICM” at P<.05

Mean use of mental health services during the first 12-months post-enrollment [matched sample, N=4,204]

No MHICM(N=2,102)

MHICM(N=2,102)

Any Service Use (%)

Inpatient Psychiatric 57.9 61.2a

Partial Hospital 18.4 29.9a

Other Mental Health Outpatientb 85.3 80.5a

Nursing Home 10.4 4.3a

Days of Service Usec

Inpatient Psychiatric

Per year 50.1 36.3a

Per stay 26.3 15.1a

Partial Hospital 59.9 52.3a

Other Mental Health Outpatientb 19.2 17.5a

Nursing Home 184.4 98.8a

a Different from No MHICM at P<.05

b All non-MHICM outpatient mental health services

c Among persons with some use of services in category

Estimated marginal effects of MHICM services on use of other mental health services in the first 12 months post enrollment, by service category [matched sample, N=4,204]

Standard Instrumental Variables

Service Category μ0a ME t P ME z P

Any Service Use

Inpatient Psychiatric 0.648 0.033 * 2.11 0.035 -0.084 -1.57 0.117

Partial Hospital 0.169 0.128 *** 10.16 <0.001 0.486 *** 24.08 <0.001

Other Mental Health Outpatientb 0.927 -0.050 *** -4.41 <0.001 -0.256 *** -6.89 <0.001

Nursing Home 0.061 -0.047 *** -6.72 <0.001 -0.011 -0.51 0.607

Days of Service Usec

Inpatient Psychiatric

Per Year 56.0 -7.5 ** -3.22 0.001 -29.5 ** -2.73 0.006

Per Stay 30.4 -9.1 *** -7.18 <0.001 -21.7 *** -3.76 <0.001

Partial Hospital 38.5 -5.8 -1.67 0.095 -1.1 -0.10 0.924

Other Mental Health Outpatientb 14.7 -0.1 -0.14 0.890 7.9 *** 3.54 <0.001

Nursing Home 159.1 -28.1 1.61 0.107 -94.3 -1.37 0.171* P<.05; ** P<.01; *** P<.001a Estimated average among MHICM-eligible non-enrolleesb All non-MHICM mental health outpatient servicesc Estimates for persons with any service use in each category

Adjusted effects of MHICM on average service days, first 12 months post-enrollment (matched sample)*

Without MHICM

With MHICM Δ %Δ

StandardInpatient psychiatric 28.0 24.2 -3.8 -14%

Partial hospital 7.0 10.6 +3.6 +52%

Other mental health outpatient 16.1 14.7 -1.4 -9%

Instrumental VariablesInpatient psychiatric 28.0 14.8 -13.2 -47%

Partial hospital 7.0 28.0 +21.0 +300%

Other mental health outpatient 16.1 10.3 -5.8 -36%

* N=4,204

Adjusted effects of MHICM on VA costs, first 12 months post-enrollment (matched sample)*

Without MHICM

With MHICM Δ %Δ

StandardTotal cost $31300 $32541 +$1241 +4%

Inpatient psychiatric $25913 $19976 -$5937 -23%

Partial hospital $1007 $1585 +$577 +57%

Other mental health outpatient $1365 $1254 -$111 -8%

Instrumental variablesTotal cost $31300 $33666 +$2366 +7%

Inpatient psychiatric $25913 $16985 -$8927 -34%

Partial hospital $1007 $5468 +$4460 +443%

Other mental health outpatient $1365 $1447 +$83 +6%

N=4,204

Average MHICM cost is $7669

-$20,000

-$10,000

$0

$10,000

$20,000

$30,000

$40,000

360 330 300 270 240 210 180 150 120 90 60 30

Net savings from MHICM during the 1st year of Net savings from MHICM during the 1st year of enrollmentenrollment

Average effect

Clients' Inpatient Psych Days of Stay 1 Yr Prior to MHICM

Inpatient Use in the Year Prior to MHICM

HIGHER LOWER

ImplicationsImplications• MHICM is a cost-effective programMHICM is a cost-effective program

• However, financial savings from MHICM have decreasedHowever, financial savings from MHICM have decreased

• Future expansions should continue to focus on the Future expansions should continue to focus on the disableddisabled

• Enrollment in MHICM increases Enrollment in MHICM increases subsequent use of partial hospitalization subsequent use of partial hospitalization program services program services • Unclear whether this effect is desirable Unclear whether this effect is desirable

• Thousands of MHICM-eligible VA patients Thousands of MHICM-eligible VA patients are not enrolled in MHICMare not enrolled in MHICM• Persons who are homeless, have concurrent substance Persons who are homeless, have concurrent substance

use conditions, and reside further away from MHICM use conditions, and reside further away from MHICM teams may have less access than othersteams may have less access than others

Future WorkFuture Work• What happens to utilization/costs in the What happens to utilization/costs in the

second year of MHICM?second year of MHICM?

• What predicts disengagement from What predicts disengagement from MHICM? MHICM?

• Does fidelity to the ACT model matter? Does fidelity to the ACT model matter?

Thank you!Thank you!

Contact Information: Contact Information: Eric SladeEric Slade

VISN5 Capitol Network MIRECCVISN5 Capitol Network MIRECC

Baltimore, MarylandBaltimore, [email protected]

410-706-2490410-706-2490