Managing Cost and Outcome of Managing Cost and Outcome of Post Hospital CarePost Hospital Care

Reg Warren PhD,Rick Glanz and Amy Leibensberger

The National Predictive Modeling SummitSeptember 13, 2007

The dilemma for post hospital care?The dilemma for post hospital care?

MCO’s spent $7.9B on SNF and Home Health in 2005 (MedPac 2007)

1/3 of hospitalized seniors will receive post acute care

1/3 of those will not go to the most appropriate setting

Misalignment of incentives: per diem High degree of practice variation Over-utilization unnecessarily exposes members

to institutional risks

Managing Cost and Outcome of Managing Cost and Outcome of Post Hospital CarePost Hospital Care

Diagnosis vs Function The Predictive Model

Regression Severity adjustment

Application Real-time decision support Retrospective comparison

Influence on cost and outcome

People get postacute care because they are frail and Care Dependent:

Function Driven

People go to the Hospital because they are Sick:

Disease Driven Treat the Illness

Restore the Ability

Acu

teP

osta

cute

Key PredictorKey Predictor

Functional MeasurementFunctional Measurement

EatingGroomingBathingDressing Upper BodyDressing Lower BodyToiletingBladder ManagementBowel ManagementBed, Chair, WC TransferToilet TransferTub/Shower TransferWalk/WCStairs

• Expression• Comprehension• Social Interaction• Problem Solving• Memory

Functional Independence Measure - FIM (18-126)

Function Burden of Care Function Burden of Care

Functional needs Drives >90% of skilled utilization Admit function- predict outcome Discharge setting : 5 points FIM equate

to one hour caregiver burden/day

Calibrate the ContinuumCalibrate the Continuum

Home

Post-acute

Other

HH

SNF

20% of SNF patients would get the same result at home

SNF LOS can be reduced by 40% without impacting functional result

HH Cost can remain stable even with increased referrals

Most Acute Rehab cases would get the same result in SNF

Acute DischargesSr. Population

Managing Cost and Outcome of Managing Cost and Outcome of Post Hospital CarePost Hospital Care

Diagnosis vs Function The Predictive Model

Regression Severity adjustment

Application Real-time decision support Retrospective comparison

Influence on cost and outcome

• Leader in post-acute outcome measurement since 1995

• Manage over 900,000 Senior lives in SNF, Acute Discharge, Acute Rehab, Home Health

• Database of 250,000+ post-acute cases• Over 30,000 new records added each year

• California, Colorado, Washington, Maryland, District of Columbia, Virginia, Arizona, Pennsylvania and Tennessee

• Kaiser Permanente, PacifiCare, Health Net, Group Health Coop, and AmeriGroup

• MHS Participant

Improvement in Function in SNF: Improvement in Function in SNF: PredictablePredictable

ADMFIM

120100806040200

DIS

FIM

140

120

100

80

60

40

20

0

Correlations

1.000 .818**. .000

500 500.818** 1.000.000 .500 500

Pearson CorrelationSig. (2-tailed)NPearson CorrelationSig. (2-tailed)N

ADMFIM

DISFIM

ADMFIM DISFIM

Correlation is significant at the 0.01 level(2-tailed).

**.

80 yr old female with CHF, cellulitis, UTI

and prior stroke

Discharge Site and Functional LevelDischarge Site and Functional Level(FIM 18-126)(FIM 18-126)

Discharge Site Considerations FIM Score

Home AloneOP Therapy/Home Safety >108

Home with AssistOP (Outpatient) Therapy >90

Home w/Assist or ALFHome Health Services >80

Home, SNF, Custodial, B&C w/24-hour Assistance <79

Diagnosis, medical complexity or other social, caregiver or medical issues may influence the functional level at which the patient is discharged.

ADMFIM

120100806040200

EPIS

OD

E

120

100

80

60

40

20

0

Length of SNF Stay: Length of SNF Stay: Less Predictable Less Predictable

Correlations

1.000 -.265**. .000

500 500-.265** 1.000.000 .500 500

Pearson CorrelationSig. (2-tailed)NPearson CorrelationSig. (2-tailed)N

ADMFIM

EPISODE

ADMFIM EPISODE

Correlation is significant at the 0.01 level (2-tailed).**.

Medical Complexity ScaleMedical Complexity Scale

Level 0: No systemic disease other than primary diagnosis

Level 1: Pre-morbid, inactive and/or irrelevant Level 2: Active, relevant. Not limiting function Level 3: Active, relevant. Limiting function Level 4: Active, relevant. Severely limiting function Level 5: Moribund/Terminal

Relevant ConditionsRelevant Conditions

Restricted Weight BearingRestricted Weight Bearing Pressure Wound: II, III or IVPressure Wound: II, III or IV Vascular (non-pressure) woundVascular (non-pressure) wound IVIV Vent Vent

Hospital: not currently dependent Hospital: not currently dependent Currently DependentCurrently Dependent

Severe ObesitySevere Obesity HemodialysisHemodialysis

Regression: Length of Skilled StayRegression: Length of Skilled StayIndependent

VariableCoefficient Patient’s

Actual Value

Result

Random Error(constant)

24.45

Admission FIM -.234 65 -15.21

Age .034 82 2.78

Days Post Onset

.565 4 2.26

Condition(IV/Obese)

13.1

Predicted Episode LOS =

27.38

X

X

=

=

=

+

+

+

X =

=+

82 y/o femaleUTIAcute: 6 days

(9.24)

(12.02)

(14.28)

How powerful is the model?How powerful is the model?

60%

38%

2%

SMTX Model Treatment Unknown

TherapyIntensity

DisabilityCo-morbidityDPODx (stroke)AgeCondition Grouper

EACH patient is case adjusted:

Impairment Group

Age

DPO

Adm FIM

Med Complex

Pt. #30: CVA ,75 yrs, DPO 12, Adm FIM 30, and Med Complex 4

Pt. # 1: UTI, 82 yrs, DPO 31, Adm FIM 57, and Med Complex 3

Query

LOS: 15 days DC FIM: 50

LOS: 11 days DC FIM: 81

Best Practice Calculation

LOS: 13 days DC FIM: 62.5

LOS: 21 days DC FIM: 50

LOS: 14 days DC FIM: 81

Actual Calculation

LOS: 18 days DC FIM: 62

SMTX National Comparison

(60% most efficient facilities)Actual Values

Pts #2-30

VARIANCELOS: 28%DC FIM: 1%

REGRESSION

Managing Cost and Outcome of Managing Cost and Outcome of Post Hospital CarePost Hospital Care

Diagnosis vs Function The Predictive Model

Regression Severity adjustment

Application Real-time decision support Retrospective comparison

Influence on cost and outcome

Expe

ctat

ions

High Practice VariationHigh Practice Variation

Actual Result

Ass

essm

ent

Treatment

Dis

char

ge

Plan

ning

Admission Discharge

Postacute Episode

Full

Rec

over

yN

o R

ecov

ery

Real-Time Decision Support

Reduced Practice VariationReduced Practice VariationA

sses

smen

t

Mos

t Lik

ely

Res

ult

Treatment

Expe

ctat

ions

Dis

char

ge

Plan

ning

Actual Result

Admission Discharge

Traditional Timeframe

Value Added

Postacute Episode

12

Retrospective ComparisonRetrospective ComparisonSeverity-Adjusted ComparisonSeverity-Adjusted Comparison

Case

s in

SM

TX

Reha

b St

art L

ag

Opt

imal

Reh

ab C

ycle

Acut

al R

ehab

Cyc

le

Reha

b Cy

cle

Varia

nce

Disc

harg

e La

g

Reha

b AL

OS

Expe

cted

Dis

char

ge F

IM

Actu

al D

isch

arge

FIM

FIM

Dis

char

ge V

aria

nce

% D

isch

arge

d to

Com

mun

ity

Ther

apy

Hrs

per D

ay

Jefferson Ave 45 1.4 7.9 10.2 28% 1.1 12.7 101.2 96.0 -5% 0.83 0.97

Mercy Court 67 1.3 11.7 12.1 3% 0.8 14.2 77.0 76.1 -1% 0.83 1.43St AllensGarden RidgeSan AngeloPacific CrestCareBrook

Total/Average

Efficiency Quality

Managing Cost and Outcome of Managing Cost and Outcome of Post Hospital CarePost Hospital Care

Diagnosis vs Function The Predictive Model

Regression Severity adjustment

Application Real-time decision support Retrospective comparison

Influence on cost and outcome

SNF LOS Variance Trend

-10

0

10

20

30

40

50

July

Aug

Sep Oct

Nov

Dec Ja

n

Feb

Mar

Apr

May

June July

Aug

Sep

t

Month

Perc

ent V

aria

nce/

Que

ry

0

0.2

0.4

0.6

0.8

1

1.2

1.4

Ther

apy

Hour

s/Da

y

cycle dcFIM Hours/Day

Client Q3, 2006 Source: SMTX

73

96.7 93.2

105.8115.5

5060708090

100110120130140

SNF Admit SNF DC HH Admit HH DC Follow-up

Assessment

Tota

l FIM

Sco

reFunctional recovery Across Settings:

SNF thru HH to Follow-up*11-01 to 9-05

Source: SMTX

SNF: all dc homeHH: all admitted from SNFFollow –up: all records (N=1259)

Improving Acute DC PlacementImproving Acute DC Placement

49%

58% 58%61% 63%

60%

67%64%

57%

40%

34%37%

32%28% 30% 28% 30% 31%

11%8%

5% 7% 9% 10%5% 6%

12%

0%

10%

20%

30%

40%

50%

60%

70%

80%

J an(N=291)

Feb(N=264)

Mar(N=260)

April(N=247)

May(N=284)

Jun(N=321)

July(N=334)

Aug(N=317)

Sep(N=195)

Home Skilled Other

Influence on UtilizationInfluence on Utilization

Average Medicare PlanLOS: 22 daysSNF Admits/k: 50-65SNF Days/k: 900-1100PMPM: $33

Predictive Model ResultsLOS: 16 daysSNF admits/k: 40-50SNF days/k: 600- 800PMPM: $22.50

Reducing Practice Variation Using Reducing Practice Variation Using Predictive ModelsPredictive Models

11

SNF Days of Treatment SNF Days of Treatment

Patie

nt H

ealth

Impr

ovem

ent

(Fun

ction

al Im

prov

emen

t Mea

sure

men

t)

Pre-Implementation Post-Implementation

SNF Days of Treatment SNF Days of Treatment

Patie

nt H

ealth

Impr

ovem

ent

(Fun

ction

al Im

prov

emen

t Mea

sure

men

t)

Pre-Implementation Post-Implementation

$5,655 average cost per case

23 point average gain in Functional Improvement Measurement (“FIM”), an internationally recognized scale of disability

$4,485 average cost per case (20.7% decrease)

23 point average gain in FIM (unchanged)

One Year