Case-Mix and Risk Adjustment in Primary Health Care

Case-Mix and Risk Adjustment in
Primary Health Care
A tool for equitable budget setting, performance
improvement and chronic care management
Jonathan P. Weiner, DrPH
Professor of Health Policy & Management
Johns Hopkins Bloomberg School of Public Health
Baltimore, Maryland 21205 USA
[email protected]
Presented in Lisbon, 29 October 2010
© Copyright 2010, Johns Hopkins University.
Goals of Presentation
• To introduce concepts surrounding population-based risk
adjustment and “predictive risk” modeling in the primary
health care context.
• To describe the Johns Hopkins “ACG” case-mix
adjustment / predictive risk modeling methodology as a
method widely used within primary care sector.
• To show practical, real-world examples of how these tools
can be applied.
• To begin to discuss potential application of these methods
within the Portuguese context.
© Copyright 2010, Johns Hopkins University.
2
Not all persons have the same
need for health care
3
% of Population
% of Resources
1%
30%
10%
72%
50%
97%
© Copyright 2010, Johns Hopkins University.
4
Working Definitions
• Case mix / risk adjustment is the process by which the
health status of a population is taken into consideration
when setting budgets or capitation rates, evaluating
provider performance, or assessing outcomes of care.
• Predictive risk modeling is the prospective (or
concurrent) application of risk adjustment measures and
statistical forecasting to identify individuals with high
medical need who would likely benefit from care
management interventions.
© Copyright 2010, Johns Hopkins University.
5
The risk measurement pyramid
Management Applications
High
CaseDisease
Management
Burden
Single High
Impact
Disease
Users
Users & Non-Users
Population Segment
© Copyright 2010, Johns Hopkins University.
Disease
Management
Practice
Resource
Management
Needs
Assessment
Quality
Improvement
Payment/
Finance
Types of risk adjustment
applications within health care

Financing, Payment,
Planning
 Morbidity-adjusted
capitation

Allocation of budgets

Service targets
Forecasting healthcare
spending
Provider Performance
Assessment
 Profiling



Pay-for-Performance
© Copyright 2010, Johns Hopkins University.
Identification of high risk
patient
Disease management
 Case management




Care Management

Quality

Quality assessment

Quality monitoring
Research and Program
Evaluation
6
7
OVERVIEW OF THE JOHNS
HOPKINS ACG SYSTEM
© Copyright 2010, Johns Hopkins University.
Overview of Johns Hopkins ACG System
• The ACG - “Adjusted Clinical Group” system provides conceptually
simple, statistically valid, and clinically relevant measures of need/risk
for health services.
• The grouping process has been computerized. The “grouper” and
“predictive modeling” software requires diagnosis information from
encounter data, electronic medical records, or insurance records.
• ACGs are generally applied using all diagnoses describing the person.
They do not focus on individual visits. Ideally they are derived from
primary and specialty ambulatory contacts as well as inpatient
• There is a comprehensive ACG “suite” of risk/case-mix measures.
© Copyright 2010, Johns Hopkins University.
8
ACG System’s
International Presence
•
•
•
•
•
•
•
•
•
•
•
Several Provinces in Canada
Numerous County Councils in Sweden
Several Regions of Spain
Multiple Primary Care Trusts in the UK
Two Sickness Funds in Germany
The largest Health Plan in Israel
Two Medical Schemes in South Africa
The Ministry of Health in Malaysia
Active piloting in Brazil and Chile
Research in Lithuania, Korea, Thailand, Taiwan
Interest expressed in numerous other countries
© Copyright 2010, Johns Hopkins University.
9
No similar method is more fully tested by
“real world” applications
•
Billions of dollars per year are now routinely
exchanged using ACGs in US and Canada and
in several other nations.
•
Healthcare of 80+ million patients is actively
managed and monitored using ACGs on
several continents.
•
The practices of hundreds of thousands of
physicians in over a dozen many nations are
now more equitably assessed on an ACG
case-mix adjusted basis.
© Copyright 2010, Johns Hopkins University.
10
The Johns Hopkins ACG System:
11
An Expanding Suite of Measures and Tools
Diagnosis (ICD) based:
• ADGs classify diagnoses into a limited number of clinically meaningful, but not diseasespecific, morbidity groups. (For example “chronic unstable”)
• EDCs classify diagnoses based on specific diseases. They represent disease markers
and can be used to determine disease prevalence (For example, Type I Diabetes, w/o
complications).
• ACGs (Adjusted Clinical Groups) represent a single, mutually exclusive actuarial cell
based on overall disease burden. (For example, 6-9 ADG Combinations, Age >34, 2
major ADGs) The system also includes national references for average cost by ACG
• Dx-PM – a “predictive model” of future risk and need based on ACGs, EDCs and special
high risk markers. (Formerly referred to as ACG-PM.)
Pharmacy based:
• Rx-PM - a predictive model of future risk and expected resources use based only on
pharmacy.
(See http://www.acg.jhsph.edu for more information on methodology).
© Copyright 2010, Johns Hopkins University.
Key components of the Johns Hopkins
ACG System
Resource
Use
$ - Counts
Patient Info
ID – Age – Gender –
Resource Use
ACGs
100
Diagnosis
ICD 9 - ICD 10 - ICPC
ACG
System
EDCs
300 / 30
Markers
Pharmaceuticals
Frailty – Hosdom –
Chronic
Pregnancy - Delivery
ATC
Rx-MG
60
© Copyright 2010, Johns Hopkins University.
Predictive
Models
12
ACGs capture the essence of a person’s
health status
Time Period (e.g., 1 year)
Treated
Morbidities
Visit 1
Diagnostic
Codes
13
Morbidity
Groups
Code A
ADG10
Code B
Visit 2
Code C
ADG21
Visit 3
Code D
ADG03
Clinician
Judgment
© Copyright 2010, Johns Hopkins University.
Clinical
Grouping
ACG
Category
Data
Analysis
14
ADGs for Diabetes: ICD Codes
D iabetes Mellitus
ICD-9-CM
Code
Label
ADG
Code
Label
2500
DIABETES MELLITUS
UNCOMPLICATED
10
CHRONIC MEDICAL:
STABLE
25003
DIABETES MELLITUS
WITHOUT
COMPLICATIONS
UNCONTROLLED
11
CHRONIC MEDICAL:
UNSTABLE
2501
DIABETES WITH
KETOACIDOSIS
09
LIKELY TO RECUR:
PROGRESSIVE
3620
DIABETIC
RETINOPATHY
18
CHRONIC SPECIALTY:
UNSTABLE-EYE
© Copyright 2010, Johns Hopkins University.
Examples of ACG Categories
ACG Code
Description
0200
Acute Minor, Age 2-5 years
0600
Likely to Recur, without Allergies
1722
Pregnancy: 2-3 ADGs, no major ADGs, not delivered
2800
Acute Major and Likely to Recur
4430
4-5 ADGs, Age > 44, 2+ major ADGs
5322
Infants: 0-5 ADGs, 1+ major ADGs, low birthweight
© Copyright 2010, Johns Hopkins University.
15
The EDC tool represents over 200 individual
“disease markers”: The Cardiovascular EDCs
•
•
•
•
•
•
•
CV signs/symptoms
Hypertension
Ischemic heart disease
Congenital heart disease
Congestive heart failure
Cardiac valve disorder
Cardiomyopathy
© Copyright 2010, Johns Hopkins University.
•
•
•
•
Heart murmur
Cardiac arrhythmia
Thrombophlebitis
Generalized
atherosclerosis
• Peripheral vascular
disease
• Edema
16
Co-morbidity is central to understanding
resource use: ACG risk levels and patterns of
resource use at an English Primary Care Trust
Level of Comorbidity
(Based on
ACGs)
High
Hospital
Use
% Pop.
Relative
Ratio
17
Est. % of
Admissions
at PCT
Avg. #
OutPatient
Episodes
/ Yr.
Avg. #
Prescripti
ons / Yr.
2%
11.5
25%
11.0
93
Moderate
17%
3.0
47%
7.1
66
Low
40%
.6
26%
3.0
28
None
41%
>.1
2%
.5
6
Data from several large GP practices within PCT for 2005. N= 20,500 all ages.
© Copyright 2010, Johns Hopkins University.
Distribution (%) of ACG case-mix “morbidity
bands” across UK & US sample populations:
Ages 0-65
40
35
30
25
UK
US-HMO
20
15
10
5
0
1
2
3
4
5
6
ACG- Morbidity Bands
(1- healthy, 6 – sickest)
Source: Forrest et al BMJ 2002.
Data: UK - GPRD, n = 758K. US- private HMO, n = 70K (w/ no Medicaid or uninsured)
© Copyright 2010, Johns Hopkins University.
18
•ACG Risk Scores Help Stratify Resource Use: NHS
Consultant Referral Rates by ACG Morbidity Score*
50
% Patients
referred per
year to one
or more
consultants
40
30
20
10
0.5
1.0
1.5
2.0
ACG Morbidity Score
Source: Forrest, Majeed, Weiner, et al. BMJ 2002: 325;370
(From a sample of 758,000 Electronic Patient Records from GPs offices. GPRD database)
* An “expected resource use” score, where 1.0 is average, is based on patient’s ADGs.
© Copyright 2010, Johns Hopkins University.
2.5
19
20
Risk Adjusted Performance
Profiles of Primary Care
Doctor Groups
© Copyright 2010, Johns Hopkins University.
Comparing primary group actual resource use to
expected use (based on case-mix) in US HMO based 21
on illness burden
Avg. Pt.
Cost per
Month
Unadjusted
cost / HMO
Avg.
Expected use
based on
“Illness
Burden”*
ACG Adjusted
Efficiency Ratio
(actual /expected)
#1
$157
1.22
1.02
1.20
#2
153
1.19
1.21
0.99
#3
144
1.12
0.92
1.22
#4
98
0.76
0.69
1.11
$129
1.00
1.00
1.00
Primary Care
Doctor
Group
Entire HMO
(includes
other groups)
* Case-Mix adjusted expected use based on ACG illness burden of patients in practice.
1.0 is average case-mix. This would be the commissioning budget.
© Copyright 2010, Johns Hopkins University.
•Breakdown of
cost, illness burden and efficiency ratio
for doctor group #3 ( see previous table) by service
22
category
Relative
Cost
ACG Illness
Burden
Efficiency
Inpatient
0.91
0.90
1.01
Primary Care
1.20
1.15
1.04
Surgery
2.23
0.91
2.45
Medical Specialists /
consultants
1.61
0.92
1.75
Lab & x-ray
1.77
0.85
2.08
Pharmacy
.86
0.85
1.01
1.12
0.92
1.22
Type of Service
Total
© Copyright 2010, Johns Hopkins University.
Risk-Adjusted O/E (Efficiency) Profiling Ratios
23
for GPs Across a Primary Care Trust (PCT) in UK
2.0
1.8
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
GP1 GP2 GP3 GP4 GP5 GP6 GP7 GP8
No of referrals No of unique prescriptions / month
Observed = actual avg. use by patients.
Expected = based on ACG case-mix of pts.
© Copyright 2010, Johns Hopkins University.
GP9 GP10 GP11 GP12
No of unique radiology tests
Above 1.0 = higher than expected.
Ambulatory Clinic Pharmacy Cost Profiling in
24
Spanish Region
Pharmacy cost x patient: observed (
) and expected (
Efficiency Index: 0,79
)
Efficiency Index: 1,27
21% undercost
943.000 €
27% overcost
737.000 €
400
350
300
250
200
150
100
50
0
001
002
003
004
005
006
007
008
009
Average
Mean Cost (€)
182,58
291,57
274,75
212,19
337,71
289,03
328,99
287,14
196,36
270,49
Mean cost (€) expected
231,02
271,59
293,94
243,63
296,59
295,57
258,10
280,21
241,01
270,49
Overcost or undercost, related to standard
Efficiency Index
0,79
1,07
Impact (€) 943.068 510.658
© Copyright 2010, Johns Hopkins University.
0,97
0,87
1,14
0,98
1,27
280.254 481.278
715.386
121.540
736.869
1,02
144.487
0,81
281.209
25
Capitation, & Budgeting
& Other Financial Issues
© Copyright 2010, Johns Hopkins University.
Determining the Healthcare Budget
26
for a Population Involves a Variety of Factors
- Available
Budget
- Political
Forces
- Actuarial
Forecasts
© Copyright 2010, Johns Hopkins University.
Size
of the
Healthcare
Pie
Risk adjustment can be used to fairly slice the
health care budget “pie”
Risk
Adjustment
© Copyright 2010, Johns Hopkins University.
27
Some Reasons Why Risk Adjusted Payment & Budgeting28
May Be Necessary
• To protect doctors, clinics, or organizations that care for
costlier than average patient populations.
• To help ensure that government or others that finance
care pay their fair share (neither too high or low).
• To deter providers from selectively attracting healthier
patients.
• To facilitate organizations or providers wishing to
specialize in treating people with higher than average
illness burden.
© Copyright 2010, Johns Hopkins University.
State of Maryland Medicaid program risk adjusted (ACG)
29
payment to capitated “HMO” health plans
Average
Risk
Using ACGs, risk ratios were determined for each contracting managed care organization / health plan.
Expected values were determined separately for the two enrollee groups with this State Medicaid program.
© Copyright 2010, Johns Hopkins University.
30
Care Management &
Predictive Modeling
© Copyright 2010, Johns Hopkins University.
Potential uses of ACG-based predictive risk
modeling in care management
• To identify persons for inclusion in care
management programs:
– multi-disease (case-management) and
– single disease programs.
• To provide information to help manage
their ongoing care.
© Copyright 2010, Johns Hopkins University.
31
ACG Predictive Modeling Combines
Multiple Risk Factors into a Single Risk
Score
Age
Overall
Disease
Burden
Gender
Complicated
Pregnancy
Marker
Frailty
Selected
Medical
Conditions
© Copyright 2010, Johns Hopkins University.
Hospital Dominant
Conditions
Pharmacy (Rx)
Information
(optional)
32
Distribution of Risk Scores Across a
Population
9500
40
Risk Score
30
20
10
2.5
0
0
© Copyright 2010, Johns Hopkins University.
2000
4000
6000
Cumulative % x 100
8000
10000
95th Percentile
33
Year-2 Costs of Persons with High (Year-1) ACG Risk
34
Score (Dx-PM) vs. Other Persons
Risk Ratio (High/Low) for Total Costs is 14.8
For Rx Costs Ratio is 18.3
Data derived from a 115,000 member US Medicaid HMO. “High” is the top 1% of ACG-PM
scorers.
© Copyright 2010, Johns Hopkins University.
Using PM risk stratification to target and stratify disease
management program participation for chronic
35
conditions
% Enrollees in ACG Risk
Category
Resource Use of Cohort
Relative to Total Population
Condition of
Interest
Low
Med.
High
Low
Med.
High
Diabetes
44.97
42.1
11.9
1.34
4.90
7.44
Congestive
Heart
Failure
19.75
53.5
26.75
1.14
6.02
7.93
Tier 1 Tier 2 Tier 3
© Copyright 2010, Johns Hopkins University.
35
ACG-PM Risk Score (Baseline Year) as Predictor of
Hospital Use (Year 2)
Type of Use
ACG-PM
Top 5%
ACG-PM
Lower 95%
Risk
Ratio
>1 Hospitalization
27.0%
5.7%
4.7
>1 ICU admit
1.9%
0.2%
8.1
>1 CCU admit
2.0%
0.2%
9.9
Source: British Columbia linked database (n=3.8M) 1998-99
© Copyright 2010, Johns Hopkins University.
36
ACGs Applied to EMR data to understand
morbidity patterns among regions
Fitri , 2010. Presentation at Patient Classification System
International, Munich Germany.
© Copyright 2010, Johns Hopkins University.
37
1.8
1.8
1.6
1.6
ACG Morbidity Index (1995/96)
Rural Areas
Urban Areas
1.4
1.4
1.2
1.2
1
1.0
0.8
0.8
0.6
0.6
0.4
0.2
ACG Morbidity Index
Premature Mortality Rate
0.4
0.2
Crude Premature Mortality Ratio 1991-95
ACGs as a measure of need using Mortality Rates
in rural & urban Manitoba (Reid et al, 2000)
Some Practical / Data Issues
39
• Requires computerized diagnoses from ambulatory care
sector
– Ideally in-patient data too.
• Ideally 12 –24 months of data.
• Diagnosis codes in ICD-9 or ICD-10, ICPC, or Read code
format.
• “Outcome measures” (e.g,. clinical or cost) on each
patient desirable. If not available, can apply resource
use weights from US or other settings.
• Possible applications with pharmacy data as source of
risk information. (requires WHO –ATC codes)
© Copyright 2010, Johns Hopkins University.
Possible Applications in Portugal
• Population based need-assessment across patient
populations (e.g. regions, vulnerable patient groups)
• Assessing performance of providers (e.g. primary care
clinics, USFs, regional health administrations).
• Resource allocation / budgeting across clinics, regions
or other care units.
• “Predictive Risk” measurement to assist in chronic care
management.
• Quality improvement comparisons.
© Copyright 2010, Johns Hopkins University.
40
Opportunities for learning more about Johns
41
Hopkins ACGs
• Web Site:
– www.acg.jhsph.edu
• To learn more, contact:
– Dr. Karen Kinder – Director, ACG International
• [email protected]
– Dr. Patricio Muñiz - Senior Consultant for Portugal, Spain
and Latin America, ACG International
• [email protected]
© Copyright 2010, Johns Hopkins University.
`