An Innovative Approach to Health Risk Assessment

An Innovative Approach to Health Risk Assessment

I. Overview and Background
describing a patient’s underlying medical condition
Assessment of member health risk has become a
rather than the individual services provided in its
critical need of healthcare organizations. Health
treatment. Because pharmaceutical data can be
plans and other managed care organizations are
incorporated into episodes of care, ERGs also take
increasingly adopting tools that allow them to better
advantage of the information available from a
understand and predict health risks of the members
patient’s use of prescription drugs – a feature that
they enroll and the potential medical care costs
has had only limited or no use in existing health risk
associated with those risks. Member health risk can
assessment models.
vary for a number of reasons, including a person’s
The next section of this paper provides an overview
current health, genetic make-up, socio-economic
of ERGs, including a summary of the data used in its
status, and environment.
development. An assessment of the performance of
Individuals generally do not select health plans and
ERGs along a number of dimensions, including its
medical care providers randomly. They make
ability to predict risk is presented in Section III. The
choices that are most suitable and advantageous to
final section provides a summary and conclusions.
them – attracted by factors such as benefits, access,
and levels of specialization. Simultaneously, plans
and providers may actively seek or attract individuals
of certain risk and avoid others. A health plan or
provider’s costs are affected by the particular
combination of the health risks that their enrollees or
patients represent. Whether to support accurate
payment rates, obtain meaningful comparisons of
provider performance, or identify patients of highest
risk, sound methods of health risk assessment are
critical tools for any health care organization.
Adjusting for differences in health risk can be
thought of as a two-step process. The first step, risk
assessment, involves the measurement of the
expected health care costs or utilization of an
individual or groups of individuals. Risk adjustment
is the mechanism for adjusting for differences in risk,
as measured by the risk assessment. In all
applications, risk adjustment will only be as good as
the underlying risk assessment method.
In this paper, we describe an innovative approach to
health risk assessment called Episode Risk Groups
(ERGsTM). Like many risk assessment models,
ERGs use basic inputs such as the diagnoses
recorded on medical claims and demographic
variables to predict health risk. However, a key
feature of ERGs is its use of “episodes of care” as
markers of risk rather than the diagnoses from
individual medical encounters. By using episodes of
care, the focus is placed on the key information

Confidential and Proprietary

August, 2006
Symmetry
1

---

An Innovative Approach to Health Risk Assessment

II. ERGs
The expense and complexity involved in
A good risk assessment model should have a
administering a model is also an important
number of qualities:
consideration. Further, a model should address
• Predictive Ability – Other factors considered, a
the practical realities in its application, including
model should maximize predictive accuracy –
different requirements for different business
how close actual levels of costs or utilization are
needs, such as for healthcare underwriting.
to those predicted by the model.
• Ability to Restrict Manipulation and Gaming
• Clinical Relevance – The markers used in
– A model should be sufficiently insensitive to
describing risk should be meaningful medically,
differences in diagnostic coding practice.
allowing clinicians and others to interpret and
Developing a risk assessment model that provides
understand the relationship between an
the best solution for each of these criteria is a
observed medical condition and a patient’s risk.
difficult, if not impossible task. Trade-offs are
• Incentives for Efficient and Quality Care – A
inevitable. For example, a demographic model –
method must provide appropriate incentives for
based on a member’s age and sex – is simple to
medical practice. Where discretion is present,
administer, difficult to game, and provides no
the risk assessment should not reward or
disincentives for cost-effective and quality care.

penalize treatment decisions, such as the
However, demographic models add little in terms of
decision to admit a patient to the hospital,
clinical information and also perform poorly in terms
perform a surgery or prescribe a medication.
of predictive accuracy relative to other types of
This is particularly important where results are
models. Alternatively, models based on prior
used for payment purposes or assessing
utilization and treatments may produce superior
efficiency in providing medical care.
predictions. However, these models present
• Retrospective and Prospective Models – Risk
significant issues for incentives for medical care.
assessment can be applied either prospectively
The optimal approach requires an appropriate
or retrospectively. Both types of models have
balance between predictive accuracy, clinical
importance for health care applications.

relevance, and other practical criteria.
Retrospective or concurrent risk assessment
In developing ERGs, we selected a solution to best
uses risk markers for an individual for a base
meet these objectives. ERGs combine a high level
year to produce a measure of risk for that same
of accuracy with clinical relevance. ERGs also
year. A prospective application uses markers
provide appropriate incentives for medical care by
for a base year to measure risk for a future year.
focusing on a member’s clinical conditions rather
Retrospective models are most often used for
than the services involved in treating those
comparisons of provider and health plan
conditions. The flexibility of ERGs supports a wide
performance, such as physician profiling.

range of practical applications.
Prospective models are often applied when
The fundamental building blocks of ERGs are a
setting payment rates and for risk stratification to
patient’s episodes of care – the unique occurrences
support care intervention and disease
of a medical condition or disease and the health
management.
care services involved in diagnosing and managing
• Administrative Practicality – To support
their treatment. The nature and mix of these
general use, the information used in risk
episodes provide a clinical profile for a patient that
assessment has to be routinely available – for
can serve as a marker of their current and future
example, age and sex or the diagnoses,
need for medical care.
procedures and treatments recorded on
administrative medical and pharmacy claims.

Confidential and Proprietary

August, 2006
Symmetry
2

---

An Innovative Approach to Health Risk Assessment

The episodes underlying ERGs are created using
2. ETGs to ERGs – Episodes for each member
the
Episode Treatment Groups (ETG™)
are further categorized into one of 126 episode
methodology developed by Symmetry.
risk groups (ERGs). The ERGs are markers of
member risk and combine ETG episodes of
ETGs – the industry-standard approach to episode
similar clinical and risk characteristics.
grouping – is a basic condition classification
methodology that combines related services into a
3. ERG
Profile – Age, gender and mix of ERGs
medically relevant unit describing a complete
provide a clinical and demographic risk profile
episode of care. The clinical richness of ETGs, its
for a member. Members can be assigned zero,
reliance on information readily available on medical
one, or more ERGs. Members with multiple
and pharmacy claims, and its potential as a tool for
medical conditions will have multiple ERGs.
describing relative patient morbidity makes it a
4. ERG Risk Score – Using pre-determined
sound basis for the development of a health risk
weights and a member’s ERG profile, a risk
assessment model.
score is computed. A member’s risk score is
simply the sum of the weights attached to each
The ERG approach involves four important steps:
ERG and demographic characteristic observed
1. ETGs and Episodes of Care – Using the ETG
in their profile. Retrospective and prospective
methodology, enrollment data and the diagnostic
risk scores are computed for each member.
and procedural information available on medical
and pharmacy claims, health care services for a
Figure 1 provides an overview of the ERG risk
member are first assigned to unique episodes of
assignment process. In the remainder of this
care.
section, we describe each of these steps in greater
detail.



Episode Risk GroupsTM (ERGs):
ETG-based Risk Assessment
Figure 1 -- Assignment of Member Health Risk
Identify OB and
MH Claims and
Members with
Catastrophic
Expenses
Medical
and
Pharmacy
A member's
Encounter
Presence of 1 or
Using pre-
episodes mapped
Data
Medical and
more episodes for a
determined weights
to one of 126
Pharmacy services
member in each
associated with a
Member
Episode Risk
assigned to unique
ERG and age and
particular ERG
Risk Score
Groups (ERGs),
episodes of care
gender, create an
profile, compute a
based on ETG
using ETGs
ERG clinical profile
risk score for each
assigned
for each member
member

Member
Enrollment
Data
ERG "profile"
Risk score is the
ETGs collapsed
describes a
Member
sum of the
Standard
into unique ERGs
members
assigned both a
weights attached
application of ETG
based on clinical,
demographic
retrospective
to each observed
Grouper
statistical, and
characteristics
and prospective
demographic
practical criteria
and observed mix
risk score
and ERG
of ERGs

Confidential and Proprietary

August, 2006
Symmetry
3

---

An Innovative Approach to Health Risk Assessment

1. ETGs and Episodes of Care – the Episode
The key features of ETGs are clusters, anchor
Treatment Groups (ETG) Methodology. The
records, and ancillary records.
fundamental building blocks of ERGs are the
Other than the individual service, the cluster is the
individual ETG episodes of care observed for a
basic unit of an episode. For example, consider a
member. Based on a series of clinical and statistical
patient with a condition requiring a series of
algorithms, ETGs combine inpatient and outpatient
diagnostic tests. The patient’s initial encounter is an
medical and pharmacy services into mutually
office visit with a physician who diagnoses the
exclusive and exhaustive categories called episodes
illness and subsequently orders laboratory and other
of care.1 Examples of ETGs are shown in Table 1.
tests to confirm the diagnosis. ETGs link these
Approximately 600 separate ETGs are defined –
related services to form a cluster. Clinically
each categorized into one of 22 Major Practice
homogeneous clusters are then combined to create
Categories (MPCs) (Table 2).
episodes of care.
Table 1.
Episode clusters are created using anchor records.
Examples of Episode Treatment Groups
Anchor records represent services provided by a
ETG
Description
27
Type I diabetes, with comorbidity
clinician engaging in the direct evaluation,
29
Type II diabetes, with comorbidity

Table 2.
Malignant neoplasm of central nervous system, with
153-01
surgery, with active management
Major Practice Categories (MPCs)
Malignant neoplasm of central nervous system, with
MPC Description
153-02
surgery, w/o active management
1
Infectious Diseases
Malignant neoplasm of central nervous system, w/o
2 Endocrinology
154-03
surgery, with active management
3 Hematology
Malignant neoplasm of central nervous system, w/o
4 Psychiatry
154-04
surgery, w/o active management
5
Chemical Dependency
6 Neurology

168
Migraine headache, common
7 Ophthalmology
169
Migraine headache, complicated
8 Cardiology
9 Otolaryngology
906-03
Ongoing Rx Tx w/o Prov intervention - Migraine Tx
10 Pulmonology
Coronary artery disease, w/o AMI, with coronary
260
11 Gastroenterology
artery bypass graft
12 Hepatology
262
Coronary artery disease, w/o AMI, with angioplasty
13 Nephrology
14 Urology
Coronary artery disease, w/o AMI, with cardiac
264
15 Obstetrics

catheterization
16 Gynecology
265
Ischemic heart disease, except CHF, w/o AMI
17 Dermatology
267
Congestive heart failure, with comorbidity
18
Orthopedics & Rheumatology
19 Neonatology

268
Congestive heart failure, w/o comorbidity
20
Preventative & Administrative
332 Allergic
rhinitis
21
Late Effects, Environmental Trauma & Poisoning
22
Isolated Signs & Symptoms
383
Acute bronchitis, w/o comorbidity, age less than 5

management or treatment of a patient. Office visits,
384
Acute bronchitis, w/o comorbidity, age 5 & older
inpatient stays, therapies, and surgical procedures
436
Ulcer, complicated with surgery
are examples. An anchor record indicates that a
437
Ulcer, complicated w/o surgery
clinician has evaluated a patient’s illness and has
438 Ulcer,
simple
decided on the types of services required to further
720 Osteoporosis
identify and treat the patient’s condition.
Major trauma, other than fracture or dislocation,
744-01
w/surgery - foot & ankle

Ancillary records are services that are incidental to

the direct evaluation, management and treatment of
a patient – for example x-rays, pharmaceuticals and

1
lab tests. Each ancillary record links to only one
Symmetry Episode Treatment GroupsTM. User’s Guide,
Release 6.0, 2006.

Confidential and Proprietary

August, 2006
Symmetry
4

---

An Innovative Approach to Health Risk Assessment

anchor record, based on the type of provider, the
Finally, physicians often prescribe medication for an
nature of the service performed, and the diagnoses
ongoing condition without requiring the patient to
assigned.
make an office visit. Although these instances are
not technically termed episodes of care, they are
After anchor and ancillary services are assigned to
categorized into ETGs based on the likely
clusters, clusters are grouped into episodes based
indications for the drug treatment prescribed.

on a series of rules and the diagnoses2 and
Alzheimer’s disease therapy, migraine therapy and
procedures3 found on medical claims, and the drug
diabetes mellitus therapy are examples of these
treatments included on pharmacy claims4. To do
“drug-only” ETGs.
this, ETGs categorize diagnoses as primary,
incidental, complicating, or co-morbid. In general,
The ETG technology is distributed in the form of
each diagnosis is primary to only one ETG. A
“Grouper” software. This software accepts health
primary diagnosis can begin an episode or be
care claims and returns an ETG value, along with
mapped directly into an existing one. Incidental
other information. The subsequent “grouped” data
diagnoses describe conditions present during the
can then be used as input into other reporting
treatment of another disorder and are incidental to
systems and applications such as ERGs.
that disorder. Throat pain for a patient being treated
A number of features of ETGs have importance for
for an episode of bronchitis is an example of an
its use in a health risk assessment model. By
incidental diagnosis. In this instance, the services
assigning individual medical services to unique
related to the throat pain would be included in the
episodes of care and characterizing these episodes,
bronchitis episode and not begin a new episode.
ETGs define clinically-based units that can be
Complications indicate a sicker patient that may
combined to construct a profile of risk for an
require more extensive treatment for a related
individual. Further, an important construct of ETGs
condition. Co-morbidities represent ongoing chronic
is its ability to prioritize related medical conditions,
conditions that impact treatment requirements for
allowing focus on that condition best describing the
another episode. The presence of complications
mix of services required for the ongoing evaluation,
and co-morbidities can impact the final ETG
management and treatment of an episode of care.
assigned to an episode of care.5
For example, for incidental diagnoses, rather than
An episode is termed complete based on the
indicate a separate incidence of a new condition,
absence of treatment for a condition for a specified
ETGs combines these services into the episode for
period of time. This dynamic period of time is called
the underlying, primary disorder. In this way, ETGs
the clean period and varies across ETGs. For
performs a complex, hierarchical grouping of
example, the clean period for an episode of acute
conditions that provides a “filter” for characterizing
bronchitis is 30 days. This timing and the diagnostic
markers of patient risk. Finally, ETGs link both
and other categorizations described above
medical and pharmacy services to diagnostic
determine the final grouping of services into an
conditions, allowing prescription drug data to
episode and the assignment of the episode to an
contribute to the measurement of health risk.
ETG.
2. Episode Risk Groups (ERGs) – Mapping ETGs

to ERGs. The ETG grouping provides a record of
2 International Classification of Diseases, 9th Revision
(ICD-9 Codes), 2006.
the different episodes of care identified for an
3 Current Procedural Terminology (CPT Codes).
individual. A key step in developing ERGs is
American Medical Association, 2006.
deciding how these episodes can best be used as
4 National Drug Code (NDC Codes). U.S. Food and Drug
markers of risk. One option is to use all of the
Administration, 2006.
5
approximately 600 ETGs as separate risk markers.
A defining surgery during an episode can also impact
ETG assignment.
This approach was not chosen for several reasons.

Confidential and Proprietary

August, 2006
Symmetry
5

---

An Innovative Approach to Health Risk Assessment

First, such a large number of risk factors would likely
with the ETGs for migraine headache to form a
produce relatively small sample sizes for some
migraine ERG. Diabetes mellitus drug therapy
markers, resulting in implausible or imprecise
was combined with the diabetes ETGs. In this
estimates of their contribution to risk. The level of
way, for selected conditions, distinctions were
clinical detail provided by ETGs could also produce
not made between those patients prescribed
imprecision due to the potential overlap in the impact
medication for an ongoing condition without an
of medically-related episodes on patient risk – over
associated office visit and other patients with the
or underestimating risk for members with different
same condition.
combinations of these episodes. Further, for a
• Those ETGs with relatively low prevalence were
number of conditions, ETGs define different
combined with other ETGs based on clinical
groupings based on observed treatment – for
similarity and implications for risk assessment.
example, separate ETGs are defined for chronic
• To enhance clinical relevance and also
sinusitis, with surgery and chronic sinusitis, without
homogeneity in terms of risk, in each of the
surgery. Assigning risk based on therapeutic
steps described above, ETGs were only
approach may reward certain clinical decisions,
combined with other ETGs in the same Major
including the decision to perform a procedure.

Practice Category (MPC) (see Table 2).
Finally, a number of ETGs distinguish a condition
• Both complete and incomplete episodes of care
with and without co-morbidity. Congestive heart
during a time period indicate the presence of a
failure (CHF) is one example. Since the co-morbid
condition or disease and contribute to a patient’s
conditions for CHF, such as diabetes, are described
profile of risk. Episode assignment to ERGs is
by other ETGs and are to be modeled as separate
not dependent on completion status.
risk markers in the ERG model, such a distinction
may not be required.
• ERG assignment does not vary with the number
of episodes or ETGs observed for an individual
A decision was made to combine episodes into
within the same ERG. Patients with single or
larger groups to create ERGs. In mapping ETGs to
multiple episodes within an ERG receive
ERGs the primary goal was combining episodes of
identical assignments.
similar clinical and risk characteristics, while at the
same time recognizing appropriate incentives and
Using this approach, a total of 126 ERGs were
practical issues. Both clinical input and empirical
identified. The final set of risk groups for ERGs,
evidence guided this process. The mapping
Version 6.0 are described in Table 3.
involved a number of steps and assumptions:
Finally, in assigning episodes to ERGs, steps were
• In nearly all cases, ETGs differentiated only by
taken to prioritize selected related ETGs within an
the presence of a defining surgery were
MPC. This step permits additional focus on those
combined. The exceptions were judged to both
episodes best describing a patient’s underlying
involve little discretion in the decision to perform
medical condition within a disease category.
surgery and also its contribution to patient risk.
3. ERG Profile. A member’s age, gender and mix
Organ transplants are one example.
of ERGs are used to create their ERG profile. Ten
• For relevant conditions, ETGs differentiated by
age groups are used for each gender for this
the presence of a co-morbid condition were
purpose: 0-5, 6-11, 12-18, 19-34, 35-44, 45-54, 55-
mapped to the same ERG.
64, 65-74, 75-84 and greater than 84 years of age.
• ETGs describing “drug-only” patient encounters
were combined with those corresponding ETGs
for the indicated condition. For example, the
migraine therapy, drug-only ETG was combined

Confidential and Proprietary

August, 2006
Symmetry
6

---

An Innovative Approach to Health Risk Assessment

Table 3.
Episode Risk Groups
ERG Description
ERG Description




MPC 1
Infectious Disease
MPC 10
Pulmonology
01.01
AIDS/HIV
10.01
Acute bronchitis
01.02
Non-HIV major infectious disease w cb
10.02
Asthma
01.03
Non-HIV major infectious disease, wo cb
10.03
Chronic Bronchitis
01.04
Lower cost infectious disease
10.04
Emphysema
Malignant pulmonary neoplasm w active
MPC 2
Endocrinology
10.05
management
02.01
Insulin dependent diabetes, w cb
10.06
Other lower cost pulmonology, I
02.02
Insulin dependent diabetes, wo cb
10.07
Other lower cost pulmonology, II
02.03
Non-insulin dependent diabetes, w cb
10.09
Other moderate cost pulmonology
Malignant pulmonary neoplasm w/o active
02.04
Non-insulin dependent diabetes, wo cb
10.10
management
Malignant neoplasm of
02.05
MPC 11
Gastroenterology
pancreas/pituitary/adrenal, active management
02.06
Hyperlipidemia, excluding lipidoses
11.01
Ulcer
02.07
Other lower cost endocrinology I
11.02
Hernias
02.08
Other lower cost endocrinology II
11.03
Appendicitis
Malignant neoplasm, gastroenterology,
02.09
Other moderate cost endocrinology
11.04
with active management
02.10
Other higher cost endocrinology
11.05
Other lower cost gastroenterology
Malignant neoplasm, thyroid and parathyroid,
02.12
11.06
Other moderate cost gastroenterology, I
active management
02.13
Cystic Fibrosis and Lipidoses
11.07
Other moderate cost gastroenterology, II
MPC 3
Hematology
11.08
Other higher cost gastroenterology
Malignant neoplasm, gastroenterology,
03.01
Leukemia w bone marrow transplant
11.09
w/o active management
Neoplastic blood disease and Leukemia w/o
03.02
MPC 12
Hepatology
BMT, w active management
Major non-neoplastic blood disease, excluding
03.03
12.01 Cholelithiasis

SCE and hemophilia
03.05
Lower cost hematology
12.02
Infectious hepatitis
Neoplastic blood disease and Leukemia w/o
03.10
12.03
Liver transplant
BMT, w/o active management
03.11 Sickle-cell
Anemia
12.04 Other lower cost hepatology
03.12 Hemophilia
12.05 Other
moderate cost hepatology
MPC 4
Psychiatry
12.06
Other higher cost hepatology
Malignant neoplasm of the hepatobiliary
04.01
Mood disorder, depressed w/o psychosis
12.10
system with active management
04.02
Personality and eating disorders
MPC 13
Nephrology
04.03
Dementia and mental retardation
13.01
Acute renal failure
04.04
Child psychiatric disorders
13.02
Chronic renal failure
04.05
Psychotic and schizophrenic disorders
13.03
Kidney Transplant
04.06
Lower cost psychiatry
13.04
Lower cost nephrology
04.07
Other moderate cost psychiatry
13.05
Moderate cost nephrology
04.08
Mood disorder with Bipolar or psychosis
MPC 14
Urology
MPC 5
Chemical Dependency
14.01
Lower cost urology, I
05.01
Moderate and higher cost substance abuse
14.02
Lower cost urology, II
05.02
Other drug dependence
14.03
Moderate cost urology
Malignant neoplasm, urology, with active
MPC 6
Neurology
14.05
management
06.01
Migraine headache
MPC 15
Obstetrics
06.02
Major brain and spinal trauma
15.01
Normal pregnancy, delivery
Malignant Neoplasm of central nervous system,
06.03
15.02
Complicated pregnancy, delivery
with active management
Non-cranial nerve inflammation, incl carpal
06.04
15.03
Normal pregnancy, non-delivery
tunnel

Confidential and Proprietary

August, 2006
Symmetry
7

---

An Innovative Approach to Health Risk Assessment

Table 3.
Episode Risk Groups
06.05
Other lower cost neurology
15.04
Complicated pregnancy, non-delivery
06.06
Other moderate cost neurology
15.05
Abortion
06.07
Other higher cost neurology
MPC 16
Gynecology
Malignant neoplasm, breast and female
06.08
Multiple sclerosis and ALS
16.03
genital tract, with active management
MPC 7
Ophthalmology
16.04
Other lower cost gynecology, I
07.01
Glaucoma
16.05
Other lower cost gynecology, II
07.02
Other lower cost ophthalmology
16.06
Other moderate cost gynecology, I
07.03
Moderate cost ophthalmology
16.07
Other moderate cost gynecology, II
07.04 Retinopathy
16.08 Other
moderate cost gynecology, III
Malignant neoplasm, gynecology, wo
07.05 Malignant
neoplasm of the eye
16.10
active management
MPC 8
Cardiology
MPC 17
Dermatology
08.01
Hypertension, benign and malignant
17.01
Lower cost dermatology, I
08.03
Congestive heart failure
17.02
Lower cost dermatology, II
Malignant skin neoplasm, major with
08.04
Coronary heart disease, w AMI
17.03
active management
08.05
Coronary heart disease, incl ischemia, wo AMI
17.04
Higher cost dermatology
08.06 Atherosclerosis

MPC 18
Orthopedics & Rheumatology
08.08
Heart and/or Lung Transplant 18.01
Arthritis,
Rheumatoid
08.09
Low Cost Cardiovascular
18.02
Lower cost orthopedics, I
08.10
Other lower cost cardiology, II
18.03
Lower cost orthopedics, II
08.11
Other moderate cost cardiology, I
18.04
Other moderate cost orthopedics, I
08.12
Other moderate cost cardiology, II
18.05
Other moderate cost orthopedics, II
MPC 9
Otolaryngology
18.07
Higher cost orthopedics
Malignant neoplasm of the bone and
09.01 Rhinitis/sinusitis

18.10 connective tissue, active management
Malignant neoplasm ENT, with active
09.02
MPC 19
Neonatology
management
09.03
Other lower cost ENT, I
19.01
Higher cost neonatal
09.04
Other lower cost ENT, II
19.02
Other neonatal
09.05
Moderate cost ENT
MPC 21
Late Effects, Environmental Trauma
Malignant neoplasm ENT, w/o active
09.06
21.01
Late effects and complications
management


21.02
Environmental trauma


21.03
Poisonings and toxic effects of drugs


MPC 22
Isolated Signs and Symptoms


22.01
Isolated signs and symptoms
MPC


Pharmacy Only
RX


RX.01
High cost pharmacy only


Every member is assigned to an age-sex group.
particular, while the retrospective model applies
Members can also be assigned to zero, one, or
some weight for all ERGs assigned, the prospective
more ERGs depending on their mix of episodes of
model applies no weight to those ERGs observed to
care. Members without claims will have no episodes
have a negligible impact on future risk. Appendicitis
of care and no ERGs. For these members, risk is
is one example.
based solely on age and gender.
4. Measuring the Contribution of ERGs to
The rules employed in mapping ETGs to ERGs are
Member Risk – ERG Risk Weights. The next step
identical for the retrospective and prospective
is the assignment of a weight to each ERG and
applications of the model. However, the final set of
demographic marker of risk. These weights
ERGs included in the two models differs. In
describe the contribution to risk of being in a specific
age-sex group or having a particular medical

Confidential and Proprietary

August, 2006
Symmetry
8

---

An Innovative Approach to Health Risk Assessment

condition included in an ERG. The model of risk can
risk weights for each model are derived separately
be defined generally as:
the a’s and b’s and the c’s in (1) and (2) above.9
(1)
RiskPi = Σas*asexi,s + Σbe*ERGi,e
ERG Models. ERGs provide significant flexibility for
supporting different business applications and data
(2)
RiskRi = Σce*ERGi,e 6
availability. In particular, ERGs include different
where RiskPi and RiskRi are the ERG prospective
model options that support the following features:
and retrospective risk scores for person i; asexis and
• Retrospective and prospective applications,
ERGie indicate their age-sex group (s); and ERG (e)
as described above.
assignments, and the a’s, b’s and c’s are the risk
weights. The age-sex and ERG markers are a
• Pharmacy benefit status, recognizing and
series of 0,1 variables, set to 1 if the marker is
adjusting for the lack of pharmacy
observed for an individual, 0 otherwise. Each
information for an individual.
member has their own profile of age-sex and ERGs.
• Length of enrollment. A member’s length of
However, for each ERG model, the risk weights are
enrollment may impact their risk. In
pre-defined and will be the same for all individuals. 7
consequence, ERGs provide partial
A person’s risk score is the sum of these risk
enrollment models which utilize different
weights for each marker observed.8
risk weights corresponding to the member’s
An ERG may have a different impact on risk
length of enrollment during the experience
depending on whether current or future risk is being
period used to measure risk.
measured. For example, acute conditions generally
• A/U timing. Actuarial and underwriting
have a greater impact on retrospective than
practices require time between the
prospective risk. Chronic conditions have similar
experience period used in measuring risk
impacts on current and future risk. As a result, the
and the future time period being predicted.
risk weights assigned to each marker vary
This allows the additional time necessary for
depending on whether the application is
claims lag and analysis prior to development
retrospective or prospective. To accommodate
of group premiums. To accommodate this
this, separate ERG models are defined for
need, ERGs employ a prospective A/U
retrospective and prospective applications and the
model that reflects a 6-month interval
between the experience period and the
prediction period.
• Dollar Thresholds. ERG models support
assumptions of both a $25,000 and
$100,000 prediction threshold. For these

6 Note that the asex group parameters were found to have
model features, the risk score provided by
minimal impact on retrospective risk and are excluded
ERGs reflects experience up to the
from all retrospective models.
threshold amount and provides flexibility in

7 The risk weights are pre-set and delivered as part of the
predicting higher level risk or risk assuming
ERG software. Alternatively, ERG customers with large

patient populations (greater than 500,000 members) might
9 As described in the following section, in addition to
want to estimate weights using their own experience.
retrospective and prospective model variants, ERGs

include a number of different models that vary depending
8 For some ERGs, differences in risk weight were
on factors such as the time period used for prediction, an
observed for elderly versus non-elderly individuals. In
individual’s pharmacy benefits status, length of enrollment
these cases, the weights were also allowed to vary
and prediction dollar threshold. Each of these models
depending on the age of the individual – less than 64
uses the same ERG array but a different set of pre-defined
years of age or age 65 and over.
weights.

Confidential and Proprietary

August, 2006
Symmetry
9

---

An Innovative Approach to Health Risk Assessment

a lower truncation. This model option
and so on. The 58-year-old male described in Table
supports applications where reinsurance or
6, Example 1 has a prospective risk score of 6.7427
other mechanisms are in place to address
– indicating a high level of health risk more than six
differences in risk and experience beyond
times that of the average for the large managed care
these amounts.
population used in developing ERGs. Note that the
risk score for the member differs between the
Estimating ERG Risk Weights. The risk weights
prospective, retrospective and A/U models reflecting
for ERGs were estimated using multiple linear
the impact of different timing for the predicted
regression and enrollment and medical and
outcome.
pharmacy claims data for a managed care
population of over 8 million members. These data
The risk score being produced, the type of claims
were also used to test the predictive accuracy of the
input into the processing engine, the risk outcome
ERG model, as described below. The ERG
desired, and the expenditure threshold are all factors
development data were obtained from the IHCIS
in determining which model is utilized when
National Managed Care Benchmarks Database, a
calculating a given risk score. Table 5 summarizes
large national database comprised of claims and
the various models available within the 6.0 release
membership information aggregated from over 35
of ERGs. The cells in the table represent the
sources.
expenditure thresholds available with each of the
various models.
Table 6 provides examples of how retrospective,
prospective and A/U ERG risk would be calculated.

The examples use medical and pharmacy data for
Table 5. ERG Models
input and output, a threshold assumption of
$100,000, and a minimum of 10 months of
Input/Output
Retrospective
Prospective
A/U
continuous enrollment.
MedRX/MedRX
25K, 100K
25K, 100K
100K
As shown, in example 1, over a 12-month period, a
Med/MedRX
25K, 100K
25K, 100K
100K
male, age 58 was observed to have five unique
episodes of care, covering four different ETGs –
Med/Med
25K, 100K
25K, 100K
100K
diabetes, CHF, an ulcer, and two episodes for a
minor skin problem. These ETGs map to four
different ERGs. The individual’s age and sex and
these four ERGs describe their profile of risk. The
sum of the weights assigned to these risk markers
provide the overall risk scores for the individual –
separate risk scores for the retrospective,
prospective and A/U models.10
The scores in Table 6 reflect each individual’s
measure of risk relative to that of the overall
population used in developing ERGs. A score of
1.00 indicates risk comparable to that of the
development population, a score of 1.10 indicates
10 percent greater risk, 0.85, 15 percent lower risk,

10 Note that the second minor skin episode in Example 1
does not contribute any additional weight in calculating
overall risk.

Confidential and Proprietary

August, 2006
Symmetry
10

---