Episodes of Illness
Farrokh Alemi, [email protected]
Objectives
This presentation trains you in using our procedures for measuring episodes of illness
Based on United States patent application 10/054,706 filed on 1/24/2002 by George Mason University. We grant permission to individual scientists within university, Federal and State governments settings to use these procedures free of licensing fees. Permission is also granted to all students using this procedure as part of an educational class.
Existing Approaches
Prospective Risk Adjustment Ambulatory Visit Groups Disease Staging Products of Ambulatory Care Ambulatory Diagnosis Groups Ambulatory Care Groups.
New Approach Easy to implement
Built using Standard Query Language operations on existing data within your organization
Tailored to the special populations served by your organization
Dynamically changing Changing as the nature of diseases change
Advantage: Built on Existing Data
Simple database manipulations can produce the desired episodes of illness from Existing Organization’s Data Can be used within electronic health
records Works on any administrative database,
which has information on date of visit and diagnoses
A Mathematical Theory
Not a black box, shows in detail how episodes are measured
Makes it possible for researchers to build on each other’s work
No Clusters Existing approaches
Schneeweiss and colleagues classified all diagnoses into 92 clusters. Otitis media infection not same as wound
infection Not limited to the etiology of the disease
All operations are defined on individual diagnoses without need for broad clusters
Not a Measure of Treatment Intensity
Not intended to classify patients into homogenous resource use groups
All short visits do not belong to same episode
Intensity-based measures can measure if length of visit is appropriate but not if number of visits are appropriate.
Terminology Episode of care
Does not depend on the nature of services Does not assume that temporally contiguous
Anchor diagnosis Trigger diagnosis Stopping point Rate of progression Peak severity Outcomes
Theory
Pia= function {Tia, Sia}
Pro
bab
ility of d
iagn
osis
i and
a bein
g p
art of
same ep
isod
e
Theory
Pia= function {Tia, Sia}
Tim
e betw
een
diag
no
sis i and
a
Similarity of
diagnosis i and a
Theory
Pia=Sia/(1+βTia) Probability of
diagnosis i and a being in same
episode
Pia= function {Tia, Sia}
Theory
Pia=Sia/(1+βTia) S
imil
arit
y o
f D
iag
no
sis
i an
d a
Pia= function {Tia, Sia}
Theory
Pia=Sia/(1+βTia)
A c
on
stan
t
Time
betwee
n
diagnosi
s i a
nd a
Pia= function {Tia, Sia}
Theory
Pia=Sia/(1+βTia)
Pia= function {Tia, Sia}
Theory
When a patient presents with several diagnoses … Probability that any two of the
diagnoses may belong to an episode is calculated
Pair-wise probabilities are used to classify diagnosis into groups
Severity of an Episode
Overall severity of episode=1-пi (1-Sevi)
Sev
erit
y o
f d
iag
no
sis
i
Why Multiply Severity Scores?
Overall severity of episode=1-пi (1-Sevi)
Sym
bo
l fo
r m
ult
iplic
atio
n
Evaluation of the Theory
565 Developmentally delayed children who were enrolled in the Medicaid program of one Southeastern State Randomly sampled Included both in-patient and outpatient Medicaid
payments for the patient State paid $9,296 per patient per year. The standard error of the cost was $2,238
Constructing Episode Measures
Time between two diagnoses Severity of each diagnosis Similarity of the two diagnoses
The number of times the two diagnoses co-occur within a specific time frame
Mean number of episodes was 147 (standard error = 320).
Results of Test of Theory
Coefficients P-value
Intercept -7297 0.003
Average severity of episodes -33.58 0.000
Number of episodes 444971 0
Interaction between number of episodes & severity of
episodes
756 0
Regression of "Amount paid by the State" on severity and number of episodes
Number of observations = 565, Adjusted R Squared = 53.11%
Conclusions of Pilot Test
Episodes of care can be constructed Explained a large percentage of variance in
cost of care 53% versus typical 10%-20%
Take Home Lesson
Simple database queries can create a measure of episodes of illness that could explain a large portion of variation in outcomes