Transcript Burt

Deriving practice-level
estimates from physician-level
surveys
Catharine W. Burt , EdD and Esther Hing, MPH.
Chief, Ambulatory Care Statistics Branch
Session 32
June 20, 2007
ICES III, Montreal, Canada
U.S. DEPARTMENT OF HEALTH AND HUMAN SERVICES
Centers for Disease Control and Prevention
National Center for Health Statistics
Topics







Introduction
Multiplicity theory
Re-weighting methods
Application to NAMCS
Assumptions
Analytical example
Limitations
Multiplicity theory

Multiplicity occurs when the same observation
unit can be counted multiple times among the
selection units


eg., same patient is counted in multiple records of
visits/discharges or same medical practice is counted
in records of multiple physicians
Using principles of network sampling, you can
adjust weights to estimate the observations of
interest rather than the selection units
Desired observation units
Survey selection units
Greek for the Geeks
X   i
N
i
X j (i)  X j
X j (i)  0
1
i

Mi
j
X j (i)
Mj
= the selection probability of physician i (i =
1, …, N)
and
if physician i is affiliated with practice j, and
if physician i is not affiliated with practice j.
Weight adjustment to estimate
X
Observation weight
= selection weight / M
where M is the multiplicity information for
the selection unit
Re-weighting methodology
 Assumptions


and definitions
Use multiplicity information from the physician
data to adjust physician-level estimates into
practice-level estimates
Dividing the physician sampling weight by
number of physicians in the practice provides
a measure of practices
Physicians ► practices example…
Samples of physician records in medical
practices
 Physician data have the same practice
included in multiple observations.
 If we knew how many physicians were in
the same practice as the sampled
physicians, then we can adjust the
estimator to account for the multiplicity.

Application to NAMCS





National Ambulatory Medical Care Survey
Annual survey of 3,000 nationally representative
office-based physicians in patient care
Excludes radiologists, anesthesiologists, and
pathologists and federally-employed physicians
Face-to-face induction interview asks physicians
questions about his/her office practice
Records are weighted by the inverse of the
probability of selection, adjusted for nonresponse
(~60% RR), with a calibration ratio to annual totals
Induction interview content
 Number



of locations
Number of other physicians
Ownership
Type of office
• Private, clinic, HMO, faculty practice plan, etc
 EMR
adoption
 Revenue sources
Assumptions



Used the first location reported
Assumes practice information provided by
sample physician is a constant for the practice
Does not account for multiplicity of practices
within a physician
• i.e., Ignores the fact that some physicians are
affiliated with multiple practices (about 1% of
physicians)
3 medical practices with a total of
7 physicians
Solo practice
Partner practice
Group practice
Probability of selecting a practice
1/7
2/7
4/7
Multiplicity factor
1/7
2/7
4/7
1
.5
.25
Multiplicity information
 How
many other physicians practice with
you at this location?
 M=
1+ # of other physicians
 Practice
weight = physician weight / M
Re-weighting example
Practice
size
Physician
weight
solo
10
Multiplicity Practice
adjustment weight
1
10
partner
20
.5
10
3
40
.333
13.3
4
40
.25
10
Sum = 110 physicians
43 practices
 Practice
weight =
physician weight / practice size
 physician
 practice
weight → 311,200 physicians
± 8,000
weight → 161,200 practices
± 5,300
Percent distribution of office-based medical
physicians and practices by size
70
Percent
69
60
50
40
36
30
27
20
15
12
14
11
10
11
4
1
0
Solo Partner '3-5
'6-10
Physicians
11+
Solo Partner '3-5
Practices
'6-10
11+
Computerized administrative and clinical
support systems
80
70
69.2
74.2
Practices
Physicians
60
50
40
30
20
15.0
19.0
6.5
10
9.2
0
Uses electronic billing
Uses EMR
Uses CPOE
Limitations of NAMCS data…

Good


National estimates of
practices
Characteristics that
are common among
physicians

Bad



Characterizing
practices
Underestimates larger
practices
Be careful how you
define size
• First-listed location
• Location with most
visits
• Location of the visit