0990-Surveyof Physiciantime use PatternsMedicare fee_Supporting Statement B_SSS_6-11-13

B.

Collection of Information Employing Statistical Methods

1. Respondent Universe and Sampling Methods
The respondent universe for the survey consists of physicians in each of the following
five specialties: Family Medicine, Diagnostic Radiology, Orthopedic Surgery, Cardiology,
and Ophthalmology. The number of physicians in the U.S. as a whole and for each of the
study specialties is shown in Table B1 here, along with the targeted number of
completed surveys.
Table B1. Number of Entities in Universe Covered by Collection and in Corresponding
Samples, by Proposed Strata
U.S. Total *

Targeted number of completes
from--

AMA
Masterfile
300
60
60

Integrated
delivery system
300
60
60

Total
anticipated
sample
(completes)

All
950,000
600
Family Practice
85,931
120
Diagnostic
120
radiology
28,385
Orthopedic
surgery
21,475
60
60
120
Cardiology
21,445
60
60
120
Ophthalmology
17,963
60
60
120
*Source: AMA Masterfile, approximate count of practicing physicians (office- and
hospital-based), in specialty with contact information available.
Based on past experience conducting physician surveys, we are targeting a response
rate between 55 and 60 percent. The sample sizes were derived based on available
resources for conducting the survey and power calculations shown in Table B2. The
analytic measure of interest is the mean ratio of the respondent-reported clinical time
to the time in the Medicare Fee Schedule, calculated at the service level. The analysis
will compare the ratios across services, both within and across specialties, and across
the two components of the sample.
In order to conduct the power calculations, we relied on data provided to the project
team by the American Medical Association from past RVS Update Committee (RUC)
surveys. The data they provided included the first quartile, median, and third quartile
values for physician reports of time spent for approximately 500 different services. The
inter-quartile ranges varied greatly according to service, with the time reports for some
services tending to be far more homogeneous than for others. Some CPT-defined
service categories are far narrower in their time requirements than others and reporting
physicians clearly differed in the level of complexity they were considering when asked
about a particular service designation. On the basis of the inter-quartile ranges and the

apparent shapes of the distributions it was possible to make rough approximations of
the standard deviations of the responses for particular services. We calculated the
power of t-tests for three different “typical” sizes of standard deviation on the basis of
our approximations derived from the RUC data. The sample sizes appear sufficiently
large to detect most substantively important differences with a beta error of 20 percent
or less even for services with responses that may have relatively large standard
deviations although the level of beta error may be somewhat higher for a few highly
heterogeneous service categories.
Table B2. Power calculations for survey of physicians, comparison of means of ratios
of respondent-reported clinical time to fee schedule time, for a given service
Comparison

Anticipated
sample sizes

Detectable difference in means of
ratios at 80% powera,b
Assumed size of standard deviations of
ratios
Large—0.9 Medium—0.5
Small—
0.2

All service-level estimates from
Masterfile sample and all service-level 1800 vs. 1800
0.08
from IDS sample
Between service-level estimates for 2
specialties, within one sample (i.e.,
0.19
either Masterfile or IDS)
360 vs. 360
Between service-level estimates for
evaluation and management, within
0.33
one sample (i.e., either Masterfile or
120 vs. 120
IDS)
a
.
Calculations were made using SAS 9.2 Proc Power
b
Power for two-tailed t test of difference between means, alpha = .05,
between means =0.

0.05

0.02

0.10

0.04

0.18

0.07

null hypothesis: difference

2. Procedures for the Collection of Information
The sample will include two independent components. The first component sample will
be drawn from the Physician Masterfile, which includes current and historical data for
more than 1.4 million physicians, residents, and medical students in the United States.
The second sample will be drawn from lists of physicians affiliated with several large
multispecialty medical group practices. (The project team is currently in discussions with
a number of practices in order to gain access to their physician rosters.) With both
sampling frames, the samples will be stratified by the following five specialties: Family
Practice, Diagnostic Radiology, Orthopedic Surgery, Cardiology, and Ophthalmology
(also indicated above in Table B1). For each of the sample components, within each of
the specialty strata, we will use simple random sampling in order to target an equal
number of cases from each specialty.

3. Methods to Maximize Response Rates and Deal with Nonresponse
We plan to have a number of procedures in place to maximize response rates. The
survey will be conducted using multiple modes to minimize the burden on respondents
and make it as easy and convenient as possible to respond. Initial contacts will be sent
by mail in a FedEx or Priority mail envelope to get the respondent’s attention. A link to a
web version of the survey will be provided for those who prefer to complete the survey
online. Respondents will be given the opportunity to return the hardcopy survey in a
prepaid envelope or by fax. We will also have skilled telephone interviewers conducting
phone prompting and available to complete the survey with the respondent by phone if
requested.
We will use multiple mailings of professionally formatted materials, and will alternate
mailings of the survey with reminder postcards to encourage participation. We plan to
include a prepaid incentive of $100 will be included in the initial mailing as an indication
of respect for the respondent’s valuable time and effort. The physician survey literature
clearly demonstrates that monetary incentives, and prepaid ones in particular, increase
response rates and that higher response rates are more effective than lower ones. For
the sample derived from medical group practices, we will also include a letter of
endorsement from the organization in the survey packet mailings. Given our past
experience with physician surveys, we anticipate a response rate of between 55 and 60
percent.
In terms of non-response, we note that nonresponse imparts bias in survey estimates
only to the extent that non-responders differ from responders with respect to the
analytic variables of interest. As such, any non-response adjustment is effective only to
the extent that responders with specific characteristics respond like the non-responders
would have responded. In other words, the nonresponse adjustment assumes that the
available variables used in the adjustment are correlated with non-response bias. We
also note that there is an empirical literature on early and late responders to physician
surveys which indicates that it is reasonable to assume that there will not be significant
non-response bias. Using variables available from the frames, we will compare
responders to non-responders in order to assess how the two groups may differ. There
will be a limited set of variables available for this purpose—age and specialty of
physician, geographic practice location (e.g., U.S. Census region, metro area vs nonmetro).

4. Tests of Procedures or Methods to be Undertaken
Where available, we have drawn questions for the survey instrument from existing
surveys. We have also conducted a limited pre-test of the survey instrument to gather
additional information on clarity of wording, completeness of response categories, and
ease of understanding and responding. The pre-test was limited to nine respondents

from the selected specialties. The physician respondents were recruited through the
professional networks of the project team. These individuals were contacted by email by
one of SSS’s senior interviewing staff to set up a time for a phone interview. They were
then sent a copy of the draft questionnaire. Prior to the phone interview, they were
asked to complete the questionnaire. During the phone interview, using a prepared
protocol, an SSS interviewer went through the survey instrument section by section,
asking if there was any confusing wording, unclear questions, missing response
categories, questions that were difficult to answer, and so on. Responses were compiled
and shared with the project team and some questions in the instrument were revised.
The revised survey instruments—one per specialty—are included as part of this
package. (See Appendix C for survey instruments. See Appendix D for respondent
communications.)
5. Individuals Consulted on Statistical Aspects and Individuals Collecting and/or
Analyzing Data
Data collection design
Marc Berk, PhD
Claudia Schur, PhD
Jacob Feldman, PhD

301-628-0410
301-629-0414
301-628-0416

[email protected]
[email protected]
[email protected]

Data collection
Social & Scientific Systems, Inc., under direction of Marc Berk and Claudia Schur
Data analysis
Claudia Schur
Katie Merrell

301-538-2042

[email protected]

Agency personnel responsible for receiving and approving contract deliverables:
Donald Cox
202-690-6597
[email protected]