METHODS
Study Sample
The target population was U.S. hospital laboratories. The
sampling frame was hospitals from the 1999 directory of the American
Hospital Association (AHA). This database is not limited to the AHA members
and includes 95% of all U.S. hospitals (personal communication, AHA,
November 1999 and as verified against the Online Survey, Certification and
Reporting database of CLIA-registered hospitals). This database includes the
number of beds, permitting us to stratify hospitals by this variable as a
surrogate measure of hospital size. We stratified hospitals into those with
> 200 beds (large hospitals) and those with <200
beds (small hospitals). The randomly selected large and small hospitals in
the study sample constituted 26% and 9% of the large and small hospitals
listed in the sampling frame, respectively.
Questionnaire Development
We developed this survey questionnaire based on the
recommendations of an advisory panel and numerous discussions and
deliberations within the CDC. We began the process to develop the survey by
formulating an initial set of questions that reflected these deliberations.
The initial version of the questionnaire then underwent numerous revisions
to improve clarity, brevity and formatting. We also asked a survey expert at
the CDCs National Center for Health Statistics to review the clarity,
framing and formatting of the survey questions. After revising the survey
based on the comments we received, we pretested the questionnaire by sending
it to laboratory management personnel of 9 hospital coagulation
laboratories. (See the Appendix for a copy of the actual survey used.)
Following the recommendation of the U.S. Office of
Management and Budget (OMB), we developed an Internet based questionnaire
and offered it to all the respondents as an alternative means of responding
to the survey. This electronic version mirrored the printed questions,
answer selections and formatting. Additionally, it incorporated logical
constraints to prevent the respondents from entering conflicting or
contradictory answers.
Sample Size and Number of Beds
The rationale for sample size was based on the criterion of
ensuring a maximum confidence interval of 6% for any estimated proportion
within the selection strata. We used a random sample stratified by large
hospitals (with > 200 beds) and small hospitals
(with <200 beds) to ensure sufficient numbers of the respondents for the 2
strata so that we could analyze data separately for each. The purpose of
examining large and small hospitals as separate groups was to determine
whether they had different practice profiles regarding coagulation testing:
Type of facility |
Population size (1999 AHA directory) |
Size of selected sample (percent) |
Small Hospitals (< 200 beds) |
4,245 |
375 (9%) |
Large Hospitals (> 200 beds) |
1,662 |
425 (26%) |
Total |
5,907 |
800 (14%) |
Below is the distribution of the number of beds for the
final 632 respondents that composed 79% of the study sample.
Distribution of Number of Beds |
Number of beds |
Number of hospitals |
Proportion of total |
<50 |
101 |
16% |
5099 |
98 |
16% |
100149 |
71 |
11% |
150199 |
41 |
6% |
200249 |
86 |
14% |
250299 |
64 |
10% |
300349 |
45 |
7% |
350399 |
34 |
5% |
400499 |
27 |
4% |
500599 |
25 |
4% |
600799 |
23 |
4% |
800999 |
11 |
2% |
10001399 |
6 |
1% |
01399 |
632 |
100% |
Survey Implementation
Upon receiving the OMB approval for the survey, the CDC
directed ASI to take the following steps:
- identify and contact each laboratory director while verifying the
hospitals mailing address,
- print the survey questionnaires and post it on a secure CDC website,
- mail the survey packets to the entire sample, providing each
laboratory the option of responding electronically,
- mail reminder postcards to each (potential) participant 1 week after
mailing the printed survey, and
- contact the non-respondents by telephone and, if needed, send
additional survey packets.
Before selecting the study sample, we excluded laboratory
personnel who were involved in pretesting the questionnaire. We also removed
the institutions of members of the expert panel from the sampling frame.
We sent each questionnaire along with a cover letter
addressed to the laboratory director by name. This letter included the
identification code and the password for each hospital to enable them to
respond electronically if they elected to do so. We administered this survey
and collected results between June 2001 and October 2001. One week after the
initial mailing of the questionnaire, ASI sent a reminder postcard to each
respondent, and telephoned the non-respondents within 2 weeks after mailing
the postcard. The purpose of the call was to confirm that the respondent
received the survey, to encourage the hospital laboratory to participate,
and to secure a commitment from the laboratory director to complete and
return the survey. We followed up with non-respondents between July 2001 and
September 2001. We were not able to contact 7 non-respondents.
Data Entry and Management
Two different individuals keyed the data from each paper
survey. Data management personnel had the 2 separate data entries compared,
identified any discrepancies, and documented these. A data entry supervisor
then performed the adjudication process. The second quality assurance (QA)
system used to assess the integrity of the data entry process consisted of
examining 10% of the survey booklets after the rekey and adjudication of
discrepancies. The auditor compared the survey booklet answers to the
database to determine if the answers were correct.
Data Exclusion
We combined the databases containing responses from the 2
modes of data collection (paper and electronic) to create the final analysis
database. We found 10 instances when the participant had responded both
affirmatively and negatively to a question. We excluded these responses.
There were also 9 other instances when the respondent had checked more than
1 selection, despite the instructions specifying that the respondent could
check only 1 selection. We excluded all such responses. We also excluded all
responses to a gate question and those relating to it if the survey
instruction had specified that those subsequent questions not be answered.
Due to such a problematic response pattern, we excluded all responses to 1
of the surveys 43 questions (Question 31) and the 3 subsequent
sub-questions involving a total of 1,300 responses. We also excluded a total
of 568 other responses because, after responding negatively to a gate
question, participants provided answers to 1 or more subsequent questions.
After excluding the question and sub-questions noted above,
there remained 153 questions and sub-questions involving 96,696 possible
responses. The following illustrates how participants responded:
Response Pattern |
Type of response |
Number of responses |
Proportion of total responses |
Valid |
62,321 |
64% |
Blank |
33,788 |
35% |
Invalidresponse pattern |
568 |
0.6% |
Invalidly responding both yes and no |
10 |
0.01% |
Invalid response to more than 1 selection |
9 |
0.01% |
All |
96,696 |
100% |
Data Presentation and Statistics
Reported frequencies relate to affirmative responses we
observed for yes-or-no questions and all responses checked in
multiple-response questions. These frequencies did not include the excluded
data noted above. Percent response in yes-or-no questions is percent of
affirmative responses. Percent response in multiple-choice questions is
percent of all those responding to 1 or more selection(s). For the
check-all-that-apply type questions, these percentages may total >
100% if 1
or more respondents checked more than 1 selection. For the choose-only-one
type questions, these percentages total 100% because we excluded all
multiple responses from further analysis. We compared responses from large
and small hospital laboratories using 2-tailed
χ2
test. When significant (P < 0.050)
differences existed, we provided response frequencies and percentages for
large and small hospitals separately. We compared the number of professional
certifications of the large and small hospital respondents by using
2-tailed t-test.
|