SEM was used to model patient safety management using causal models derived by experts and validated using monitoring survey data. Once defined, SEM was used to regress the defined strategies against each other to produce final model outputs. As a result, a series of causa relations arrows are produced, each associated with a standard coefficient representing its impact on a scale of -1 (high negative impact) to +1 (high positive impact) (16,17).
A questionnaire intended to identify factors affecting primary care was developed. It comprised 38 items which were adopted from multiple scales used elsewhere in the literature. The instrument comprises the following five constructs which were analyzed by SEM: patient safety management, facilities in the practice, generic conditions, communication and collaboration and education.
The questionnaire was completed by 248 individuals (Table 1). The respondents differed significantly in the number of patients cared for and the size of their counseling centers, measured by the number of patients. The number of patients ranged from 0 to 4,000 (mean 1.763 patients; SD = 873). The number of patients per facility ranged from 300 to 16,220 (mean 5091; SD=3,096). In addition, 49.6% of respondents conducted their medical practice in a city with over 100,000 people.
Table 1. Sample profile (N = 248)
Characteristics
|
n
|
%
|
Gender
|
|
|
|
Male
|
132
|
53.2
|
|
Female
|
116
|
46.8
|
Age [value from 29 to 63], mean and SD
|
43
|
7
|
Current professional discipline
|
|
|
|
GP
|
141
|
56.9
|
|
General internist
|
102
|
41.1
|
|
Other primary care physician
|
51
|
20.6
|
|
Medical teacher
|
10
|
4
|
|
Scientific researcher
|
10
|
4
|
|
Other
|
6
|
2.4
|
Current professional discipline *
|
|
|
|
Individual
|
61
|
24.6
|
|
Group
|
53
|
21.4
|
|
counseling centers
|
164
|
66.1
|
Number of doctor's patients [value from 0 to 4,000], mean (SD)
|
1,763
|
872.96
|
Number of patients in the facility [value from 300 to 16,220] , mean and SD
|
5,091
|
3,095.5
|
Area of practice
|
|
|
|
city with over 100,000 inhabitants
|
123
|
49.6
|
|
city of 30,000 to 100,000 inhabitants
|
37
|
14.9
|
|
city with less than 30,000 inhabitants
|
44
|
17.7
|
|
small town / village
|
44
|
17.7
|
Most of the 38 presented items (Table 1A in appendix) were seen as important by the majority of participants, but the use of the strategies in daily practice varied widely (5). The mean scores for all questions were low (see table 1A in appendix). The highest scoring strategy concerned patient medical records (mean=3.33 and SD=1.067). The following items were found to have a mean score above 2: planned checks of safety of equipment medication and other facilities in the practice, the use of a properly-kept computerized medical record system, the use of ICPC-coded electronic record standards, a strong understanding of patient safety in health professionals. Particularly how it varies depending on treatment models complications and the availability of information technology in general practice and the ability to properly use them.
Measurement of latent variables
The instrument used in the study consists of 38 items measuring the five latent variables (strategies). All measures are scored on a four-point Likert scale providing sufficient variance and covariance for better data analysis (18). In addition, all the manifest variables included in the instrument, i.e. the items, reflect the changes of their corresponding latent variables and therefore can be seen as being caused by constructs (19). In addition, all the constructs are operationalized as first-order latent variables because first, this way, the complexity of the whole model is reduced as the number of latent variables did not increase. Secondly, as no blocks of indicators were found to share specific common characteristics, all items were treated as a single latent variable. A more detailed specification of items is presented in appendix Table A1.
The unrotated principal component factor analysis (CFA), principal component analysis with varimax rotation, and principal axis analysis with varimax rotation all revealed the presence of three distinct factors with eigenvalue greater than 1.0, rather than a single factor (20). The seven factors together accounted for 67.84 percent of the total variance; the first (largest) factor did not account for a majority of the variance (11.324%). While the results of these analyses do not preclude the possibility of common method variance, they do suggest that common method variance is not of great concern and thus is unlikely to confound the interpretations of results.
After 248 observations, the accuracy of each of the hidden variables, was found to be at least 0.7, as measured by Cronbach’s alpha (each variable also demonstrated CR>0.7 and AVE>0.5). The latent variable was not constructed from all of the observable variables proposed by (1). The strongness of these deliberations is to meet the threshold conditions by the constructed variables and the observable variables that represent it.
Table 2. Discriminant validity for constructs and their correlations
|
Mean
|
Std. Deviation
|
R2
|
FP
|
CC
|
PSM
|
GC
|
EPS
|
FP
|
1.493
|
.591
|
.356
|
.711
|
|
|
|
|
CC
|
1.221
|
.522
|
.412
|
.488
|
.927
|
|
|
|
PSM
|
1.529
|
.653
|
.365
|
.543
|
.688
|
.710
|
|
|
GC
|
1.245
|
.466
|
.356
|
.503
|
.590
|
.555
|
.709
|
|
EPS
|
1.077
|
.365
|
.593
|
.408
|
.486
|
.615
|
.627
|
.710
|
Note: CC: communication and collaboration, CR: composite reliability, EPS: education on patient safety, FP: facilities in the practice, GC: generic conditions, the square root value of AVE is shown on the diagonal, under the diagonal of the Pearson correlation coefficient. For all p <0.001.
The square root of the AVEs are compared with the appropriate correlation factors in Table 2. They have much higher values, indicating positive divergent validity, i.e. the individual latent variables differ significantly from one another. In addition discriminant validity analysis was performed to determine whether the measures of each construct differ sufficiently from those of other constructs (17).
The part of the model that examines relationship between the latent variables and their measures is known as the measurement model. A previous CFA based on a sample of respondents in Poland confirmed that the measurement model demonstrates satisfactory construct validity, discriminant validity and internal consistency (21,22). In the case of the measurement models, there is no reason to reject the hypothesis that the standardized residual values of the empirical and theoretical matrix are equal to zero (χ2 = 564.812; p = 0.000). The model was found to demonstrate a good fit to the data, as indicated by a root mean square of approximation error (RMSEA) of 0.061<0.08 (LO=0.054; HI=0.069), and to demonstrate good acceptability, as indicated by χ2/ss=1.928<2, GFI=0.951>0.9 and AGFI=0.921>0.9 (16,21,23). All latent variables in the model are significantly correlated, as shown in Table 3.
Table 3. Covariance and correlation between latent variables
Relation
|
Covariances
|
S.E.
|
C.R.
|
Correlations
|
EPS
|
<-->
|
CC
|
.127
|
.029
|
4.428
|
.486***
|
EPS
|
<-->
|
PSM
|
.119
|
.023
|
5.302
|
.615***
|
EPS
|
<-->
|
FP
|
.138
|
.031
|
4.491
|
.408***
|
CC
|
<-->
|
PSM
|
.125
|
.027
|
4.595
|
.688***
|
CC
|
<-->
|
FP
|
.154
|
.036
|
4.234
|
.488***
|
PSM
|
<-->
|
FP
|
.127
|
.027
|
4.790
|
.543***
|
EPS
|
<-->
|
GC
|
.189
|
.040
|
4.790
|
.627***
|
CC
|
<-->
|
GC
|
.167
|
.041
|
4.019
|
.590***
|
PSM
|
<-->
|
GC
|
.116
|
.028
|
4.161
|
.555***
|
FP
|
<-->
|
GC
|
.183
|
.045
|
4.109
|
.503***
|
Note: *** mean p<0.0001, CC: communication and collaboration, CR: composite reliability, EPS: education on patient safety, FP: facilities in the practice, GC: generic conditions