3. Results

#### 3.1. Demographic data

A total of 7299 respondents provided feedback (response rate of 32.2%), 38% were nurses, 27% physicians, and 19% secretary/clerk (Table 2).

3.2. Data screening and pre-analysis

Table 3. Composite average positive responses.

Total 7215 Missing values 84 Total 7299

Table 2. Demographic characteristics.

3.3. Reliability analysis

From an initial data set of 7299 respondents, it was removed 587 surveys with missing values and 2408 surveys with more than 2 answers on the option "not applicable," getting a final data set with 4304 surveys, exceeding the minimum necessary. The surveys with one answer in the option "not applicable" were replaced by the middle category in a five-point Likert scale.

Respondents

Patient Safety Culture in Portuguese Primary Care: Validation of the Portuguese Version of the Medical Office…

Physicians 1954 27 Nurses 2729 38 Assistant 456 6 Secretary 1380 19 Technicians 560 7 Others 136 2

Composite Average positive responses (%)

1. Teamwork 76 2. Patient Care Tracking/Follow Up 76 3. Organization Learning 71 4. Overall Perceptions of Patient Safety and Quality 69 5. Staff Training 44 6. Owner/Managing Partner/Leadership Support for Patient Safety 31 7. Communication about Error 54 8. Communication Openness 52 9. Office Processes and Standardization 53 10. Work Pressure and Pace 21

N %

http://dx.doi.org/10.5772/intechopen.80035

19

Reliability analysis using Cronbach's α was performed on the 10 composites to ensure that individuals were responding consistently to items (Table 4). Considering Cronbach's α, all composites had values higher than 0.6, where composite 1 achieved the highest value and

Average composite positive responses were obtained (Table 3). The lowest positive scores were found in composites Work Pressure and Pace, Owner/Managing Partner/Leadership Support for Patient Safety, and Staff Training. The composites with highest scores were Teamwork, Patient Care Tracking/Follow Up, and Organization Learning.

Patient Safety Culture in Portuguese Primary Care: Validation of the Portuguese Version of the Medical Office… http://dx.doi.org/10.5772/intechopen.80035 19


Table 2. Demographic characteristics.

matrix is used. Pearson correlations assume that data have been measured on, at least, an equal interval scale, and a linear relationship exists between the variables. These assumptions are typically violated in the case of variables measured using ordinal rating scales. Pearson correlations have been found to underestimate the strength of relationships between ordinal items.

EFA is useful for assessing the dimensionality of survey scales that measure underlying latent variables. This factor analysis gives an indication of the number of factors that the survey appears to measure of its intended subject. In this way, through EFA, we can investigate if the

Since the data are ordinal, it was used a polychoric correlation matrix for EFA analysis and a Varimax rotation. To decide on the number of factors, it was used a parallel analysis [15, 16]. Items with a factor loading lower than 0.4 on all factors were excluded. Libraries psych and

We used confirmatory factor analysis (CFA) for ordinal data to compare the Portuguese sample factor structure to the factor structure reported for the original HSOPSC. CFA for ordinal data will use diagonally weighted least squares (DWLS) to estimate the model parameters, but it will use the full weight matrix to compute robust standard errors and a mean- and variance-adjusted test statistic. We used the goodness-of-fit index (GFI), which accounts for the proportion of observed covariance between the manifest variables (items), explained by the fitted model (a concept similar to the coefficient of determination in linear regression). Generally, GFI values between 0.9 and 0.95 indicate good fit, and GFI values above 0.95 indicate a very good fit. Bentler's comparative fit index (CFI) was used to correct the underestimation that can occur when samples are small. CFI is independent from the sample size. Values between 0.9 and 0.95 indicate good fit, and values equal to or above 0.95 indicate a very good fit. The Tucker-Lewis index (TLI) varies between 0 and 1; values close to 1 indicate a good fit. Parsimony GPI (PGFI) is obtained to compensate for the "artificial" improvement in the model, which is achieved simply by adding more parameters, i.e., a more complex model may have better fit than a simpler model (parsimonious). Values between 0.6 and 0.8 indicate a reasonable fit and values above 0.8 a good fit. The index root mean square error of approximation (RMSEA) was used to adjust the model simply by adding more parameters. Empirical studies suggest that the model fit is considered good for values ranging between 0.05 and 0.08 and very good for values less than 0.05. The lavaan library from R was used [19].

A total of 7299 respondents provided feedback (response rate of 32.2%), 38% were nurses, 27%

Average composite positive responses were obtained (Table 3). The lowest positive scores were found in composites Work Pressure and Pace, Owner/Managing Partner/Leadership Support for Patient Safety, and Staff Training. The composites with highest scores were Teamwork, Patient

Portuguese data will produce different factors from the American structure.

polycor from R were used [17, 18].

18 Vignettes in Patient Safety - Volume 4

3. Results

3.1. Demographic data

physicians, and 19% secretary/clerk (Table 2).

Care Tracking/Follow Up, and Organization Learning.


Table 3. Composite average positive responses.

#### 3.2. Data screening and pre-analysis

From an initial data set of 7299 respondents, it was removed 587 surveys with missing values and 2408 surveys with more than 2 answers on the option "not applicable," getting a final data set with 4304 surveys, exceeding the minimum necessary. The surveys with one answer in the option "not applicable" were replaced by the middle category in a five-point Likert scale.

#### 3.3. Reliability analysis

Reliability analysis using Cronbach's α was performed on the 10 composites to ensure that individuals were responding consistently to items (Table 4). Considering Cronbach's α, all composites had values higher than 0.6, where composite 1 achieved the highest value and composite 9 the lowest. Analyzing AIIC coefficient, only composites 1 and 3 obtained values outside from the reference. In terms of global consistency, both coefficients lead to a good overall consistency.


Table 4. Internal consistency statistics.

#### 3.4. Exploratory factor analysis

To examine whether a different structure would give a better fit to the data, an exploratory factor analysis was performed. To determine how many composites should be retained, it was obtained the path diagram in Figure 1, where a new structure is proposed. Eight composites were obtained with 37 items (item F6R was not considered since he had an eigenvalue lower than 0.4). Comparing this structure with the one proposed by MOSPSC, composites 1, 5, and 6 did not suffer any changes, composites 2 and 10 gained one item each, composite 4 lost 2 items, composite 8 gained one item and changed other, and composite 3 gained several items from composites 4, 7 and 9.

It was obtained the coefficients for internal consistency for the new proposed structure by EFA (Table 5). In a general way, it was obtained better internal consistency coefficients than with the original structure.

#### 3.5. Confirmatory factor analysis

The fit of the data to the dimensional structure proposed in the original instrument was analyzed using structural equations models through confirmatory factor analysis (CFA). Correlations between composites are presented in Table 6, where it can be observed that there are

Figure 1. Path diagram of exploratory factor analysis. Rectangles represent items, circles represent factors (composites),

Patient Safety Culture in Portuguese Primary Care: Validation of the Portuguese Version of the Medical Office…

http://dx.doi.org/10.5772/intechopen.80035

21

and the values on the arrows are the eigenvalues.

Patient Safety Culture in Portuguese Primary Care: Validation of the Portuguese Version of the Medical Office… http://dx.doi.org/10.5772/intechopen.80035 21

composite 9 the lowest. Analyzing AIIC coefficient, only composites 1 and 3 obtained values outside from the reference. In terms of global consistency, both coefficients lead to a good

Composite No of items Cronbach's α AIIC 1. Teamwork 4 0.82 0.53 2. Patient Care Tracking/Follow Up 4 0.71 0.38 3. Organization Learning 3 0.79 0.56 4. Overall Perceptions of Patient Safety and Quality 4 0.69 0.38 5. Staff Training 3 0.69 0.43 6. Owner/Managing Partner/Leadership Support for Patient Safety 4 0.69 0.36 7. Communication about Error 4 0.75 0.43 8. Communication Openness 4 0.73 0.40 9. Office Processes and Standardization 4 0.63 0.31 10. Work Pressure and Pace 4 0.75 0.42 Total 38 0.92 0.24

To examine whether a different structure would give a better fit to the data, an exploratory factor analysis was performed. To determine how many composites should be retained, it was obtained the path diagram in Figure 1, where a new structure is proposed. Eight composites were obtained with 37 items (item F6R was not considered since he had an eigenvalue lower than 0.4). Comparing this structure with the one proposed by MOSPSC, composites 1, 5, and 6 did not suffer any changes, composites 2 and 10 gained one item each, composite 4 lost 2 items, composite 8 gained one item and changed other, and composite 3 gained several items from

It was obtained the coefficients for internal consistency for the new proposed structure by EFA (Table 5). In a general way, it was obtained better internal consistency coefficients than with

The fit of the data to the dimensional structure proposed in the original instrument was analyzed using structural equations models through confirmatory factor analysis (CFA). Correlations between composites are presented in Table 6, where it can be observed that there are

overall consistency.

20 Vignettes in Patient Safety - Volume 4

3.4. Exploratory factor analysis

Table 4. Internal consistency statistics.

composites 4, 7 and 9.

the original structure.

3.5. Confirmatory factor analysis

Figure 1. Path diagram of exploratory factor analysis. Rectangles represent items, circles represent factors (composites), and the values on the arrows are the eigenvalues.


\*Composites who did not suffer any changes after EFA.

Curve brackets represent added items and rectangular brackets represent removed items from the composite.

Table 5. Internal consistency statistics after structure proposed by exploratory factor analysis.


Table 6. Correlations of the 10 composites.

high values between some composites. This will produce a nonpositive definite matrix of the covariances of the latent variables. In this sense, composite 9 was removed.

Figure 2 shows the relation of the individual items to the composites. The standardized path between coefficients shows the strength of these relations. A coefficient less than 0.1 indicates a low effect; coefficients around 0.3 indicate a medium effect, while large effects are suggested by coefficients higher or equal of 0.5. In this model, coefficients ranged between 0.45 and 0.87.

Figure 2. Confirmatory factor model where composite 9 was removed (34 items).

Patient Safety Culture in Portuguese Primary Care: Validation of the Portuguese Version of the Medical Office…

http://dx.doi.org/10.5772/intechopen.80035

23

Patient Safety Culture in Portuguese Primary Care: Validation of the Portuguese Version of the Medical Office… http://dx.doi.org/10.5772/intechopen.80035 23

Figure 2. Confirmatory factor model where composite 9 was removed (34 items).

high values between some composites. This will produce a nonpositive definite matrix of the

10 0.147 0.117 0.191 0.270 0.357 0.326 0.230 0.153 0.530 1

Composite No. of items Cronbach's α AIIC 1. Teamwork\* 4 0.82 0.53 2. Patient Care Tracking/Follow Up + 8. Communication Openness (D1) 5 0.73 0.35

4. Overall Perceptions of Patient Safety and Quality [F2, F6R] 2 0.76 0.61 5. Staff Training\* 3 0.69 0.43 6. Owner/Managing Partner/Leadership Support for Patient Safety\* 4 0.69 0.36

10. Work Pressure and Pace + 9. Office Processes and Standardization (C12R) 5 0.79 0.42 Total 38 0.92 0.243

1 2 3 4 5 6 7 8 9 10

Curve brackets represent added items and rectangular brackets represent removed items from the composite.

Table 5. Internal consistency statistics after structure proposed by exploratory factor analysis.

8 0.88 0.48

5 0.78 0.41

3. Organization Learning + 7. Communication about Error (D11, D12) + 9. Office Processes and Standardization (C9, C15) + 4. Overall Perceptions of Patient Safety

8. Communication Openness [D1] + 7. Communication about Error (D7R) + 2.

and Quality (F2)

22 Vignettes in Patient Safety - Volume 4

1 1

2 0.495 1

3 0.758 0.588 1

4 0.591 0.533 0.859 1

Table 6. Correlations of the 10 composites.

5 0.570 0.368 0.538 0.516 1

6 0.405 0.302 0.520 0.515 0.549 1

7 0.736 0.679 0.841 0.659 0.499 0.487 1

Patient Care Tracking/Follow Up (D8)

\*Composites who did not suffer any changes after EFA.

Figure 2 shows the relation of the individual items to the composites. The standardized path between coefficients shows the strength of these relations. A coefficient less than 0.1 indicates a low effect; coefficients around 0.3 indicate a medium effect, while large effects are suggested by coefficients higher or equal of 0.5. In this model, coefficients ranged between 0.45 and 0.87.

covariances of the latent variables. In this sense, composite 9 was removed.

8 0.788 0.622 0.773 0.648 0.498 0.463 0.893 1

9 0.820 0.606 0.929 0.765 0.669 0.571 0.798 0.763 1

Table 7 shows the fit of the confirmatory factor analysis for the model proposed in Figure 2. The indices CFI and GFI showed a very good fit; RMSEA and TLI showed a good fit and PGFI a reasonable fit.


Table 7. Confirmatory factor analysis model fit indices.

It was also obtained a good overall internal consistency (Cronbach's α = 0.91, AICC = 0.243).

Considering the model proposed by EFA (Figure 1), it was obtained the CFA model in Figure 3. In this model, coefficients ranged between 0.45 and 0.88.

The goodness-of-fit indices (Table 8) obtained for EFA model (Figure 3) are very similar to the ones obtained for the model proposed in Figure 2.
