The HSOPS is one of the most frequently used questionnaires to assess safety culture in health care settings. Until now, this questionnaire has been used to evaluate safety culture from employees' points of view. There exist an increasing number of studies testing how consistently the HSOPS questionnaire measures safety culture dimensions [18, 29, 31–33]. However, these surveys have all been tested with medical staff only. Because it is important to test whether the HSOPS is also applicable for assessing single views of a hospital's safety culture, the purpose of our study was to test the psychometric properties of the HSOPS adapted for hospital management (HSOPS_M).
The CFA indicated that the factor structure of the original HSOPS fits the data of the German version for medical directors. The factor model exhibited an acceptable-to-good global data fit. Furthermore, the local fit indices were considered acceptable. Regarding the indicator reliability, most indicators, except one, exceeded the acceptable values. All factors reached the recommended critical values for the factor reliabilities, but only four factors reached adequate AVE values. These results suggested a good convergent validity. The results of the local fit indicators, especially for the AVE, found in this study are comparable to the results of the Swiss-German version of the HSOPS .
According the construct validity, only three dimensions reached satisfactory values for the FLR. Therefore, the construct validity of the factor model can be considered less acceptable. The values of FLR indicated that the factors measured not readily distinguishable dimensions. The values of the FLR are similar to the Swiss data analysis . Furthermore, the inter-correlations between all 12 safety culture dimensions ranged between .13 and .64. We found several correlations higher than .5, which supported the result that the dimensions were not at all independent of each other. One possible reason for the high values of FLR and high intercorrelations could be that theoretically correlated dimensions (e.g. Feedback and communication about error and Communication openness; Hospital management support for patient safety and Supervisor/manager expectations/actions) are measured with different constructs. This suggested further investigation, especially on the question whether these dimensions should be measured in one dimension. Nonetheless, as expected, all 12 safety culture dimensions correlated with the outcome variable Patient safety grade. We found a high correlation between Patient safety grade and Overall perceptions of safety, which is a good indication of the validity of the latter dimension.
Finally, the analysis of Cronbach's alpha signified that the dimensions have an acceptable level of reliability. In nine out of the 12 dimensions hypothesised in the origin factor model, the Cronbach's alpha ranged between .73 and .87. In addition, the alpha of the factor Supervisor/manager expectations/actions was not much below the recommended cut-off value of .70. In particular, for Communication openness and Organizational learning, the lower values of Cronbach's alpha can probably be attributed to the different survey designs (e.g. measuring management perception versus the perceptions of frontline staff). Nevertheless, a comparison of these reliabilities with other European HSOPS surveys showed that Communication openness [18, 32] and Organizational learning [18, 31, 32] repeatedly had low Cronbach's alpha values.
Overall, the construct validity indicated that further scale refinement is needed to improve the questionnaire. To minimise differences between the survey versions, we refrained from reducing or adding any scales within the instrument. Nonetheless, model modifications should not generally be excluded. Especially in cases of high intercorrelations between dimensions, which are theoretically high correlated, further scale refinement could lead to better psychometric properties. In this respect, we agree with Pfeiffer and Manser  that the set of dimensions within the HSOPS still has to be optimised.
The findings of our study are limited by the following aspects. The results of this study are based on a cross-sectional mail survey with a response rate of 45%. Although little is known about potential non-response bias with these kinds of surveys, we assumed that the attitudes of the responding medical directors do not differ from those of non-responding medical directors .
Within the scope of this study, we were not able to examine the relationship between patient safety culture and objective patient safety outcomes, such as patient safety indicators or frequencies of medical errors. Therefore, we agree with previous suggestions [7, 29] that more evidence is needed on the relationship between patient safety culture and patient safety outcomes.
Comparing the psychometric properties of the HSOPS_M to those of the original HSOPS for hospital employees means not only comparing different countries, but also different methods. Most safety culture surveys are used to measure safety culture from the frontline staff's points of view. Assessing safety culture only with medical directors excludes the views of frontline staff and does not take the potential differences between hospital units [35–38] into account. Therefore, we think the area of application of the HSOPS_M is different from traditional hospital-related safety culture instruments. According to Rousseau , we presupposed that the points of view of key informants, such as medical directors, were representative of hospital professionals in identifying safety culture for the whole hospital. Hospital managers are expected to make decisions regarding quality improvement and patient safety issues. In addition, essential decision makers - such as medical directors - have a comprehensive knowledge about their organisations. Therefore, questioning the top management offers a different approach to measuring safety culture and providing aggregated organisational data. For analysing the safety culture in different hospitals units, further research should consider using the HSOPS_M for hospital unit managers as well.
Finally, the HSOPS_M was embedded in a larger questionnaire within the HOSQua-study, which could be a possible factor that influenced the responses of the medical directors. According to Linsky , we assumed that the length of the questionnaire would not necessarily influence the validity or reliability of the HSOPS. Nonetheless, further analyses on validity and reliability should be performed using the HSOPS_M questionnaire only.