Scale of adverse events associated to nursing practices: a psychometric study in Portuguese hospital context *

Objective to contribute to the validation study of the Scale of Adverse Events associated with Nursing Practices in the hospital context. Method cross-sectional study, in public hospital units, in the central and northern regions of Portugal. The exploratory factor analysis of the Scale of Adverse Events associated to Nursing Practices was conducted with a sample of 165 nurses and the confirmatory factorial analysis was made with a sample of 685 nurses. Reliability, internal consistency and construct validity were estimated. The invariance of the model was evaluated in two subsamples to confirm the stability of the factorial solution. Results the global sample consisted of 850 nurses aged between 22 and 59, mostly licensed professionals. The model had a good overall fit in the subscales (Nursing Practices: χ2/df = 2.88, CFI = 0.90, GFI = 0.86, RMSEA = 0.05, MECVI = 3.30; Adverse Events: χ2/df = 4.62, CFI = 0.93, GFI = 0.95, RMSEA = 0.07, MECVI = 0.39). There was a stable factor structure, indicating strong invariance in the subscale Nursing Practices and structural invariance in the subscale Adverse Events. Conclusion the refined model of the Scale of Adverse Events associated with Nursing Practices revealed good fit and stability of the factorial solution. The instrument was adjusted to evaluate the perception of nurses about adverse events associated with health care, precisely nursing care, in the hospital setting.


Introduction
Health care safety has become one of the priorities of national and international health organizations in recent decades. Scientific evidence indicates high rates of adverse events (AE) arising from health care provision, with an impact on patients' health and economic-financial systems, being an important indicator of the safety of care measures. However, the reporting of adverse events is still incipient, making it difficult to estimate their impact (1)(2)(3) .
Health-related AE result from a succession of occurrences that favor unexpected/unwanted events arising from health care interventions due to failure or omission in its provision instead of factors associated with the patients' underlying pathology. These can cause adverse effects/harm to the patients, including permanent damages or even death, influencing the increase in morbidity and mortality, hospitalization time and consequent associated costs, with an impact on the health systems (4)(5) .
AE result from the combination of several factors in highly complex environments, including individual factors related to the patient, factors related to the health professionals such as professional skills, but also economic-financial constraints and institutional weaknesses such as insufficient human resources, overcrowding of patients, inadequate structure and equipment, misfit accommodation care, poor hygiene conditions, among others. There are also aspects related to the work environment, safety culture, leadership style and structure and development of the care process as determinants of health care safety (1)(2)(6)(7)(8) .
The development of indicators and management support instruments for the measurement of care quality and safety is essential to minimize the risks associated with health care, supporting the decisionmaking process with a view to continuous improvement. This is particularly relevant in hospital settings, and  (4) .
However, the initial exploratory factor analysis (EFA) developed by the authors of the scale resulted in a factorial solution slightly different from the predicted, evident mainly in the subscale of "risk perception and occurrence of AE", due to the absence of homogeneity in the criterion of grouping of items according to dimensions. In some dimensions, grouping by type of AE was verified, with association between perception of risk and occurrence. However, with regard to falls and pressure ulcers, the perception of risk is isolated from the perception of occurrence. It was also evidenced the need to remove some items from the original scale, and the suggestion to include new items and restructure previous items. It was then proposed the development of a revised version of the scale, inciting the need for new psychometric evaluation studies (4) .
In this context, given the scarcity of instruments for evaluation of adverse events associated with nursing practices, it is fundamental to evaluate the factorial structure and the invariance of measurement of this instrument, given the importance of obtaining valid and reliable instruments with external and internal validity. The study is of decisive importance given the high potential of the SAEANP to monitor the nurses' perception of AE, taking the instrument as a reference to evaluate the quality of nursing care.
Thus, the present study aims to contribute to the validation of the SAEANP in the hospital context.

Method
A cross-sectional study was carried out to evaluate the psychometric properties of the SAEANP in 12 public hospital units in the central and northern regions of Portugal.
The target population includes nurses who perform functions in the provision of direct care to patients in 71 hospitalization, general surgery, internal medicine and orthopedic services of the hospitals studied.
As inclusion criterion in the sample, only nurses who provide direct nursing care were included. Nurses with management roles ("nurse managers") were excluded.
Data collection took place between January 15 th and September 15 th , 2015.
The sample size was calculated based on the objectives of the study, considering the need for the development of EFA and confirmatory factor analysis (CFA). A sample of 165 individuals was considered for the EFA, taking into account a ratio of three observations www.eerp.usp.br/rlae 3 Neves T, Rodrigues V, Graveto J, Parreira P.
per variable (10) . In the case of the CFA, the sample size was based on a formula for the analysis of structural equations (11) , obtaining an estimate of 151 individuals.
However, because the objective was to perform a psychometric evaluation, the selected sample consisted of the maximum number of participants in the target population, i.e. 685 nurses, to ensure the external validity of the results and the generalization of the conclusions for the population under study.
The data collection instrument was delivered personally to the nurse manager (who had the mediating role in the delivery and collection of questionnaires) of each service, who passed in to all nurses. The instrument was filled according to availability and then delivered in a sealed envelope.
The self-completed instrument includes sociodemographic questions and the revised SAEANP, after an initial evaluation of the psychometric properties, consisting of 55 items (4,12) . This is composed of two independent subscales, with process and result indicators, respectively, nursing practices (NP) and AE. The items are answered in a Likert-type scale of five points, where the score (1) corresponds to "Never" and the score (5) to "Always".
The revised version of the NP subscale (41 items) integrates two new items to evaluate the fulfillment of preventive practices and failures in the application of professional norms, considering the original 10 dimensions, according to Figure 1 (4) .
In the AE subscale (14 items), a new item was included, considering six dimensions, according to Figure 2 (4) .
In a second phase of the study, we performed the CFA and invariance analysis to verify the adequacy of the data to the model under study.
Adherence to the normal distribution of variables was determined by the asymmetry (Sk) and kurtosis (Ku) coefficients, considering that |Sk| <3 and |Ku| <10 did not indicate significant deviations from the normal distribution, which impedes the analysis by the method of maximum likelihood. The presence of outliers was evaluated by the Mahalanobis' square distance (D 2 ). Omitted values were replaced by the mean of the series due to the small percentage in the sample (less than 3%) (14) .
The quality of the overall goodness-of-fit was evaluated according to different indices, considering acceptable values of χ 2 /df < 5, values of CFI and GFI > 0.90, RMSEA < 0.08, where the lowest MECVI indicates the model with the best external validity (14)(15)(16) . The modifications introduced to fit the model were supported by the modification indices (MI) (MI > 11; p < 0.001) produced by the AMOS software as well as theoretical considerations (14) .  intercepts, residuals and variances/covariance of the two groups were fixed. The statistical significance of the difference between the two models was determined by the chi square test (14) .
The reliability and internal consistency of the construct were evaluated by composite reliability (CR) and Cronbach's alpha (α), considering values above 0.70. The validity of the construct was determined in three subcomponents: convergent validity, calculated by the average variance extracted (AVE) by each factor, considering values greater than 0.50 (14)(15) as indicators of convergent validity; discriminant validity was evident when the AVE of each of two factors was equal to or greater than the square of the correlation between these factors; and factorial validity was assessed considering the standardized factor loadings (λ) and the individual reliability (λ 2 ), being also indicators of the goodness of the local fit. Usually, λ above 0.50 and subsequently λ 2 higher than 0.25 (14.17) are considered appropriate, but in the social sciences sometimes lower values are accepted (18) . In the initial SAEANP study, the authors proposed λ greater than 0.30 (4) , an option that was maintained in this investigation.  Regarding the representativeness of the sample, the results of the chi-square test did not show significant differences between the study sample and the Portuguese nurses' population (19) Figures 3 and 4).     (20)(21)(22) .
It was also identified the constitution of a new dimension, PC, composed of two new items of the revised version, regarding patients' privacy and confidentiality, increasing the specificity of the analysis of the instrument in a similar way to an earlier study (12) .
As for item 2.5, this was eliminated because it presented higher saturation in a factor different from the original one (UA), thus conditioning its interpretation. Two previous studies in which this item was eliminated due to a low factor loading (12,23) were also used to support this decision.
In the AE subscale, we opted for a model with six dimensions, similar to the original model. Differences in the RWFS and RWFA dimensions are evident, making it possible to capture these differences, with a subsequent increase in the instrument's specificity, similar to an earlier study (12) . On the other hand, the RFPU and OFPU dimensions were grouped into a single factor, consistent with the other dimensions, which associate the risk It is also pointed out that the factorial solution of the CFA shows a better fit to the characteristics of the study sample compared to the original model (4) . The MI analysis, supported by the theoretical, semantic and conceptual basis, also allowed the refinement of the model through the correlation of the errors of some items.
The GPS factor was eliminated given its internal consistency and the factor loading of item 7.2. This strategy is also based on the results of previous scale evaluation studies, which also excluded this dimension given the values of internal consistency and/or factor loading of the items, suggesting the analysis of the items as indicators of general perception (4,12,23) .
In the NP subscale it was necessary to correlate the measurement errors of items 5.C.1 and 5.C.2, which is theoretically justified by their similarity; both refer to "failures in medication surveillance", and constitute an autonomous factor in the original version (4) .
It was also chosen to correlate the errors of items 5.1 and 6.1 because both refer to nurses' perception of commitment with patients' safety, that is, the risk of occurrence of two types of AE (medication errors and HAIs). It is important to note that, contrary to the "Risk of falls and pressure ulcers", reflecting essentially the clinical condition of the patient, the "risk of medication errors and HAIs" is particularly associated with the intervention of health professionals, thus justifying the correlation among their errors, although they integrate different factors.
Regarding internal consistency, the EFA results showed low values in UA and ROME factors. However, these are similar to those of the initial evaluation of the instrument (UA: α = 0.51, ROME: α = 0.68) (4) and the revised version for the UA factor (α = 0.56) (12) , being even slightly higher in the present study.
In the CFA, there was adequate internal consistency in most of the subscales; however, slightly lower values in the UA and CPPE dimensions of the NP subscale, and ROFPU and ROME of the AE subscale are recognized. It can be seen that the internal consistency of the UA and ROME dimensions is at the threshold of acceptability.
However, there is a higher CR than a previous study in the ROME dimension (FC = 0.63). As for the perception of ROFPU, the same study analyzes them in two independent factors, according to the original version of the scale, also showing threshold values of acceptability (CR: RFPU = 0.70; OFPU = 0.67) (23) . The small number of constituent items of these dimensions is identified as a factor determining reliability, with only two items being identified. However, although low, some authors report that, in the social sciences, α values of 0.60 may be acceptable, provided the results are interpreted with parsimony (24) .
Regarding the construct validity, only one item of the AE subscale is identified with a value slightly less than 0.50, conditioning the individual reliability. Some authors consider factor loadings equal to or greater than 0.30 or 0.40 in EFA acceptable in the social sciences (18,25) .
However, in the CFA, values lower than 0.50 influence factorial validity and, subsequently, convergent validity, by conditioning the AVE value (14) . The item 3.4 (There is a risk of falls in patients) (λ = 0.47) conditioned the AVE value in the ROFPU dimension, but for theoretical reasons and due to its importance to guarantee the evaluation of the latent construct of risk of occurrence of falls associated with this dimension, we opted for its maintenance in the model.  (19) , and there are no

Conclusion
The present study contributed to the evaluation of the psychometric qualities of the SAEANP, an instrument for evaluating the nurses' perception about the AE associated with nursing care in the hospital setting. This scale is useful for management as a tool to support decision making, with a view to improving the work processes and, subsequently, the quality of health care and patient safety.