Structural equation modeling for identification of patient safety antecedents in primary care

In this study, patient safety is understood as an activity that minimizes and removes possible errors and injuries to patients. In this sense, it comprises prevention of medical errors and avoidable adverse events, the protection of patients from harm or injury, collaborative efforts by individual healthcare providers and a strong, well‐integrated healthcare system [1]. However, in the US alone, approximately 5.08% of all adult outpatients, i.e. 12 million adults, experience diagnostic errors every year. It has been estimated that about half of these errors could potentially be harmful [2].

In Poland, patient safety research is scarce. Only general data has been collected. In one region of Poland, 5748 out of 8062 patients’ complaints regarding medical care cases in the years 2006–2008 were recognized as medical errors [3]. To avoid errors and improve patient safety, health care organizations need to implement corrective activity. Despite increasing understanding of patient safety strategies, little is known about the precedents for improving safety in primary care.

A number factors have been found to influence patient safety management strategies, including the facilities available in the practice, communication and collaboration, education regarding patient safety and generic conditions [4, 5]. The present study examines these factors, and some others, with the aim of developing more effective interventions and identifying routes to implementing better strategies.

The principle of concept analysis states that identifying the antecedents and consequences would help in that [6, 7]. Certain antecedents have already been found to positively influence patient safety in primary care [8‐10]: documentation of safety incidents, provision of adequate staffing and facilities in the practice, encouraging communication and collaboration to coordinate safe patient care, provision of in-service education and training for providers and fostering a safety culture among leaders and providers. For the purposes of the present study, these antecedents were classified into four strategies: generic conditions (GC), facilities in the practice (FP), communication and collaboration (CC) and education regarding patient safety (EPS).

Patient safety is needed at all levels of the healthcare system. Regardless of the national setting, a positive safety culture focuses on collective improvement and teamwork [11, 12]. Technology and expertise exist to manage these conditions and limit their impact on patient safety [7, 13]. However, safety problems are most effectively managed through education, appropriate facilities and communication.

The aim of the present study was to test a theoretical model of patient safety interventions based on safety antecedents.

The patient safety strategies included in the present study were adopted from earlier Dutch research [1], examining the variables affecting patient safety management (PSM): facilities in the practice, communication and collaboration, education regarding patient safety. Therefore, it is assumed that (Fig. 1):

The population used in this study was varied (Table 1). We check whether the control variables have no influence on the measured effect (Table A2, Supplementary Information) [14], it is assumed that:

The model was analyzed using structural equation modeling (SEM): an extension of linear regression that enables the modeling of complex relationships between interrelated (and thus correlated) variables. The resulting models account for direct and indirect impacts and can cope with correlated variables. It can also be used to model causal relationships. In SEM, it is possible to use more than one attribute to characterize latent variables. SEM uses an optimization process to create an index score for each latent variable using linear regression.

In this study, SEM was used to characterize latent variables and to produce a causal model incorporating several of them. The models allow the diagnosis and assessment of potential causes of safety management failure, as well as the identification of relevant management methods.

SEM enabled patient safety management to be modelled 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 causal relationship arrows are produced; each was associated with a standard coefficient representing its impact on a scale of -1 (high negative impact) to + 1 (high positive impact) [18, 21]. This model was estimated using the maximum likelihood method, assuming a multidimensional normal distribution. No suspicious response pattern was observed, and, following the outlier labeling rule, no significant outlier was observed; however, as no skewness or kurtosis values higher than one were found, the data was normal. As the sample size was small, analysis was based on the residual bootstrap method and the sandwich standard error estimator. In the case of the structural 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 = 582.298; p = 0.000). The value of root mean square of approximation error (RMSEA) is 0.063 < 0.08 (LO = 0.055; HI = 0.070), standardized RMR = 0.0762 < 0.08 indicates a good fit of the model. The values of χ2/ss = 1.981 < 2 indicate the acceptability of the model. Indices GFI = 0.947 > 0.9 and AGFI = 0.917 > 0.9 assume the values are acceptable, so the model is well suited to the data and can be used in the description [22]. Following the criteria and rules of thumb specified by [23], data analysis results are interpreted in the following sections in terms of model estimations.

The determine the predictive accuracy of the model, the determination coefficients (R²) of the latent variables were estimated. R² values higher than 0.33 are considered acceptable for latent variables [24, 25]. The R² statistics were found to be significantly higher than 0.33 (Table 3) confirming that the model has sufficient predictive accuracy.

As measures such as effect sizes could not be used in the model, it was decided to determine the strengths of impacts based on standardized path coefficients. Considering the relationships between patient safety management and its three antecedents, the results of our model calculations (Fig. 2) revealed that communication and collaboration had a large effect on patient safety (0.434), facilities in the practice had a small effect (0.215), and education on patient safety had a medium effect (0.325). Consequences as generic conditions for patient safety is large effects of education on patient safety (beta coefficient 0.675), of communication and collaboration (0.674) and of facilities in the practice (0.586). The standardized indirect impact of generic conditions for patient safety on the patient safety management is 0.638. The strength is comparable to the impact of education on patient safety.

The aim of this study was to identify the antecedents of patient safety management. Seven dimensions of communication and collaboration, five dimensions of generic conditions for patient safety and eight dimensions of education on patient safety were adapted from previous research [4, 5]. However, only four dimensions were used for practice facilities rather than nine, and seven for patient safety management instead of nine. Each of these dimensions was found to be significant, and were important components of the latent construct. Our findings validated hypotheses H1 and H2 and confirmed that the antecedents of patient safety in primary care are facilities in the practice, communication and collaboration, education and antecedents of them are generic conditions. While all had a positive impact on patient safety, communication and collaboration had the strongest effect, and hence appears to be an important aspect of patient safety management [10].

Primary care modeling of patient safety is complex because causes and effects are interrelated and effects are often mitigated by a set of attributes that cannot be easily modeled using standard techniques. Traditional statistical techniques such as linear regression mainly model the direct impact of uncorrelated variables. Indirect effects are difficult to model and the presence of correlations between explanatory variables may lead to biased coefficient estimates and incorrect model interpretation [19, 26]. Difficulties in measuring patient safety and the complexity of primary care make it desirable to use new methods in the patient safety model [13, 22].

Hypothesis 3 “The relationships with strategies in all subgroups studied are identical” was rejected because only in the case of strategies: general condition and communication and collaboration relationships are identical for each control variables. The impact is different for education and practice facilities strategies.

The literature review recommends choosing a complementary approach through the use of an effective system of reporting incidents from patients [9, 10, 13]. This is not in our model. The most important and pressing action is to implement an information infrastructure that enables capture of adverse events and harm across patient safety systems, thus allowing information to flow freely between GP (general practitioner) and patients on various data platforms [11]. Improved patient safety in primary care results in better quality of care, prevention of injury or harm, and greater patient satisfaction [8].

It has been noted in the literature [1, 9] that reporting incidents, ensuring proper communication among health care providers and fostering a safety culture within the organizational structure allowed medication errors and safety incidents to be minimized, thus improving quality of care in primary healthcare.

Our model shows that it is possible to prevent incidents related to patient safety through professional development and training. The key to improving patient satisfaction is proper communication with the patient and cooperation in the provision of healthcare. Facilities in the practice are also important [8].

The integrated safety model described in the study can provide a greater insight into patient safety mechanisms. Both our present findings and the literature suggest that one of the leading interventions for improving patient safety management is generic condition empowerment. Involvement can be achieved through facilities, communication and collaboration and education. In all contexts, implementing and sustaining these changes requires management of patient safety focused on collective improvement and teamwork. Our proposed model can achieve this at all levels of the healthcare system. Due to the high importance of patient safety, there are pragmatic reasons for using this model.