Background
There is substantial evidence of unsafe care in health systems globally[
1‐
3]. Harms resulting from unsafe care include infections, incorrect medicines or procedures, missed or delayed diagnosis, falls, and are often preventable[
4]. Understanding the factors that influence those behaviors associated with patient safety is an important first step in the development of strategies to improve care[
5]. However, interventions to improve patient safety have typically been developed intuitively and have relied on managers or other experts using strategies such as education, persuasion, or reminding people to change their behavior[
6], rather than adopting a more theoretical approach to understanding and addressing key barriers and levers to behavior change[
7]. Although there is some support for the effectiveness of the aforementioned strategies[
8], evidence suggests there are more and less appropriate times to use particular intervention techniques depending on the specific factors (
e.
g., motivation, confidence, environment, emotion) affecting behavior change[
7,
9]. However, individuals tasked with designing and implementing behavior change interventions may find it difficult to choose from the abundance of health behavior theories, which are often insufficiently specified to determine when or how to modify factors that are to be targeted through an intervention[
10].
The theoretical domains framework (TDF) was developed using an expert consensus and validation process to rationalize and reconceptualize the theoretical constructs from psychological and organizational theory that influence behavior and behavior change[
9]. The framework was developed to make a plethora of behavior change theories more accessible for interdisciplinary audiences involved in implementation, and can be used to understand the barriers and levers to change in a range of contexts[
11,
12]. Researchers applying the TDF to healthcare practitioner (HCP) behavior have, to date, relied on qualitative interviews to understand the factors influencing HCP behavior change[
5,
13]. Although interviews are useful for gaining a detailed understanding of barriers and levers to change, they are resource intensive, time consuming, and often allow for a small sample size, limiting generalizability of findings across, for example, a hospital Trust. A questionnaire based on the TDF, on the other hand, might be a quicker way to identify key domains of behavior change among a larger sample. A questionnaire approach also has the potential to be used in practice by HCPs, improvement teams, or others who have been tasked with facilitating behavior change in their organization. For these groups, the knowledge, skills, and time required to use qualitative interviews might be prohibitive.
Although the TDF has been used to develop a handful of questionnaires examining HCP barriers and levers to working with patients to improve health behaviors such as smoking[
14‐
16], to our knowledge, a questionnaire to understand the factors affecting healthcare practitioner behavior change for patient safety does not exist in the literature. Furthermore, although there are other validated patient safety questionnaires e.g.,[
17], these tend to measure general attitudes, culture, and climate within a particular ward or organization, rather than to understand barriers to performing a specific patient safety behavior. Given the extensive range of behaviors associated with ensuring the safety of patients, it was deemed necessary to address this gap in the literature and develop a measure that accounts for this factor. Therefore, this study reports on the development and initial validation of the Influences on Patient Safety Behaviors Questionnaire (IPSBQ).
Discussion
Based on a theoretical framework of behavior change, 11 scales measuring the psychosocial domains of patient safety behavior change among HCPs were developed and tested. With the exception of Taylor, Lawton, and Conner[
31], to our knowledge, this is first time that CFA has been used to validate the TDF domains, which resulted in good construct validity, discriminant validity, and internal consistency.
This is the first study to develop a measure of the barriers to practitioner behavior change for patient safety (a full version of the IPSBQ can be found in Additional file
2), the novel design of which allows for application to a range of patient safety behaviors. Following further testing and confirmation of its validity, the IPSBQ may be used to identify barriers across a large sample—this information might be complemented by a smaller sample of focus groups to cross validate, and further understand the details about, the key barriers identified. The IPSBQ could potentially act as a tool for developing theoretically underpinned large-scale interventions or, at a local level, for working with staff to co-develop realistic and feasible strategies to address key barriers. The latter approach may be especially relevant if there are differences in key barriers to behavior change between or within organizations, because this would allow for tailoring of interventions for specific contexts.
Despite establishing initial reliability and validity of the IPSBQ, a number of limitations should be noted. First, although the sample size was adequate, a larger sample would allow for increased confidence in the reliability of the measure. Second, we did not directly ask participants about the regularity with which they checked the position of NG tubes, so were unable to apply a filter to assess differences in barriers according to the level of performance of the target behavior. Third, the RMSEA index of fit result did not meet agreed standards for construct validity[
23]; however there are variations for optimal levels in the literature, for example 0.06 has been suggested as the ideal maximum[
24], but 0.08 is also considered an acceptable upper limit[
32]. The weak RMSEA index may be partly related to the attempt to ensure each domain contained two items in order to reduce the participant burden of completion. Even so, several measures reduced to two items per domain have demonstrated valid and reliable properties, and been successfully used in healthcare research[
33]. Furthermore, some domains (
e.
g., beliefs about consequences, social influences) are represented by a large number of constructs, which might be viewed as relevant to more than one domain (
e.
g., the ‘anticipated regret’ construct can be found in both the beliefs about consequences and emotion domains; the ‘social/group norms’ construct can be found in both the social influences and social and professional role and identity domains); this makes it difficult to both generate a questionnaire that represents all of the constructs, does not take a considerable amount of time for participants to complete, and which demonstrates good discriminant validity. We have worked with each of these constraints to produce the first validated version of this measure. For example, we worked with clinicians in the development of all of the items for each domain, and consulted the interview questions from the TDF[
9], literature regarding particular domains (
e.
g., Theory of Planned Behavior for motivation and goals, beliefs about capabilities, social influences, beliefs about consequences), and work which has recently been undertaken to develop measures based on the TDF e.g.,[
11,
31]. Nonetheless, these points highlight the need for additional work to improve the reliability and validity of the tool, and also demonstrate some of the implications that operationalizing heterogeneous domains into a parsimonious questionnaire can have for selecting some items and omitting others. Fourth, this study has not confirmed the criterion validity of the IPSBQ on patient safety behavior. Although these early results indicate that the IPSBQ can detect higher reported barriers for individuals within Trusts demonstrating lower compliance with the target patient safety behavior, further work is needed to establish whether this measure can explicitly demonstrate criterion validity. In the first instance, this could be achieved by adding a self-report measure of behavior to the questionnaire.
The NG tubes alert provides 17 recommendations for NHS organizations/individuals to achieve, many of which involve a range of behaviors. This can make it difficult for HCPs to define an appropriate target behavior to address, because it requires consideration of compliance—(identifying which elements of current practice did not meet the recommendations[
34]), specificity (focusing on a specific behavior that it is possible to change following identification of associated barriers[
9]), and impact (defining a behavior that is likely to have most effects on outcomes[
35]). During this work, other possible key behaviors were identified due to low compliance, but teams decided that targeting the use of pH as the first line method for checking tube position was the crucial aspect of the process. This was because teams recognized that in addition to low compliance, changing this behavior had the potential to not only prevent the need for X-ray (and therefore reduce the risk of misinterpretation), but also to improve the chances of pH being used for checking tube position for subsequent feeds, thus reducing the need for multiple X-rays and further risk of misinterpretation (impact). Nonetheless, this highlights the complexities associated with defining a specific target behavior for change, especially if attempting to use a description of the behavior in relation to a set of questionnaire items that relate to 11 domains of behavior change. Future work with the TDF in the context of patient safety should investigate how appropriate it is to select a single behavior if clinicians are performing multiple behaviors, or whether it is possible to operationalize the TDF to elicit behavior change when multiple behaviors are targeted.
While a questionnaire of this kind might be a useful method for identifying the relative strength of barriers, the absolute strength of a barrier is more difficult to measure. None of the domains assessed in this questionnaire had a mean score above the mid-point of the scale, despite compliance with this guideline being low according to case note audit results. This might imply a tendency for people to underestimate the barriers to behavior change, or alternatively an affinity to respond in a socially desirable way. Future work might assess the impact of using a four-point Likert scale on reporting of barriers, because evidence has demonstrated that this can reduce social desirability response bias[
36]. The potential for this measure to be used or adapted to identify levers to support behavior change might also be an area worth investigating.
We have presented an indicator of criterion validity by demonstrating that the total barriers score for each Trust increases as compliance with the target behavior decreases. However, a sum score may not be entirely appropriate because this indicates that all domains are equivalent proximal predictors of behavior. Nevertheless, although many of the constructs from within different domains stem from complex mediating processes from the theories from which they are sourced and are arguably inter-related, to our knowledge this has not yet been tested. Therefore, the current results perhaps provide a basis or rationale for future research to examine the extent to which each domain (and which of the associated constructs) predicts behavior in the context of patient safety.
In addition to undertaking further work to test the 23-item measure, and improve the reliability, validity, and generalization of the IPSBQ, the next stage of this research should also aim to establish whether the key barriers identified by this measure can be targeted with theoretically underpinned and pragmatic interventions. Following this, the impact of these interventions on changing healthcare practitioner behaviors, as well as their reported barriers[
37], should be tested. Given the IPSBQ has been used to identify the factors affecting change for one patient safety behavior, work should also be undertaken to understand if the measure can be used to reliably identify barriers to other target behaviors; this next phase is currently underway in areas relating to patient safety and midazolam, gentamicin, and medicines reconciliation. Finally, since the development of the IPSBQ, an updated version of the TDF has been published[
38]; this includes three new domains (optimism, goals, and reinforcement) and revisions to some of the constructs associated with each domain. Therefore, the new aspects of the TDF will need to be considered for inclusion in a revised measure.
Endnotes
aThe ‘nature of the behavior’ determinant was, as in the Michie et al. (2005) paper, accorded a different order to the rest, as it describes the dependent variable, which in this case is ‘using pH as the first line method to check tube position’. It is therefore not treated as a domain of behavior change, but its constructs (such as habit, stages of change, and representation of tasks) were considered throughout the development of the questionnaire in relation to the target behavior.
bMIs were provided by AMOS for all parameters constrained to zero and indicate when an item may cross load or load onto a different factor[
20]. The standardised residual matrix identifies pairs of items that are either under or over-predicted by the model[
21], for which values > +/−2.58 are considered to be large[
22].
cTwo constructs display discriminate validity if the average of the estimate of variance extracted exceeds the square of the correlation between the two latent constructs, and the confidence interval around the correlation estimate between the two factors includes 1.0. Inter-item correlations were used to test for internal consistency, with values above 0.15-0.50 being the optimal range[
23‐
25]. Full workings out and results for discriminant validity are available from the author.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
NT led the design and coordination of the study, performed the statistical analysis, and led the writing process. SP performed the statistical analysis, and helped to draft the manuscript. VR advised on questionnaire item content, and helped to draft the manuscript. BS participated in the design of the study, and helped to draft the manuscript. RL participated in the design of the study, advised on questionnaire item content, and helped to draft the manuscript. All authors read and approved the final manuscript.