Article Text

Download PDFPDF

Modifying head nurse messages during daily conversations as leverage for safety climate improvement: a randomised field experiment
  1. Dov Zohar1,
  2. Yaron T Werber1,
  3. Ronen Marom2,
  4. Bruria Curlau2,
  5. Orna Blondheim2
  1. 1Faculty of Management, Technion Israel Institute of Technology, Haifa, Israel
  2. 2Emek Medical Center, Afula, Israel
  1. Correspondence to Dr Dov Zohar, Faculty of Management, Technion Israel Institute of Technology, Technion City, Haifa 32000, Israel; dzohar{at}


Background Recent literature reviews lament the paucity of high-quality intervention studies designed to test safety culture improvement in hospitals. The current study adapts an empirically supported strategy developed for manufacturing companies by focusing on patient care and safety messages head nurses communicate during daily conversations with nurses.

Methods The study was designed as randomised control trial coupled with before-after measurement of outcome variables. We randomly assigned 445 nurses working in 27 inpatient departments in a midsize hospital in Israel to experimental and control groups. Ten randomly chosen nurses in both groups filled a brief questionnaire referring to last conversation with head nurse. One month later, head nurses in the experimental group received individual feedback, comparing individual with mean hospital scores, coupled with self-set goals for the following feedback session. Head nurses in the control group received no feedback, except for a summary report by the end of intervention.

Results Patient care messages increased by 16% and professional development messages by 12%, accompanied by 17% decline in nurse-blaming messages in the experimental group, remaining unchanged in the control group. Such changes led to statistically significant increase in patient care behaviours (17%), safety climate (13%), teamwork (9%) and supervisory leadership quality (18%). Rule-compliance messages and workaround behaviours remained unchanged in experimental and control departments.

Conclusions These data support the utility of our intervention strategy for improving patient safety climate and resultant caring behaviours in healthcare organisations. The fact that our intervention used easy-to-deliver feedback requiring only two sessions minimised its organisational costs.

  • Patient safety
  • Safety culture
  • Healthcare quality improvement

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


Recent reviews of the culture intervention literature in healthcare organisations lament the slow progress in reducing medical errors in general and shortage of high-quality intervention studies in particular.1–6 For example, in a recent systematic review of the culture intervention literature, Sacks et al4 identified 47 studies, out of which only 4 studies were designed as randomised trials, with the rest being largely observational studies. Given this state of affairs, the present study was designed as a randomised control trial testing an intervention strategy that has proven effective in manufacturing organisations.

Our strategy aims at improving departmental patient safety climate, constituting the measurable dimension of safety culture. Safety climate has been shown to offer robust prediction of treatment errors,7 hospital readmissions,8 preventable complications,9 reporting of treatment errors10 and patient safety scores.11 In fact, recent meta-analyses of safety climate studies indicated that it offers a strong and reliable prediction of safety performance across a variety of industries, including healthcare.12–15

The original paper on safety climate defined it as ‘shared employee perceptions about the relative importance of safe conduct in their occupational behavior’ (p. 96).16 This definition identifies safety climate as consensual or shared social cognition regarding the relative importance or priority of (employee or patient) safety versus competing goals such as efficiency or profitability. In healthcare organisations, such socially shared perceptions inform employees of organisational and departmental commitment to patient safety, guiding appropriate role behaviours.17 Given that ‘doing no harm’ by protecting patient safety is an ethical norm and a professional code of conduct, practically all hospitals adopt patient safety policies.

At the same time, however, it has been repeatedly observed that adoption of such policies does not necessarily result in significant improvement in patient safety behaviours. In fact, literature reviews indicated that the majority of hospitals that adopted formal policies for treatment quality or patient safety have failed to demonstrate subsequent improvement in respective medical performance criteria.18 For example, research testing the relationship between adoption of Total Quality Management policies and patient safety outcomes, using a sample of >3000 hospitals across the USA, indicated that, except for early adopters, effect size was minimal or non-existent.19 Similar results have been reported with regard to ISO-9000 effectiveness.20 Surveys of senior management in >500 organisations concluded that about two-thirds of respondents consider this standard's adoption as a failure due to lack of effects on quality outcomes.21 Such results suggest widely spread gaps or discrepancies between formally adopted policies or care quality standards and its daily implementation at organisational subunits, labelled as policy–practice decoupling.22 ,23 Simply stated, decoupling refers to failing to ‘walk the talk’.24 ,25

Reviews of the decoupling literature suggest that one of its primary causes is the high costs of policy implementation stemming from the competing demands for efficiency or profitability.26 Furthermore, employees become aware of such discrepancy between words and actions and use the latter as an indicator of true or realised (vs formally espoused) priorities in the organisation.27 ,28 Such findings have three implications: (a) patient safety decoupling can be frequently observed in healthcare organisations; (b) hospital employees take notice of such policy–practice misalignment; and (c) employees use managerial or supervisory actions, rather than words, as indicators of true (ie, implemented) priorities. Considered jointly, these results suggest a link between safety-related decoupling and climate perceptions. Such a link implies that since climate perceptions concern assessment of the true (vs formally espoused) priority of safety behaviours, any attempt at improving safety climate perceptions of front-line staff must use an intervention strategy that promotes a change in supervisory safety behaviours as leverage for change. In the context of healthcare organisations, attempts at improving safety climate at hospital wards must be based on modification of head nurse behaviours in each ward. Furthermore, such change must be demonstrated by head nurse behaviours or actions in situations presenting competing demands for increased efficiency or cost-cutting.

Supervisory messages in daily conversations

Given the fact that language serves as the primary medium for interpersonal processes,29 it follows that leaders must talk with followers to clarify their role expectations and identify desired role behaviours, turning language and discourse into an inherent dimension of leader behaviour.30 Consequently, attempts to improve employee climate perceptions can use modification of work-related discourse through which supervisory role expectations are communicated to group members as leverage for change.

Intervention-induced changes in supervisory role expectations must remain stable and consistent, offering group members multiple opportunities for its observation and validation. In other words, changed supervisory role expectations must be communicated in a variety of situations, especially those presenting competing (safety vs efficiency) demands. When one's head nurse remains committed to patient safety across a variety of situations, it offers multiple opportunities for ward nurses to observe it and develop socially shared climate perceptions, promoting a climate change. One method for measuring such cross-situational stability is by focusing on verbal exchanges taking place in routine or daily conversations with nurses as opposed to those taking place in formal occasions such as weekly departmental meetings. Whereas the latter is likely to be used for communicating formally espoused policies, it is during the former that supervisory actions revealing of genuine commitments are likely to be revealed. Given the prevalence of word–action misalignment in hospitals, nurses are assumed to use supervisory role expectations communicated during the latter in order to untangle espoused policies from enacted behaviours. Given all of the above, our intervention targeted the modification of supervisory work-related messages communicated during daily conversations as the leverage for safety climate change. A similar intervention strategy was tested by the lead author in a number of other industries, resulting in significant and stable safety climate and safety performance improvements.31 ,32

Feedback intervention for climate change

The primary leverage for modifying nurse-manager messages in routine conversations was individual feedback, coupled with a benchmark (mean message scores) and self-set goals for desired message scores. Performance feedback is one of the best studied and most effective intervention strategies,33 provided its recipients consider it accurate, non-threatening and suggesting a need for behaviour change.34 ,35

Feedback data in our project indicated the (perceived) priority or significance of four different nursing aspects (ie, patient care, professional development, rule compliance and nurse blaming) based on messages communicated by head nurse in the most recent conversation with ward nurses. Use of subjectively perceived, rather than objectively recorded, messages is based on the prevalence of safety policy–practice misalignment. In such cases, although communicated messages may refer to patient safety, its subtext could imply that meeting other goals is no less important, resulting in diminution of the (perceived) priority of supervisory patient safety expectations. For example, a safety-enhancing message is one in which the head nurse indicates that dispensing of medications ought to follow standard procedures such that if patient is not in bed it should not be left on its side table. However, the safety-enhancing effect of such a message might be mitigated when it is followed by an additional message asserting that medicine administration has to be completed before the doctors' daily round. When such accompanying message is communicated on a day in which there is excessive overload due to a missing staff member, it creates a subtext likely to diminish the assessed priority of patient safety.

This example highlights the fact that the distinction between sent and received messages is especially important when such messages relate to priorities associated with competing demands. Assessment of priorities requires an interpretive sense-making process on behalf of employees stemming largely from the difficulty of untangling or discriminating between espoused and enacted priorities.27 ,36 Considering that the objective of our project was to improve departmental safety climate, our feedback data were based on perceived, rather than literally sent, messages. As noted above, our feedback intervention included four key dimensions of the nursing role in hospitals (ie, patient care, professional development, rule compliance and nurse blaming).37–39 These dimensions were chosen as the targets for intervention due to its ascribed role in promoting care quality and patient safety.1 ,3 ,5 ,6

Assuming a performance facilitation effect for the feedback intervention and goal setting, it is expected that a stable, cross-situational increase in supervisory safety messages will lead nurses to modify their safety climate perceptions. Furthermore, given the well-documented effect of safety climate on safety behaviour,12–15 we expect that higher safety climate will improve patient caring behaviours, accompanied by improved team leadership. The latter is based on studies reporting a strong, bidirectional leadership–climate relationship such that modifying the one affects the other.15 ,40 In the context of healthcare organisations, this relationship is expected due to the fact that a leader who prioritises staff's patient care behaviours, professional development and using errors as opportunities for learning rather than blaming is likely to be considered a better leader than one who fails to refer to such issues.5 ,9 In the present project, leadership was measured with one of the most widely used leadership scales, known as Leader Member Exchange (LMX).41 This scale reflects the quality of leader–member relationships, operationalised by openness, reciprocity and trust between leader and followers. As will be noted below, given that feedback data refer to communication aspects such as inviting nurses to request (professional, personal) help whenever needed or clarifying that the head nurse is interested in professional development of nursing staff, it is expected that our intervention would improve supervisory leadership, reflected by higher LMX scores.

Given all of the above, the primary objectives of this study are

  1. test the effect of our feedback and goal-setting intervention on supervisory messages communicated during randomly chosen routine conversations with nurses;

  2. test the effect of our feedback and goal-setting intervention on front-line staff safety climate perceptions and patient-caring behaviour;

  3. test whether simplifying feedback provision such that it can be conducted by non-experts and limiting it to two sessions would result in a low-cost intervention.



The sample included 445 nurses working in 27 inpatient departments in a midsize hospital in Israel. Using random selection of departments into experimental and control groups, the former included 232 nurses in 13 departments and the latter included 213 nurses working in 14 departments. Average nurse tenure was 14.4 years (range 1–45), and average head nurses tenure was 10.2 years in that role (range 1–30). Most nurses were female (91%) and worked full time (62%).


The project was designed as randomised trial coupled with before-after measurements. During the intervention, randomly selected nurses in both groups were approached at randomly determined times during work shifts and asked to fill a one-page survey, taking some 3–4 min to complete. Items referred to the last conversation they have had with their head nurse, as well as their own behaviour over the last hour. The survey was anonymous and could only be identified by hospital department. After collecting data from 10 nurses in each department, head nurses in the experimental group received feedback, whereas those in the control group received no feedback. Data collection and analysis was followed by feedback provision, taking 1 month to complete each, resulting in a bimonthly cycle. Two such cycles were performed in this intervention. Head nurses in the control group received a summary feedback by the end of intervention. Before-after measurements were conducted two months before and after the intervention. Feedback sessions included the following information: (a) departmental ratings of perceived supervisory messages communicated in the last 10 routine conversations, (b) a benchmark representing the average ratings for all hospital departments and (c) self-set goal for desired ratings to be achieved by the second feedback session.


Before-after intervention measures

Patient safety climate was measured with a brief 15-item scale of the empirically validated Patient Safety Climate in Healthcare Organizations .42 Items of this scale were accompanied by a five-point rating scale ranging from strongly agree (5) to strongly disagree (1). Sample items are ‘My unit does a good job managing risks to ensure patient safety’ and ‘I am rewarded for taking quick action to identify a serious mistake’ (Cronbach's α=0.78).

Supervisory leadership was measured with the seven-item LMX (LMX-7) scale.41 Items were followed by a five-point rating scale ranging from strongly agree (5) to strongly disagree (1). Some words were modified to fit the context of healthcare industry. Solid psychometric properties of the LMX-7 scale are well documented in a meta-analytic study comparing different leadership measurement scales.43 Sample items are ‘My head nurse understands my problems and needs well enough’ and ‘Working relationships with my head nurse is effective’ (α=0.94).

Teamwork was measured with nine items taken from the Team Climate Inventory subscales of Team Interaction Frequency and Openness.44 Responses were based on the same five-point Likert scale described above. Sample items are ‘We interact frequently’ and ‘We share information generally in the team rather than keeping it to ourselves’ (α=0.82).

Intervention phase measures

Head nurse messages were measured with an eight-item empirically validated checklist developed by the lead author in the context of manufacturing organisations,31 ,32 after its adaptation to healthcare organisations. Using a series of interviews at the beginning of this project, we developed two items for measuring each of the four message types selected for this intervention. Sample items are ‘Made me understand she expects me to request help whenever needed (patient care)’; ‘Head nurse was interested in my professional development, not just finding a solution to the matter at hand (professional development)’; ‘I understood that she would blame me for any error or mistake (nurse blaming)’; and ‘Got the impression that she would be upset for any rule violation even when no harm has been done (safety rule compliance)’. A three-point scale was used, rating the perceived priority for each item, ranging between 0 (none) and 2 (high). Considering that each message type was measured with two items, its score was based on item-pair average. These scores were presented as bar graphs to participants during feedback sessions.

Nursing care behaviours were measured with four items included in the same checklist described above. In this case, nurses were requested to describe their activities over the last hour, focusing on two primary activities identified in our review of the nursing literature, that is, quality of patient care and workarounds or rule violations, constituting a dominant strategy for coping with workload. As before, we developed two items for measuring each of these behaviours. Sample items include ‘I tried to listen caringly to patient/family regardless of how busy I was at that time (patient care)’ and ‘I tried to expedite caregiving by ignoring some guidelines or rules (workaround)’. Notably, these items refer to actual behaviours having been performed over the last hour, attempting to minimise self-report bias. Items were accompanied by a three-point scale, rating the extent to which such behaviour was considered valuable and praiseworthy in the department, ranging between 0 (low) and 2 (high). As before, we used the average of each item-pair to derive the nursing behaviours score.


Table 1 presents the descriptive statistics for our study's outcome variables, broken down by group (experimental and control) and time (pre intervention and post intervention). Such breakdown indicates that our analysis constitutes testing of the statistical significance of difference in differences (ie, intervention vs control and before vs after). As can be seen, correlations between variables have remained unchanged although its mean values have changed, reflecting the effect of intervention in experimental group departments and lack thereof in control group departments.

Table 1

Means, SDs and correlations of before-after intervention variables

Table 2 presents descriptive statistics regarding changes in nursing care behaviours and supervisory messages as a result of our intervention. Notably, in the experimental group, supervisory messages relating to patient care and professional development increased between feedback sessions (ie, 0.92 vs 1.24 and 1.58 vs 1.69, respectively) and nurse blaming decreased (ie, 0.44 vs 0.28), remaining little changed in the control group. At the same time, rule-compliance messages increased equally in both groups (ie, 0.98 vs 1.18 and 1.09 vs 1.38, respectively). Table 2 indicates that such change in supervisory messages was accompanied by increased patient-caring behaviours in the experimental group (ie, 1.74 vs 1.88) accompanied by an unanticipated decrease in the control group (ie, 1.79 vs 1.72). Contrary to expectations, workarounds remained unchanged in both groups (ie, 0.74 vs 0.73 and 0.84 vs 0.83, respectively), apparently due to high workload affecting all hospital departments.

Table 2

Means and SDs (in parentheses) of nursing care behaviours and supervisory messages during the two-feedback intervention phase

Intervention effects on supervisory messages have been tested with Hierarchical Mixed Model analysis using SAS Mixed Procedure software (SAS V.9.3). Linear mixed models were applied for each of the four message types, that is, patient care, professional development, nurse blaming and rule compliance. The independent variables were time (T=0 for first feedback; T=1 for second feedback); group (G=1 for experimental group; G=2 for control group). The repeated option was used to account for differences between times regarding the variances between hospital departments and for correlations between times within departments. Such a statistical model allows testing of interaction effects between time and group for each of the message types.

Table 3 presents the outcomes of this analysis. As can be seen, except for rule-compliance messages, interactions for the other three message types were significant. Figures 13 present the shape of these interactions. These figures indicate that there was significant increase in patient care and professional development messages, accompanied by a significant decrease in nurse blaming messages in experimental departments. Effect size estimates were computed by using the R2LR statistic for comparing the effect of intervention on supervisory messages in experimental and control groups.45 This statistic compares the R2 values offered by a main effect model with that offered by a model that includes the time×group interaction. R2LR differences accounted for 16% increment in explained patient care messages, 17% decline in nurse blaming messages and 12% increment in professional development messages. Control department data indicate either no change or an unexpected decline in the frequency of professional development messages.

Table 3

Results of mixed-effect model analysis of intervention effect on supervisory messages in experimental and control groups

Figure 1

Interaction between group and time for perceived supervisory patient care messages.

Figure 2

Interaction between group and time for perceived supervisory nurse-blaming messages.

Figure 3

Interaction between group and time for perceived supervisory professional development messages.

Table 4 presents the outcomes of the hierarchical mixed model analyses using department nurses' patient care and workaround behaviours as outcome variables. As can be seen in this table, whereas the interaction term for patient care behaviours was significant, the same was not true for workaround behaviours. The latter apparently results from having to cope with high workload in all hospital departments. Figure 4 describes the shape of interaction for patient care behaviours, indicative of improvement in experimental group departments while remaining little changed in control group departments. Using the same R2LR statistic for estimating effect size, it indicates a 17% increment in patient care behaviours by nurses in the experimental group.

Table 4

Results of mixed-effect model analysis of intervention effect on nursing behaviours in experimental and control groups

Figure 4

Interaction between group and time for patient care nursing behaviours.

Finally, table 5 presents the results of the same hierarchical mixed model analyses, using before and after intervention data as outcome variables (ie, safety climate, teamwork and supervisory leadership). As can be seen, all interaction effects are significant. Figures 57 presenting these interactions suggest the same pattern of improvement in experimental group departments accompanied by no or negative change in control group departments. Using the R2LR statistic to estimate effect sizes, it indicates 13% improvement in safety climate, coupled with 9% improvement in teamwork. The unexpected negative change in supervisory leadership scores for control group head nurses is described in figure 7, resulting in 18% difference in leadership scores between experimental and control groups.

Table 5

Results of mixed-effect model analysis of intervention effect on preintervention and postintervention variables in experimental and control groups

Figure 5

Interaction between group and time for departmental patient safety climate.

Figure 6

Interaction between group and time for departmental teamwork behaviours.

Figure 7

Interaction between group and time for supervisory leadership (LMX).


Results of this intervention project indicate that modification of nurse manager role expectations communicated in routine conversations across different situations can induce statistically significant change in patient care nursing behaviours and departmental safety climate. Specifically, increased frequency of patient care and professional development messages, accompanied by reduced frequency in nurse blaming messages, has led to significant increase in patient care behaviours, safety climate, teamwork and supervisory leadership, measured two months following the intervention.

Having adopted a randomised control trial design, coupled with before-after measurement of outcome variables, it is possible to assume causality for this messages–practices relationship due to the fact that changes have taken place in experimental group departments, remaining unchanged in control group departments. The fact that such changes have taken place following two easy-to-deliver feedback sessions suggests cost minimisation of our intervention strategy, even though there is a need for further testing of its long-term effects. At the same time, it should be noted that rule compliance messages and workaround behaviours have failed to discriminate between experimental and control departments, stemming, apparently, from contextual effects in the hospital environment.

Reported results offer empirical support for two theoretical claims. First, it supports the claim that employee climate perceptions focus on assessment of enacted (vs formally espoused) priorities at the workplace. Given that patient safety is expected to assume highest priority in healthcare organisations, virtually all hospitals adopt policies and procedures to that effect. At the same time, as noted above, due to competing demands such as profitability or efficiency, policy–practice decoupling has become widely spread. Consequently, assessment of true (vs formally espoused) priorities requires hospital employees to focus on work experiences in situations involving such competing demands. In the present case, our intervention uses supervisory role expectations communicated in randomly chosen routine conversations as a means for safety climate change. Specifically, using feedback coupled with self-set goals, we demonstrate that modified cross-situational supervisory messages can lead to consequent changes in nurse patient care behaviours and safety climate perceptions.

Second, our results support the claim that, due to the prevalence of policy–practice decoupling, changes in safety climate require a stable change in managerial commitment that can be observed by front-line employees across a variety of situations presenting competing demands. Formal declarations that are not implemented on a daily basis would lead employees to realise that management does not walk the talk, leading them to focus on truly prioritised aspects of their work to the detriment of patient safety. Such results suggest that climate change interventions must be based on modification of managerial behaviours rather than adoption of new formal policies or offering a one-time training session or workshop. It is the stability of change attesting to a genuine commitment to patient safety that can serve as leverage for safety culture change.

Practically speaking, our intervention was designed to minimise organisational costs due to the following attributes: (a) data collection was designed to minimise disruption of nursing activities; (b) feedback sessions could be delivered by graduate students, requiring no special skills; and (c) intervention results could be observed after few feedback sessions. This design was based on previous intervention projects conducted by the lead author in manufacturing organisations, in which a two-feedback design offered similar improvement in safety climate and safety behaviour.31 Furthermore, the stability of postintervention change was shown to extend for at least 1 year.32 Our study offers a viable, evidence-based intervention strategy filling the gap in culture improvement research.

Study strengths and limitations

Study strengths largely relate to its randomised control design coupled with before-after measurement of outcome variables. Such design allows inference of causality between changed supervisory (verbal) behaviours due to individualised feedback and resultant changes in nurse patient caring behaviours, safety climate perceptions and team working. Study weaknesses largely refer to the following issues: (a) changes in nurse behaviour and safety climate perceptions were measured two months after intervention, requiring further testing of longer-term effects. Although such effects were indicated in the manufacturing industry, it needs to be tested in the healthcare sector; (b) measurement of nurse patient caring and workaround behaviours was based on self-report rather than observational data. Although such self-reporting related to recent actual behaviours, designed to minimise single-source bias, there is a need for complementing it with more objective data; (c) our study did not include measurement of objective nursing-based patient outcomes such as medication errors and recordkeeping. As before, there is a need to include such measurement in future studies; and (d) the fact that our intervention failed to effect workarounds, resulting in no difference between the experimental and control groups, requires further research to test whether different supervisory messages or intervention strategies would have resulted in such a change. Despite such limitations, we hope this study stimulates similar attempts in the healthcare industry.



  • Contributors All contributors are listed as co-authors.

  • Ethics approval The Technion Ethics Committee (IRB) approved this study (decision letter issued on 29 May 2014).

  • Funding Israel National Institute for Health Policy Research.

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.