Compassionate care intervention for hospital nursing teams caring for older people: a pilot cluster randomised controlled trial

Objective Compassionate care continues to be a focus for national and international attention, but the existing evidence base lacks the experimental methodology necessary to guide the selection of effective interventions for practice. This study aimed to evaluate the Creating Learning Environments for Compassionate Care (CLECC) intervention in improving compassionate care. Setting Ward nursing teams (clusters) in two English National Health Service hospitals randomised to intervention (n=4) or control (n=2). Intervention wards comprised two medicines for older people (MOPs) wards and two medical/surgical wards. Control wards were both MOPs. Participants Data collected from 627 patients and 178 staff. Exclusion criteria: reverse barrier nursed, critically ill, palliative or non-English speaking. All other patients and all nursing staff and Health Care Assistant HCAs were invited to participant, agency and bank staff were excluded. Intervention CLECC, a workplace intervention focused on developing sustainable leadership and work-team practices to support the delivery of compassionate care. Control: No educational activity. Primary and secondary outcome measures Primary—Quality of Interaction Schedule (QuIS) for observed staff–patient interactions. Secondary—patient-reported evaluations of emotional care in hospital (PEECH); nurse-reported empathy (Jefferson Scale of Empathy). Results Trial proceeded as per protocol, randomisation was acceptable. Some but not all blinding strategies were successful. QuIS observations achieved 93% recruitment rate with 25% of patient sample cognitively impaired. At follow-up there were more total positive (78% vs 74%) and less total negative (8% vs 11%) QuIS ratings for intervention wards versus control wards. Sixty-three per cent of intervention ward patients scored lowest (ie, more negative) scores on PEECH connection subscale, versus 79% of control. This was not a statistically significant difference. No statistically significant differences in nursing empathy were observed. Conclusions Use of experimental methods is feasible. The use of structured observation of staff–patient interaction quality is a promising outcome measure inclusive of hard to reach groups. Trial registration number ISRCTN16789770.


Introduction
Healthcare systems internationally are challenged by the provision of optimal care to an aging population 1 . Research into outcomes for older people admitted to hospital is far from encouraging with hospitalised older people at significant risk of functional decline 2 and frail older patients at increased risk of mortality and re-admission 3 4 . A recent systematic review on outcomes for older people in acute care suggest there is an "urgent need for the development and evaluation of effective interventions…. that optimise the care outcomes of older patients" 5 . This review found personalised treatment plans, and clear communication strategies can reduce re-admission and mortality 5 . Since person centred care and communication are features of relational care, this study aims to pilot an intervention aimed at improving relational aspects of care, i.e. capacity to experience empathy and engage in a caring relationship.
Research indicates that the quality of relationships with staff is key to shaping older people's hospital experiences, with older people valuing being seen as people, listened to and involved in treatment 6 . All of these values fall under the scope of compassionate care, requiring "relational capacity" in practitioners 7 . Thus research into what older people value in care, and conclusions from English NHS and international reports 1 8-10 suggest that interventions to improve the delivery of compassionate care are needed to meet the care needs of older people in hospitals. Compassionate care intervention for hospital nursing teams: a pilot cluster randomised controlled trial 4 Although it is apparent that interventions to improve the delivery of compassionate care are needed, there is a lack of evidence for their efficacy, with utility limited by a seeming reluctance to use rigorous experimental methods for evaluation. A recent systematic review of evidence for compassionate nursing care interventions found that most of the 24 studies identified used uncontrolled before and after designs, with just four using randomised controlled designs 11 . Studies tended to be single-site and small-scale. A wide range of outcome measures (n=67) were deployed between the studies including staff-based outcomes (e.g. empathy), patient-based outcomes (e.g. mood) and care outcomes (e.g. patient-centeredness), indicating a lack of consensus in the field as to appropriate compassionate care outcomes and how to measure them. While most studies (79%) reported a positive effect in relation to one or more outcomes, higher quality studies were less likely to report positive effects and no interventions were evaluated more than once.
Thus the quality of the evidence for effectiveness in this field is predominantly low, hampered by a lack of experimental research of sufficient scale.
Responding to an absence of high quality evidence for the effectiveness of compassionate care interventions for older patients, the study reported here aimed to pilot the use experimental methodology to evaluate a compassionate care intervention targeted at work teams in acute care settings. We aimed to provide an evidence base to guide future trial design and implementation, including feasibility of ward level randomisation, selection of outcome measures including success in blinding, sample size calculation, minimising contamination between experimental and control clusters, and maximising participation of older patients.

Methods
A multi-site pilot cluster randomised controlled trial (CRT) was undertaken with randomisation of staff and patients at ward nursing team level. Medical and surgical wards with high proportion of older patients were eligible. Six wards in two NHS hospital Trusts in England were enrolled and allocated to intervention (n=4) or control (n=2). Randomisation was undertaken using the ralloc command in Stata 12   Compassionate care intervention for hospital nursing teams: a pilot cluster randomised controlled trial 5 The CLECC intervention is based on workplace learning theory with the ward conceptualised as a learning environment and ward team as a community of practice 13 . It is a team-based educational programme focused on developing manager and team practices that create an expansive learning environment, theorised to enhance team capacity to provide compassionate care 14 . Expansive (rather than restrictive) environments foster workplace learning and the integration of personal and organisational development [15][16][17] . The intervention aims to embed ward-based manager and team practices including dialogue, reflective learning and mutual support, such that the team has the understanding and skills to continue to improve compassionate care following the end of the programmed activities.
Research suggests that embedding such practises leads to a longer-term period of service improvement and sustainable improvements in practice 18 . CLECC training consisted of key activities, such as: monthly ward leader action learning sets; team learning activities, including local team climate analysis and values clarification; peer observations of practice and feedback to team by volunteer team members; team study days focused on team building and understanding patient experiences; mid-shift 5 minute team cluster discussions; and twice weekly team reflective discussions. A Practise Educator, seconded from the hospital, leads these activities. Throughout the 4-month implementation period, ward leaders and their teams developed a team-learning plan that included a patient feedback plan and measures for continuing to support leader and team practices that underpin the delivery of compassionate care. Usual practice continued on control wards.
Outcome measures were assessed at baseline (2 months before intervention and prior to randomisation to groups) and follow-up (4 months after completion of CLECC implementation period). Given anticipated patient and staff turnover between assessment periods, follow-up was at cluster level rather than individual participant level, and so recruitment for baseline and follow-up assessment periods was independent. There is no single validated measure for compassionate care so its impact was assesed across three  Compassionate care intervention for hospital nursing teams: a pilot cluster randomised controlled trial 6 research, that is, older people with complex needs including cognitive impairment and communication difficulties. There is insufficient literature to guide the recruitment of these groups in acute care settings, so it was not possible at the outset to predict sample size, instead more flexible target recruitment rates were used.
The quality of staff-patient interactions was assessed using the Quality of Interactions Schedule (QuIS). QuIS is a time sampling tool that measures the volume and quality of interactions through observation, enabling a calculation of how many patients experience one or more negative interactions during an observation session 19 . QuIS interactions between staff and patients are coded as positive social, positive care, neutral, negative protective or negative restrictive. Earlier piloting work has established its validity and reliability in acute settings 20 .
All adult patients on participating wards were assessed for eligibility to be included in observations. Patients were excluded if they were unable to communicate their choices about taking part in the research and a consultee could not be contacted. We also excluded  Compassionate care intervention for hospital nursing teams: a pilot cluster randomised controlled trial 7 English hospital settings 21 . The subscales are security, knowledge, personal value and connection. PEECH is sensitive to changes in service quality and in ward environment 22 . All eligible patients on the ward were invited to complete a questionnaire. Patients were excluded if there were clinical concerns (as detailed above) or if they lacked capacity to consent. If recruited, patients were offered help by the researcher in completing the questionnaire.
Nurses' self-reported empathy was measured using the Jefferson Scale of Empathy (JSE)(Physician/HP version), a 20 item inventory in a 7-point Likert-type format ranging from Strongly Disagree to Strongly Agree with higher scores reflecting a more empathic orientation 23 . The JSE was developed and validated for use by health care workers, the scale is sensitive to changes in individual empathy over time and context 24 25 . All nursing staff (registered nurses and health care assistants) were invited to complete a questionnaire, based on a staff list supplied by the ward manager. Questionnaires in individually named envelopes were distributed by ward managers and returned via an on ward postbox.
A number of measures were employed to enable allocation concealment and blinding.
Clusters were randomly allocated to group following baseline data collection. At follow-up, researchers conducting observations were blind to conditions. Researchers gathering questionnaire data at follow-up were aware of ward allocation. It was not possible to conceal allocation from ward team nursing staff. Patients were not informed of allocation.
All analyses were carried out on an intention to treat basis. Descriptive statistics were used to show the proportion of participants that consented to participate in study. The proportion of QuIS interactions rated for each of the five categories was analysed and the frequencies of patients with the lowest (most negative) scores for each subscale was calculated. The differences between groups were tested using Chi-square test. A three level mixed-effects logistic regression model was fitted to investigate the effect of the CLECC intervention on the likelihood of a negative interaction. Predictive factors were included as fixed effects and presented as odds ratios (OR) with 95% CI, after adjustment for baseline and ward consecutively. Mean PEECH and JSE scores were calculated by subscale and in total, and differences between groups at follow-up were tested using Mann-Whitney U test. Compassionate care intervention for hospital nursing teams: a pilot cluster randomised controlled trial 8 In order to determine the appropriate approach for analysis and the design effect when calculating the required sample in a definitive trial, estimates of intracluster correlation were generated for each outcome measure.

Results
Six out of seven nursing ward managers invited to take part agreed to randomisation to either intervention or control. Three wards were recruited in each Trust, and all wards remained in the study until it closed, the wards had a range of 28-32 beds and mean patients stays ranged from 6days-19days. Data were collected between March 2015 and March 2016. Procedures for allocation concealment and blinding proceeded as planned, with the exception of two (out of eight) researcher observers at follow-up reporting that they learned of ward allocation from ward staff. No staff audited following observations reported that their behaviour had changed because they were being observed. Researcher field notes reflect reports from hospital managers that discussions about CLECC between staff on intervention and control wards had the potential to influence practice on the control wards, but we did not detect evidence of contamination. Figure 1 shows the flow of clusters and participants through the pilot trial. Randomisation took place after baseline data collection, but results are presented by allocation for baseline and follow-up data to enable comparisons between groups.

FIGURE ONE: CONSORT flow diagram
For staff-patient observations, Figure 1 illustrates the number of approaches rather than individual patients, as some patients were invited more than once to be involved.
Recruitment rate for observations at baseline was 97% (152 out of 157 approaches to eligible patients), and at follow-up was 90% (157 out of 175). Recruitment rates were similar between intervention and control wards (96% versus 98% at baseline, 90% versus 88% at follow-up). Twenty-three participants declined to participation for reasons such as, "not feeling up to it" (17%), "too unwell" (4%) and "no reason" (8%). No specific reason was recorded for 70%. In 17% (63 out of 362 approaches) the patient was approached and then assessed as not having capacity to make the decision to take part in the research. In 67% Compassionate care intervention for hospital nursing teams: a pilot cluster randomised controlled trial 9 (42 out of 63) of these occasions, researchers were able to contact a consultee for advice and in 100% of these cases the consultee advised that the patient should participate. A final 273 patients were observed (133 at baseline and 140 at follow-up). The mean age of patients observed was 82 years (84 years in intervention group and 77 in control). Most patients were female (77%) and 25% had evidence of cognitive impairment, with no significant differences between patients in the different experimental groups. All observation data gathered was included in analysis. Please see table 1 for Patient characteristics. were no notable differences in these characteristics between intervention and control group staff returning questionnaires. All returned questionnaires (91 at baseline and 87 at followup) were included in analyses.

Baseline and outcome measures
As planned, 120 hours of observations took place in each assessment period, resulting in data collected on 3109 interactions between staff and patients over 240 hours. On average, each patient had 6 interactions with hospital staff per hour. Most interactions were rated as positive care (59%) and least interactions as negative protective (0.04%) for each experimental group at both assessment periods (Table 2).

TABLE TWO
At follow-up, there were higher total positive (positive social and positive care) and lower total negative (negative protective and negative restrictive) scores for intervention wards at follow-up than control wards at follow-up (78% versus 74%, 8% versus 12%). Chi square testing of these differences between groups suggested a significant difference (p=0.017).

Article Summary: Strengths and Limitations of this study
• Findings from this pilot trial make an important contribution to the evidence base on the evaluation of compassionate care interventions, particularly their impact on patient-based outcomes. • This study demonstrates that use of experimental method in this field is feasible.
• The study demonstrates where blinding was efficient, and where it was more difficult in a pragmatic hospital based intervention to control. • Only six wards were included in this study, meaning the results may not be generalizable due to the small sample size. • The study is of insufficient scale to draw meaningful conclusions about CLECC's effectiveness. The findings indicate, however, that more definitive evaluation is merited.

Introduction
Healthcare systems internationally are challenged by the provision of optimal care to an aging population. 1 Research into outcomes for older people admitted to hospital is far from Research indicates that the quality of relationships with staff is key to shaping older people's hospital experiences, with older people valuing being seen as people, listened to and involved in treatment. 5 However, evidence from English NHS and international reports 1 6-8 indicates that older people frequently fail to experience positive and caring staff attitudes and behaviours, resulting in a perceived lack of compassion. Compassion is "a deep awareness of the suffering of another coupled with the wish to relieve it". 9 Being compassionate requires "relational capacity" in practitioners, i.e. capacity to experience empathy and to engage in a caring relationship. 10 The apparent need to improve compassionate hospital care for older people has led to the development of a number of interventions, but there is a lack of evidence for their efficacy, with utility limited by a seeming reluctance to use rigorous experimental methods for evaluation. A recent systematic review of evidence for compassionate nursing care interventions found that most of the 24 studies identified used uncontrolled before and after designs, with just four using randomised controlled designs. 11 Studies tended to be single-site and small-scale. A wide range of outcome measures (n=67) were deployed between the studies including staff-based outcomes (e.g. empathy), patient-based outcomes (e.g. mood) and care outcomes (e.g. patient-centeredness), indicating a lack of consensus in the field as to appropriate compassionate care outcomes and how to measure them. While most studies (79%) reported a positive effect in relation to one or more outcomes, higher quality studies were less likely to report positive effects and no interventions were evaluated more than once. Thus the quality of the evidence for effectiveness in this field is predominantly low, hampered by a lack of experimental research of sufficient scale. Responding to an absence of high quality evidence for the effectiveness of compassionate care interventions for older patients, the study reported here aimed to pilot the use experimental methodology to evaluate a compassionate care intervention targeted at work teams in acute care settings. We aimed to provide an evidence base to guide future trial design and implementation, including feasibility of ward level randomisation, selection of outcome measures including success in blinding, sample size calculation, minimising contamination between experimental and control clusters, and maximising participation of older patients.

Methods
As part of a wider feasibility study, a multi-site pilot cluster randomised controlled trial (CRT) was undertaken with randomisation of staff and patients at ward nursing team level. 12 Medical and surgical wards with high proportion of older patients were eligible. Six wards in two NHS hospital Trusts in England were enrolled and allocated to intervention (n=4) or control (n=2). Randomisation was undertaken using the ralloc command in Stata 13 and stratified by hospital and by ward type: Medicine for Older People (MOP) or not MOP.
The CLECC intervention is based on workplace learning theory with the ward conceptualised as a learning environment and ward team as a community of practice. 14 It is an educational programme focused on developing manager and team practices that create an expansive learning environment, theorised to enhance team capacity to provide compassionate care. 15 Expansive (rather than restrictive) environments foster workplace learning and the integration of personal and organisational development. [16][17][18] The intervention aims to embed ward-based manager and team practices including dialogue, reflective learning and mutual support. Research suggests that embedding such practices leads to a longer-term period of service improvement and sustainable improvements in practice. 19  activities through a four-month implementation period, aiming to develop a team-learning plan that included measures for continuing to support leader and team practices that underpin the delivery of compassionate care beyond the initial programmed activities.
Usual practice continued on control wards. Further detail on development of the CLECC intervention can be found in Bridges and Fuller. 20 Outcome measures were assessed at baseline (2 months before intervention and prior to randomisation to groups) and follow-up (4 months after completion of CLECC implementation period). Given anticipated patient and staff turnover between assessment periods, follow-up was at cluster level rather than individual participant level, and so recruitment for baseline and follow-up assessment periods was independent. There is no single validated measure for compassionate care so its impact was assessed across three complementary core outcomes: researcher-rated observations of the quality of staff-patient interactions, patient-reported evaluations of emotional care and nurse-reported measures of empathy. Baseline and follow-up data were also gathered on individual and ward team characteristics including patient age, cognitive impairment, ward leadership and staff turnover. We aimed to maximise the participation of older people with cognitive impairment and communication difficulties through recruitment procedures that optimised capacity to make decisions about taking part in the study. 12 Because there is insufficient literature to guide the recruitment of these groups, it was not possible at the outset to predict sample size. Instead, more flexible target recruitment rates were used.
The quality of staff-patient interactions was assessed using the Quality of Interactions Schedule (QuIS), a time sampling tool that measures the volume and quality of interactions through observation. 21 Staff-patient interactions are rated as positive social, positive care, neutral, negative protective or negative restrictive. Earlier piloting work has established its validity and reliability in acute settings. 22 All adult patients on participating wards were assessed for eligibility to be included in observations. Patients were excluded if they were unable to communicate their choices about taking part in the research and a consultee could not be contacted. We also excluded patients who were unconscious or where there were clinical concerns (critically ill, in receipt  Nurses' self-reported empathy was measured using the Jefferson Scale of Empathy (JSE)(Physician/HP version), a 20 item inventory in a 7-point Likert-type format ranging from Strongly Disagree to Strongly Agree with higher scores reflecting a more empathic orientation 25 . The JSE was developed and validated for use by health care workers, the scale is sensitive to changes in individual empathy over time and context 26 27 . All nursing staff (registered nurses and health care assistants) were invited to complete a questionnaire, based on a staff list supplied by the ward manager. Questionnaires in individually named envelopes were distributed by ward managers and returned via an on-ward postbox.
A number of measures were employed to enable allocation concealment and blinding.
Clusters were randomly allocated to group following baseline data collection. At follow-up, researchers conducting observations were blinded to allocation, but researchers gathering Compassionate care intervention for hospital nursing teams caring for older people: a pilot cluster randomised controlled trial 6 questionnaire data were aware of ward allocation. It was not possible to conceal allocation from ward team nursing staff. Patients were not informed of allocation.
All analyses were carried out on an intention to treat basis. Descriptive statistics were used to show the proportion of participants that consented to participate in study. The proportion of QuIS interactions rated for each of the five categories was analysed and the frequencies of patients with the lowest (most negative) scores for each subscale was calculated.
The differences between groups were tested using Chi-square test. A three level mixed-effects logistic regression model was fitted to investigate the effect of the CLECC intervention on the likelihood of a negative interaction. Predictive factors were included as fixed effects and presented as odds ratios (OR) with 95% CI, after adjustment for baseline and ward consecutively. Mean PEECH and JSE scores were calculated by subscale and in total, and differences between groups at follow-up were tested using Mann-Whitney U test.
In order to determine the appropriate approach for analysis and the design effect when calculating the required sample in a definitive trial, estimates of intracluster correlation were generated for each outcome measure.

Results
Six out of seven nursing ward managers invited to take part agreed to randomisation to either intervention or control. Three wards were recruited in each Trust, and all wards remained in the study until it closed, the wards had a range of 28-32 beds and mean patients stays ranged from 6days-19days. Data were collected between March 2015 and March 2016. Procedures for allocation concealment and blinding proceeded as planned, with the exception of two (out of eight) researcher observers at follow-up reporting that they learned of ward allocation from ward staff. No staff audited following observations reported that their behaviour had changed because they were being observed. Researcher field notes reflect reports from hospital managers that discussions about CLECC between staff on intervention and control wards had the potential to influence practice on the control wards, but we did not detect evidence of contamination.  Figure 1 shows the flow of clusters and participants through the pilot trial. Randomisation took place after baseline data collection, but results are presented by allocation for baseline and follow-up data to enable comparisons between groups.

FIGURE ONE: CONSORT flow diagram
For staff-patient observations, Figure 1 illustrates the number of approaches rather than individual patients, as some patients were invited more than once to be involved.  (missing data=6%). There were no notable differences in job role by experimental group. All returned questionnaires (91 at baseline and 87 at follow-up) were included in analyses.

Baseline and outcome measures
As planned, 120 hours of observations took place in each assessment period, resulting in data collected on 3109 interactions between staff and patients over 240 hours. On average, each patient had 6 interactions with hospital staff per hour. Most interactions were rated as positive care (59%) and least interactions as negative protective (4%) for each experimental group at both assessment periods (  At follow-up, there were higher total positive (positive social and positive care) and lower total negative (negative protective and negative restrictive) scores for intervention wards than control (78% versus 74%, 8% versus 12%). Chi square testing suggested these differences were significant (p=0.017). However, multilevel logistic regression results  Table 3 QuIS multilevel logistic regression results: odds ratios (OR) of a negative interaction Table 4 shows the mean patient evaluations of emotional care (PEECH) values by experimental group. Higher scores indicate better patient-reported experiences.
Connection subscale scores were consistently lower than on other subscales. Differences between groups at follow-up favor CLECC in total score and three of the four subscales, but these differences were not significant. Hospitalised older patients with cognitive impairment are a traditionally hard-to-reach group and even though they appear more prone to negative experiences of hospital care, 29 they are often excluded from research. 5 30 31 It is estimated that up to 25% of beds in acute hospitals are occupied by people with dementia, with the figure likely to be higher on specialist older people's wards. 32 33 While cognitive deficits may limit some people's ability to share their experiences, our study has been successful in devising recruitment and data collection methods that maximise their inclusion. Overall 25% of patients observed in this study had evidence of cognitive impairment, suggesting a sample representative of the wider hospital population. Twelve percent of patient questionnaires returned were completed by patients with cognitive impairment, indicating the questionnaire method was inclusive than observation methods. Participating in an observation does not require any particular state of health, abilities or performance form the patient in question, whereas participating in a questionnaire about one's care experiences requires a minimum orientation to place, language skills and attention. 31 In addition, using questionnaire methods may be psychologically threatening to patients still in receipt of care, regardless of cognitive status. 34 The validity of observer ratings as accurate representation of patient experiences merits attention. Because main study observation and questionnaire data were gathered from different patient groups, it was not possible to test the validity of observer ratings against patient-reported experience. However, in earlier piloting work we found 79% agreement (weighted kappa 0.40: P < 0.001; indicating fair agreement) between patients' and  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60 F o r p e e r r e v i e w o n l y For each group, the numbers of participants who were randomly assigned, received intended treatment, and were analysed for the primary outcome For each group, the numbers of participants who were approached and/or assessed for eligibility, randomly assigned, received intended treatment, and were assessed for each objective

Introduction
Healthcare systems internationally are challenged by the provision of optimal care to an aging population. 1 Research into outcomes for older people admitted to hospital is far from encouraging with hospitalised older people at significant risk of functional decline 2 and frail older patients at increased risk of mortality and re-admission. 3 A recent systematic review on outcomes for older people in acute care suggest there is an "urgent need for the development and evaluation of effective interventions…. that optimise the care outcomes of older patients". 4 This review found personalised treatment plans, and clear communication strategies can reduce re-admission and mortality. 4 This study aims to pilot an intervention aimed at improving compassionate hospital care for older people.
Research indicates that the quality of relationships with staff is key to shaping older people's hospital experiences, with older people valuing being seen as people, listened to and involved in treatment. 5 However, evidence from English NHS and international reports 1 6-8 indicates that older people frequently fail to experience positive and caring staff attitudes and behaviours, resulting in a perceived lack of compassion. Expressed simply, compassion is "a deep awareness of the suffering of another coupled with the wish to relieve it". 9 There are four key components to the narrative of nursing compassion. 10 The first focuses on ideas about the moral attributes of a 'compassionate' nurse, including wisdom, humanity, love, and empathy. These moral attributes are expressed through a kind of situational awareness in which vulnerability and suffering are perceived and acknowledged. These perceptions underpin participation of the nurse in responsive action that is aimed at relieving suffering and ensuring dignity, and which involves the nurse in a participatory relationship in which the nurse exercises relational capacity through which empathy is experienced and a caring pastoral relationship is constructed. 10,11 The apparent need to improve compassionate hospital care for older people has led to the development of a number of interventions, but there is a lack of evidence for their efficacy, with utility limited by a seeming reluctance to use rigorous experimental methods for evaluation. A recent systematic review of evidence for compassionate nursing care interventions found that most of the 24 studies identified used uncontrolled before and after designs, with just four using randomised controlled designs. 10 Studies tended to be single-site and small-scale. A wide range of outcome measures were deployed between the studies including staff-based outcomes (e.g. empathy), patient-based outcomes (e.g. mood) and care outcomes (e.g. patient-centeredness), indicating a lack of consensus in the field as to appropriate compassionate care outcomes and how to measure them. While most studies (79%) reported a positive effect in relation to one or more outcomes, higher quality studies were less likely to report positive effects and no interventions were evaluated more than once. Thus the quality of the evidence for effectiveness in this field is predominantly low, hampered by a lack of experimental research of sufficient scale.
Responding to an absence of high quality evidence for the effectiveness of compassionate care interventions for older patients, the study reported here aimed to pilot the use experimental methodology to evaluate a compassionate care intervention targeted at work teams in acute care settings. We aimed to provide an evidence base to guide future trial design and implementation, including feasibility of ward level randomisation, selection of outcome measures including success in blinding, sample size calculation, minimising contamination between experimental and control clusters, and maximising participation of older patients.

Methods
As part of a wider feasibility study, a multi-site pilot cluster randomised controlled trial (CRT) was undertaken with randomisation of staff and patients at ward nursing team level. 12 Medical and surgical wards with high proportion of older patients were eligible. Six wards in two NHS hospital Trusts in England were enrolled and allocated to intervention (n=4) or control (n=2). The number of clusters was determined by funding availability and the plan to run the study in at least two hospital organisations, and at least two ward specialties.
Randomisation of clusters was undertaken using the ralloc command in Stata 13 by the team statistician (IM) blinded to hospital and ward information other than ward specialty. The CLECC intervention is based on workplace learning theory with the ward conceptualised as a learning environment and ward team as a community of practice. 14 It is an educational programme focused on developing manager and team practices at a group level that create an expansive learning environment, theorised to enhance team capacity to provide compassionate care. 15 Expansive (rather than restrictive) environments foster workplace learning and the integration of personal and organisational development. [16][17][18] The intervention aims to embed ward-based manager and team practices including dialogue, reflective learning and mutual support. Research suggests that embedding such practices leads to a longer-term period of service improvement and sustainable improvements in practice. 19  Outcome measures were assessed at baseline (2 months before intervention and prior to randomisation to groups) and follow-up (4 months after completion of CLECC implementation period). Given anticipated patient and staff turnover between assessment periods, follow-up was at cluster level rather than individual participant level, and so recruitment for baseline and follow-up assessment periods was independent. There is no single validated measure for compassionate care, the systematic review cited above identifying 18 different types of outcome measure (a total of 67 individual outcome measures) for compassionate nursing care. 10 The most commonly used nurse-based  were audited to ensure that allocation determined by randomisation was implemented in  23 Designed to measure patient views on the nature of interpersonal interactions with hospital staff and patient-reported assessment of the extent to which therapeutic emotional care has occurred, the subscales are security, knowledge, personal value and connection. PEECH is sensitive to changes in service quality and in ward environment. 24 All eligible patients on the ward were invited to complete a questionnaire.

Randomisation was stratified by hospital and by ward type: Medicine for Older People
Patients were excluded if there were clinical concerns or if they lacked capacity to consent.
If recruited, patients were offered help by the researcher in completing the questionnaire.
Nurses' self-reported empathy was measured using the Jefferson Scale of Empathy (JSE)(Physician/HP version), a 20 item inventory in a 7-point Likert-type format ranging from Strongly Disagree to Strongly Agree with higher scores reflecting a more empathic orientation. 25 The JSE was developed and validated for use by health care workers, the scale is sensitive to changes in individual empathy over time and context. 26 27 All nursing staff (registered nurses and health care assistants) were invited to complete a questionnaire, based on a staff list supplied by the ward manager. Questionnaires in individually named envelopes were distributed by ward managers and returned via an on-ward postbox.
A number of measures were employed to enable allocation concealment and blinding.
Clusters were randomly allocated to group following baseline data collection. At follow-up, researchers conducting observations were blinded to allocation, but researchers gathering questionnaire data were aware of ward allocation. It was not possible to conceal allocation from ward team nursing staff. Patients were not informed of allocation. All analyses were carried out on an intention to treat basis. Descriptive statistics were used to show the proportion of participants that consented to participate in study. The proportion of QuIS interactions rated for each of the five categories was analysed and the frequencies of patients with the lowest (most negative) scores for each subscale was calculated. The differences between groups were tested using Chi-square test. A three level mixed-effects logistic regression model was fitted to investigate the effect of the CLECC intervention on the likelihood of a negative interaction. Predictive factors were included as fixed effects and presented as odds ratios (OR) with 95% CI, after adjustment for baseline and ward consecutively. Mean PEECH and JSE scores were calculated by subscale and in total, and differences between groups at follow-up were tested using Mann-Whitney U test.
In order to determine the appropriate approach for analysis and the design effect when calculating the required sample in a definitive trial, estimates of intracluster correlation were generated for each outcome measure.

A small patient and public involvement (PPI) group and PPI representatives on the Steering
Group oversaw and advised on intervention development, study design, selection of outcome measures and research team training.

Results
Six out of seven nursing ward managers invited to take part agreed to randomisation to either intervention or control. Three wards were recruited in each Trust, and all wards remained in the study until it closed. The wards had between 28 and 32 beds and mean patients stays ranged from six days to 19 days. Data were collected between March 2015 and March 2016. Procedures for allocation concealment and blinding proceeded as planned, with the exception of two researcher observers at follow-up reporting that they learned of ward allocation from ward staff. No staff audited following observations reported that their behaviour had changed because they were being observed. Researcher field notes reflect reports from hospital managers that discussions about CLECC between staff on intervention and control wards had the potential to influence practice on the control wards, but we did not detect evidence of contamination.

Participant flow
Figure One shows the flow of clusters and participants through the pilot trial.
Randomisation took place after baseline data collection, but results are presented by allocation for baseline and follow-up data to enable comparisons between groups.

FIGURE ONE: CONSORT flow diagram
For staff-patient observations, Figure 1 illustrates the number of approaches rather than individual patients, as some patients were invited more than once to be involved. patients were "not having capacity" (43%, n=48) and "very cognitively impaired" (29%, n=32). Ninety nine percent (354 of 359) of patients who consented returned a completed

Baseline and outcome measures
As planned, 120 hours of observations took place in each assessment period, resulting in data collected on 3109 interactions between staff and patients over 240 hours. On average, each patient had 6 interactions with hospital staff per hour. Most interactions were rated as positive care (59%) and least interactions as negative protective (4%) for each experimental group at both assessment periods (   Table 3 QuIS multilevel logistic regression results: odds ratios (OR) of a negative interaction At ward level, intracluster correlations (ICCs) for QuIS, PEECH and JSE were low (<0.027).
The ICC for QuIS at ward level was higher, although still small (0.071), but high at observation session level (0.411).

Discussion:
This study aimed to deliver a compassionate care intervention in acute care settings, pilot the use of experimental methodology and assess the performance of selected outcome Hospitalised older patients with cognitive impairment are a traditionally hard-to-reach group and even though they appear more prone to negative experiences of hospital care, 28 they are often excluded from research. 5 29 30 It is estimated that up to 25% of beds in acute hospitals are occupied by people with dementia, with the figure likely to be higher on specialist older people's wards. 31  whereas participating in a questionnaire about one's care experiences requires a minimum orientation to place, language skills and attention. 30 In addition, using questionnaire The validity of observer ratings as accurate representation of patient experiences merits attention. Because main study observation and questionnaire data were gathered from different patient groups, it was not possible to test the validity of observer ratings against patient-reported experience. However, in earlier piloting work we found 79% agreement (weighted kappa 0.40: P < 0.001; indicating fair agreement) between patients' and observers' ratings of interaction quality. 22 Our earlier work did not include people with a cognitive impairment and validation of QuIS ratings with this patient group may be a necessary next step in the tool's development. In addition, if the proportion of negative interactions is the primary outcome measure in a future study, understanding which interactions are rated by observers (and, where possible, patients) as negative, and why, is an important next step, as is working with patient representatives to establish their views on the size of a meaningful reduction in negative interactions. Further study can also be used to develop more effective procedures to blind observers from experimental allocation in advance of an experimental study. In addition, the high intrucluster correlation we found at an observation session level merits the exploration of the cause of this variance and the feasibility of different approaches to data collection that reduce its impact, for instance, shorter observation sessions. Our findings echo those of Goldberg et al. 29 that structured non-participant observation appears to be the most promising method to describe the experiences of older people with cognitive impairment in the general hospital setting, and so further evaluation and testing of QuIS across these parameters would be a valuable foundation to its further use as an outcome measure in acute settings. 29 While the response rate to patient questionnaires was good (77%), of all the patient questionnaires returned, just 12% were completed by patients with cognitive impairment.
While questionnaires provide an opportunity for patient to directly rate their care, less successful recruitment of a group known to be vulnerable to more negative experiences in hospital, means that any results may not be a valid representation of this group's experiences. The response rate to nursing questionnaires was low (36%), with some larger scale studies showing response rates of European nurses to be 62%, and US nurses to be  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y Compassionate care intervention for hospital nursing teams caring for older people: a pilot cluster randomised controlled trial 15 around 39% . 34 Improving staff survey response rates through further feasibility work would improve confidence that conclusions in empathy levels across staff groups can be drawn with more confidence.
This study was piloted on a small number of wards in two hospitals so the findings are not generalisable. In addition, being observed could, in itself, change staff behaviours, and a common limitation of trials of this kind when it is not possible to conceal allocation from staff, is that bias may influence staff responses to observations and questionnaires.
Additionally the finding of possible contamination between wards means that intervention and control conditions should not run in the same organisation over the same time period.

Pg 2 and 3, and 4
Allocation concealment mechanism: 9 Mechanism used to implement the random allocation sequence (such as sequentially numbered containers), describing any steps taken to conceal the sequence until interventions were assigned Pg 2 and 3,4 and 5 Implementation: 10 Who generated the random allocation sequence, enrolled participants, and assigned participants to interventions Pg 2, 3, 4 and 5 For each group, the numbers of participants who were randomly assigned, received intended treatment, and were analysed for the primary outcome For each group, the numbers of participants who were approached and/or assessed for eligibility, randomly assigned, received intended treatment, and were assessed for each objective  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47 o n l y Numbers analysed:

16
For each group, number of participants (denominator) included in each analysis and whether the analysis was by original assigned groups For each objective, number of participants (denominator) included in each analysis. If relevant, these numbers should be by randomised group Figure 1, pg 6, 7 and 8 Outcomes and estimation: 17a For each primary and secondary outcome, results for each group, and the estimated effect size and its precision (such as 95% confidence interval) For each objective, results including expressions of uncertainty (such as 95% confidence interval) for any estimates. If relevant, these results should be by randomised group Pg 8 and 9. Tables 2, 3 and 4 17b For binary outcomes, presentation of both absolute and relative effect sizes is recommended Not applicable N/A Ancillary analyses: Research indicates that the quality of relationships with staff is key to shaping older people's hospital experiences, with older people valuing being seen as people, listened to and involved in treatment. 5 However, evidence from English NHS and international reports 1 6-8 indicates that older people frequently fail to experience positive and caring staff attitudes and behaviours, resulting in a perceived lack of compassion. Expressed simply, compassion is "a deep awareness of the suffering of another coupled with the wish to relieve it". 9 There are four key components to the narrative of nursing compassion. 10 The first focuses on ideas about the moral attributes of a 'compassionate' nurse, including wisdom, humanity, love, and empathy. These moral attributes are expressed through a kind of situational awareness in which vulnerability and suffering are perceived and acknowledged. These perceptions underpin participation of the nurse in responsive action that is aimed at relieving suffering and ensuring dignity, and which involves the nurse in a participatory relationship in which the nurse exercises relational capacity through which empathy is experienced and a caring pastoral relationship is constructed. 10,11 The apparent need to improve compassionate hospital care for older people has led to the development of a number of interventions, but there is a lack of evidence for their efficacy, with utility limited by a seeming reluctance to use rigorous experimental methods for evaluation. A recent systematic review of evidence for compassionate nursing care interventions found that most of the 24 studies identified used uncontrolled before and and care outcomes (e.g. patient-centeredness), indicating a lack of consensus in the field as to appropriate compassionate care outcomes and how to measure them. While most studies (79%) reported a positive effect in relation to one or more outcomes, higher quality studies were less likely to report positive effects and no interventions were evaluated more than once. Thus the quality of the evidence for effectiveness in this field is predominantly low, hampered by a lack of experimental research of sufficient scale.
Responding to an absence of high quality evidence for the effectiveness of compassionate care interventions for older patients, the study reported here aimed to pilot the use experimental methodology to evaluate a compassionate care intervention targeted at work teams in acute care settings. We aimed to provide an evidence base to guide future trial design and implementation, including feasibility of ward level randomisation, selection of outcome measures including success in blinding, sample size calculation, minimising contamination between experimental and control clusters, and maximising participation of older patients.

Methods
As part of a wider feasibility study, a multi-site pilot cluster randomised controlled trial (CRT) was undertaken with randomisation of staff and patients at ward nursing team level. 12 Medical and surgical wards with high proportion of older patients were eligible. Six wards in two NHS hospital Trusts in England were enrolled and allocated to intervention (n=4) or control (n=2). The number of clusters was determined by funding availability and the plan to run the study in at least two hospital organisations, and at least two ward specialties.
Randomisation of clusters was undertaken using the ralloc command in Stata 13 by the team statistician (IM) blinded to hospital and ward information other than ward specialty.
Randomisation was stratified by hospital and by ward type: Medicine for Older People (MOP) or not MOP. The allocation was communicated to the chief investigator (JB) who oversaw its implementation in practice. The CLECC intervention is based on workplace learning theory with the ward conceptualised as a learning environment and ward team as a community of practice. 14 It is an educational programme focused on developing manager and team practices at a group level that create an expansive learning environment, theorised to enhance team capacity to provide compassionate care. 15 Expansive (rather than restrictive) environments foster workplace learning and the integration of personal and organisational development. [16][17][18] The intervention aims to embed ward-based manager and team practices including dialogue, reflective learning and mutual support. Research suggests that embedding such practices leads to a longer-term period of service improvement and sustainable improvements in practice. 19   All interactions between patients and staff were directly observed by a single researcher for two hours and coded (there were 10 x 2 hour observation sessions per ward per 3 week assessment period). Observation sessions were randomly sampled over three weeks from  23 Designed to measure patient views on the nature of interpersonal interactions with hospital staff and patient-reported assessment of the extent to which therapeutic emotional care has occurred, the subscales are security, knowledge, personal value and connection. PEECH is sensitive to changes in service quality and in ward environment. 24 All eligible patients on the ward were invited to complete a questionnaire.
Patients were excluded if there were clinical concerns or if they lacked capacity to consent.
If recruited, patients were offered help by the researcher in completing the questionnaire.
Nurses' self-reported empathy was measured using the Jefferson Scale of Empathy (JSE)(Physician/HP version), a 20 item inventory in a 7-point Likert-type format ranging from Strongly Disagree to Strongly Agree with higher scores reflecting a more empathic orientation. 25 The JSE was developed and validated for use by health care workers, the scale is sensitive to changes in individual empathy over time and context. 26 27 All nursing staff (registered nurses and health care assistants) were invited to complete a questionnaire, based on a staff list supplied by the ward manager. Questionnaires in individually named envelopes were distributed by ward managers and returned via an on-ward postbox.
A number of measures were employed to enable allocation concealment and blinding.
Clusters were randomly allocated to group following baseline data collection. At follow-up, researchers conducting observations were blinded to allocation, but researchers gathering questionnaire data were aware of ward allocation. It was not possible to conceal allocation from ward team nursing staff. Patients were not informed of allocation.
In order to determine the appropriate approach for analysis and the design effect when calculating the required sample in a definitive trial, estimates of intracluster correlation were generated for each outcome measure.
A small patient and public involvement (PPI) group and PPI representatives on the Steering Group oversaw and advised on intervention development, study design, selection of outcome measures and research team training.

Results
Six out of seven nursing ward managers invited to take part agreed to randomisation to either intervention or control. Three wards were recruited in each Trust, and all wards remained in the study until it closed. The wards had between 28 and 32 beds and mean patients stays ranged from six days to 19 days. Data were collected between March 2015 and March 2016. Procedures for allocation concealment and blinding proceeded as planned, with the exception of two researcher observers at follow-up reporting that they learned of ward allocation from ward staff. No staff audited following observations reported that their behaviour had changed because they were being observed. Researcher field notes reflect reports from hospital managers that discussions about CLECC between staff on intervention and control wards had the potential to influence practice on the control wards, but we did not detect evidence of contamination. Figure One shows the flow of clusters and participants through the pilot trial.

FIGURE ONE: CONSORT flow diagram
Connection subscale scores were consistently lower than on other subscales. Differences between groups at follow-up favor CLECC in total score and three of the four subscales, but these differences were not significant.
Levels of staff self-reported empathy using Jefferson Scale of Empathy varied across individual wards at baseline and at follow-up. There was no significant difference between groups (p=0.800).
The ICC for QuIS at ward level was higher, although still small (0.071), but high at observation session level (0.411).  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   14 Hospitalised older patients with cognitive impairment are a traditionally hard-to-reach group and even though they appear more prone to negative experiences of hospital care, 28 they are often excluded from research. 5 29 30 It is estimated that up to 25% of beds in acute hospitals are occupied by people with dementia, with the figure likely to be higher on specialist older people's wards. 31  whereas participating in a questionnaire about one's care experiences requires a minimum orientation to place, language skills and attention. 30 In addition, using questionnaire methods may be psychologically threatening to patients still in receipt of care, regardless of cognitive status. 33 The validity of observer ratings as accurate representation of patient experiences merits attention. Because main study observation and questionnaire data were gathered from different patient groups, it was not possible to test the validity of observer ratings against patient-reported experience. However, in earlier piloting work we found 79% agreement (weighted kappa 0.40: P < 0.001; indicating fair agreement) between patients' and observers' ratings of interaction quality. 22 Our earlier work did not include people with a cognitive impairment and validation of QuIS ratings with this patient group may be a necessary next step in the tool's development. In addition, if the proportion of negative interactions is the primary outcome measure in a future study, understanding which interactions are rated by observers (and, where possible, patients) as negative, and why, is an important next step, as is working with patient representatives to establish their views on the size of a meaningful reduction in negative interactions. Further study can also be used to develop more effective procedures to blind observers from experimental allocation in advance of an experimental study. In addition, the high intrucluster correlation we found at an observation session level merits the exploration of the cause of this variance and the feasibility of different approaches to data collection that reduce its impact, for instance, shorter observation sessions. Our findings echo those of Goldberg et al. 29 that structured non-participant observation appears to be the most promising method to describe the experiences of older people with cognitive impairment in the general hospital setting, and so further evaluation and testing of QuIS across these parameters would be a valuable foundation to its further use as an outcome measure in acute settings. 29 While the response rate to patient questionnaires was good (77%), of all the patient questionnaires returned, just 12% were completed by patients with cognitive impairment.
While questionnaires provide an opportunity for patient to directly rate their care, less successful recruitment of a group known to be vulnerable to more negative experiences in hospital, means that any results may not be a valid representation of this group's experiences. The response rate to nursing questionnaires was low (36%), with some larger scale studies showing response rates of European nurses to be 62%, and US nurses to be around 39% . 34 Improving staff survey response rates through further feasibility work would improve confidence that conclusions in empathy levels across staff groups can be drawn with more confidence.
This study was piloted on a small number of wards in two hospitals so the findings are not generalisable. In addition, being observed could, in itself, change staff behaviours, and a common limitation of trials of this kind when it is not possible to conceal allocation from staff, is that bias may influence staff responses to observations and questionnaires.
Additionally the finding of possible contamination between wards means that intervention and control conditions should not run in the same organisation over the same time period.

16
For each group, number of participants (denominator) included in each analysis and whether the analysis was by original assigned groups For each objective, number of participants (denominator) included in each analysis. If relevant, these numbers should be by randomised group Figure 1, pg 6, 7 and 8 Outcomes and estimation: 17a For each primary and secondary outcome, results for each group, and the estimated effect size and its precision (such as 95% confidence interval) For each objective, results including expressions of uncertainty (such as 95% confidence interval) for any estimates. If relevant, these results should be by randomised group Pg 8 and 9. Tables 2, 3 and 4 17b For binary outcomes, presentation of both absolute and relative effect sizes is recommended Not applicable N/A Ancillary analyses:  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47