Multifaceted intervention including Facebook‐groups to improve guideline‐adherence in ICU: A quasi‐experimental interrupted time series study

The impact of social media, with its speed, reach and accessibility, in interventions aimed to improve adherence to guidelines such as assessment of Pain, Agitation/Sedation and Delirium (PAD) in intensive care is not described. Therefore, the primary objective of this quality improvement study was to evaluate the impact of a multifaceted intervention including audit and feedback of quality indicators (QI) via Facebook‐groups, educational events and engagement of opinion leaders on adherence to PAD‐guidelines in four ICUs.


| INTRODUC TI ON
Health status of critically ill patients depends significantly on quality of care in the intensive care unit (ICU). 1,2 Optimisation of provided critical care according to evidence-based guidelines is of utmost importance. 1,2 Quality should be monitored and measured, and action must be taken if quality is found to be suboptimal. 2,3 Current practice for providing information and feedback about quality of care is mainly based on traditional communication methods such as international, national and local meetings, e-mails, web-pages and posters in the ICU. The effects of single components or multifaceted interventions targeting common barriers such as lack of knowledge, awareness or motivation on improved adherence vary. [4][5][6][7][8][9][10] An overview of systematic reviews from 2011 showed that multifaceted interventions are more likely to improve practice than single interventions. 9 We use social media (SoMe) as a daily way of communication. SoMe can improve communication and information sharing, 11 and provide an educational medium for improving health care personnel (HCP) knowledge, research evidence adherence and clinical behaviour. 12 However, use of SoMe in an integrated approach aimed to communicate with HCPs to improve ICU guideline-adherence has not been studied. [13][14][15] SoMe can be an alternative communication method with its speed, reach and accessibility via their smartphones. 12,16,17 In a recent Norwegian survey, 93% of ICU nurses and physicians reported having a SoMe profile, with Facebook being the most popular. 18 In particular, ICU nurses reported a positive attitude towards receiving content on critical care topics in work-related closed Facebook-groups. 18 To our knowledge, no study has tested use of Facebook-groups to improve HCP's adherence to ICU guidelines.
Assessment of guideline-adherence can be measured through quality indicators (QIs). 19 QIs are defined as 'measures to assess a particular health care structure, process or outcome', 1,20,21 and may be used as screening tools to flag potential health care quality problems needing further investigation. 22 Process indicators describe the process of care itself; whether what is known as good clinical practice has been applied. 1,20 ICU staff deal with several care processes, and pain, agitation/sedation and delirium (PAD) are typical examples. 23,24 Routines of systematic assessment of PAD with validated tools are strongly recommended in evidence-based international guidelines. 23,24 In a large study, adherence to a bundle including PAD assessment and management was associated with a clear dose-response relationship between higher bundle-adherence and improved patient outcomes. 25 In addition, significant pain was more frequently reported as bundle performance proportionally increased. 25 HCPs need both knowledge and clinical competency in understanding the complexity of PAD elements and overcome barriers to improve treatment based on PAD assessment. Therefore, a multifaceted approach would be more likely to facilitate adherence to PAD assessment, also considering that people respond differently to varying types of interventions. 9,26 The primary objective of the present quality improvement study was to evaluate the impact of a multifaceted intervention including audit and feedback of QIs via Facebook-groups and email, educational events and engagement of opinion leaders to ensure adherence to the recommended PAD-guidelines. 23 We hypothesised that process PAD-QIs would increase in the intervention period compared to the period before. In addition, we aimed to perform an exploratory process evaluation of the Facebook-intervention.

| ME THODS
This study is part of a larger quality improvement initiative through a multifaceted intervention. In addition to PAD, four other QIs were included: multi-professional ward rounds, early mobilisation, early enteral nutrition and pressure ulcers ( Figure 1, Appendix 1). Only the impact on PAD are included within the scope of this study, because PAD-QIs are applicable to all patients, less influenced by specific diagnoses and circumstances, thereby clearly reflecting the impact of the intervention on adherence.

| Design
A quasi-experimental interrupted time series study with two phases was designed, including eight monthly data points before (January'17-August'17, Before) and after initiation of the multifac-

| Setting and participation
The study was conducted in four ICUs at Oslo University Hospital (OUH). In 2017, OUH had 3390 ICU patient-stays from 11 ICUs registered in the national Norwegian ICU Registry (NIR), of which 1378 (41%) represented the four study ICUs. 27 These ICUs are organised within the same department, with two physically located at OUH

Editorial Comment
Use of social media may play a role in the dissemination and implementation of clinical practice guidelines in the ICU.

| Multifaceted intervention
The multifaceted intervention included educational events, audit and feedback of QIs via Facebook and email, and engagement with OL ( Figure 1). To increase distribution, visibility and interest, they included emojis, questions and a call to action to gain comments and/or 'likes', including offering gifts to one of those who liked/ commented. The last Facebook-post was a poll asking Facebook-members to vote for their future preferred place to receive information on critical care topics. All options known to members were available, and multiple options could be voted for.

Educational events
Involvement of local OL 7 included staff leadership and especially professional development nurses who were involved in planning, patient-inclusion and bedside follow-up. QIs were presented to OLs at two meetings during the intervention period in addition to an ongoing dialog. The importance of their involvement to optimise care was emphasised.

| Data collection
Data from each ICU patient-stay related to PAD-QI calculations, were retrieved retrospectively from the electronical patient chart system (MetaVision, iMDsoft, Israel). To describe included stays, data were retrieved from NIR, including demographic data, primary reason for ICU admission, treatment-interventions, Nursing Activities Scores (NAS), LOS, time on invasive mechanical ventilation and ICU mortality.
We intended to collect data on engagement from the closed Facebook-groups on each Facebook-post 24 h after posting. Data from the poll was summarised when there was no more activity.

| Variables and outcomes
Primary outcome was adherence to PAD guidelines, measured by the level of the three PAD-QIs in Before and Intervention. The PAD-QIs were calculated per ICU patient-stay and defined as number of

| Ethics
Study approval was obtained from the Regional Ethics Committee (2016/2281/REK sør-øst A), and the data protection officer at OUH. Permission was obtained to connect NIR data with data from MetaVision. Data were safely stored on the hospital research server.
All patients included received standard care. Written informed consent was obtained prospectively by the patient or a relative. Consent from retrospectively included patients was achieved by a letter with a request to use a defined set of their ICU data with the possibility to withdraw their study participation.
Permission was obtained from department heads. ICU nurses and physicians were informed during educational sessions and through Facebook in the four closed Facebook-groups.

| Statistical analysis
Categorical data are presented as counts and percentages.
Continuous variables are described with mean and standard deviation (SD) or median and interquartile range (IQR) depending on the distribution. Crude differences between Before and Intervention for continuous variables were assessed by independent samples ttest or non-parametric independent samples Mann-Whitney U test, when appropriate. Pairs of categorical data were compared using Pearson's chi-square test or Fisher's exact test, as appropriate.
For the ITS analysis, time was measured in months. Data from each ICU patient-stay was allocated to the appropriate month based on date of discharge. Data are depicted graphically using estimates of aggregated monthly averages with 95% confidence intervals (CI).
Differences between Before and Intervention were assessed using generalised mixed model for repeated measures with unstructured covariance matrix, and results are presented as estimated means at given time points (separately for each ICU) and overall estimated change (Intervention-Before) quantified as regression coefficient The study is considered exploratory so no correction for multiple testing was performed.

| RE SULTS
Of 1413 eligible ICU patient-stays, 1108 (78%) in 978 patients were included. Finally, 1049 ICU patient-stays were analysed; 534 in Before and 515 in Intervention, after excluding 59 ICU stays overlapping with the two time periods (Figure 2). Details from the two cohorts are shown in Table 2.

| Levels of PAD-QIs
The 16 individual monthly data points included PAD-QIs from 53 to 80 ICU patient-stays. ITS-analyses showed a significant increase in all three PAD-QIs in Intervention versus Before ( Figure 3). All three

| ICU personnel's engagement in Facebookposts
The

| DISCUSS ION
The main finding of this study evaluating the impact of a multifac- To our knowledge, this is the first study to use closed Facebookgroups as part of an intervention strategy to improve adherence to ICU recommendations. The QIs in Before were relatively low for pain (40%) and especially low for delirium (10%), and these increased by 31% and 34%, respectively. To improve an activity that prior to intervention is low is less challenging and not surprising. 4  QIs from the main investigator (AP) to ICU colleagues. In addition, educational events and OLs were also included. In two Cochrane reviews, median-adjusted risk difference in adherence to desired practice was 6% with educational events 5 and 10.8% improvements in adherence to evidence-based practice with OL interventions. 7 The impact of Facebook on the observed changes in PAD-QIs in this study is unclear, and studies of Facebook-use to improve adherence to patient care are lacking. In a study, communicating evidencebased practice points via Facebook and Twitter, 70% of respondents reported that SoMe had changed their practice. 12 We found that a large proportion of Facebook-group members saw the posts indicating that the intervention was adopted. However, we do not know whether 'seen' actually means that they read the content or just

| Strengths and limitations
Randomisation was not possible due to logistical issues. The ability to attribute the change to the intervention is strengthened with ITS including multiple measurements by reduced uncertainty of unstable measurements at only two time-points. [32][33][34] However, we cannot completely exclude that the difference in PAD-QIs could be caused by something else and not the intervention due to the history threat. We are, however, not aware of any structural changes made in Intervention versus Before.
Characteristics of included ICU patient-stays were similar in both cohorts. However, mechanical ventilation including tracheostomies and use of vasoactive infusions were more common, in addition to longer time on mechanical ventilation and higher ICU mortality in Before.
Whether this is relevant to PAD-documentation is unclear.
Effects of the multifaceted intervention were evaluated during an ongoing intervention period, and we do not know if the effect was sustainable or if it just represented a Hawthorne effect. 33,35 At what point the effect on improved adherence should be measured is controversial, due to several confounding aspects developing over time. 34 Implementation and quality improvement initiatives are ongoing processes for several years, and more follow-up is needed after this study's intervention period both to maintain and further improve practice. Typical ITS limitations are autocorrelation and seasonality.
Autocorrelation was adjusted for using appropriate statistical methodology, and seasonal changes are not expected in the PAD-QIs except for weekends and holidays with more use of temporary staff.
Before included summer holidays, with expected lower guideline adherence, but this was not reflected in the ITS figure. Finally, blinding of ICU personnel and study-investigators was impossible.