Does Restraint Use Depend on the Hospital? A Multilevel Analysis of Multicentre Prevalence Measurements

In restraint use in the somatic acute-care hospital setting, routine and institutional culture seem to play an important role. This implies that similar patient situations would be managed with restraints in one hospital, while in another hospital the situation would be managed without restraints. This practice variation appears to be ethically and legally questionable. The inuence of organisation-specic factors such as the availability of guidelines is discussed. However, the relevance of such factors at the hospital level has been rarely investigated to date. Therefore, the aims of this study were a) to determine how much variance in restraint use can be explained on the hospital level and b) to examine the impact of organisational factors on restraint use.


Abstract Background
In restraint use in the somatic acute-care hospital setting, routine and institutional culture seem to play an important role.
This implies that similar patient situations would be managed with restraints in one hospital, while in another hospital the situation would be managed without restraints. This practice variation appears to be ethically and legally questionable. The in uence of organisation-speci c factors such as the availability of guidelines is discussed. However, the relevance of such factors at the hospital level has been rarely investigated to date. Therefore, the aims of this study were a) to determine how much variance in restraint use can be explained on the hospital level and b) to examine the impact of organisational factors on restraint use.

Methods
A secondary data analysis of cross-sectional multicentre data was performed. Data were collected during three quality measurements (2016)(2017)(2018) in acute-care hospitals in Switzerland and Austria. Hospitalised patients from different medical specialties aged 18 + with informed consent were included. Descriptive analysis and multilevel logistic regression analysis were performed.

Results
The study included 29,477 patients from a total of 140 hospitals. The 30-day prevalence rate of patients with at least one restraint was 8.7% (n = 2,577). The availability of guidelines regarding restraint use and refresher courses for nursing staff were associated with less restraint use (odds ratios = 0.60 and 0.76). By adding the hospital as a random effect, the explained variance of the model increased from 24-55%.

Conclusions
The use of restraints varies widely between hospitals, even considering patient characteristics. The identi cation of situations in which restraints were used out of routine or institutional culture appears to be an important approach in restraint reduction. Investments in appropriate structures and employee knowledge can facilitate providing restraint-free care as much as possible.
To date, quality improvement initiatives regarding restraint use are mainly known in the long-term care and mental health setting (10,11). Nevertheless, restraints are frequently used in the somatic acute care hospital setting (henceforth referred to as 'hospital') as well. Prevalence rates up to 100% are reported (1,12,13). Large differences in restraint prevalence rates can be detected depending on the ward type studied (intensive care units often have a much higher prevalence rate) and by the de nition of restraints used (e.g. only restraint belts; alternatively, bed rails and electronic monitoring can also be considered as restraints).
Surprisingly, the practice variation in restraint use among hospitals and the impact of organisational factors has rarely been investigated to date. Nevertheless, in order to promote a professional management of restraints and, thus, to develop and implement effective measures for restraint reduction, it is crucial to know the in uencing factors on different levels and their impact on the use and non-use of restraints. Therefore, the aims of this study were a) to determine how much variance in restraint use can be explained on the hospital level (hospital effect) and b) to examine the impact of organisational factors on restraint use; both aspects considered the in uence of patient characteristics on restraint use.

Study design and setting
A secondary data analysis of cross-sectional multicentre studies was performed. Data were collected within the International Prevalence Measurement of Quality of Care, called LPZ (Landelijke Prevalentiemeting Zorgkwaliteit) International (25,26). LPZ International performs an annual international quality measurement for a variety of care indicators (like pressure ulcers, falls and restraints) in various settings and countries. Healthcare institutions are invited annually by a national coordinator in several countries to participate on a voluntary basis in the measurement. For the present study, data from the hospital setting of Switzerland and Austria from three one-day measurement points in the years 2016 to 2018 were included. Other countries in the LPZ consortium were not able to provide data as very few hospitals measured restraint use.

Sample
Hospitalised patients from different medical specialties (ward types) aged 18 + with informed verbal (Switzerland) or written (Austria) consent were included. Patients were excluded if they were not available on the ward during the measurement (e.g. since they were undergoing surgery) or could not give informed consent (e.g. due to cognitive impairment or language barriers) and where no legal representative was available.

Instrument and data collection
For data collection, the LPZ 2.0 instrument was used. It is the 2016 revised version of the LPZ instrument (25). With LPZ 2.0, general and care indicator speci c information is assessed on the institutional, ward and patient levels. For this secondary data analysis, information regarding restraints of different levels was included (for details, see Table 1). Restraints were de ned as 'interventions that may infringe [on] a person's human rights and freedom of movement, including observation, seclusion, manual restraint, mechanical restraint and rapid tranquillisation' (27). Within LPZ 2.0, the data collection process is highly standardised. The whole process (e.g. recruitment and information of patients, preparing data collection including documentation of restraint use 30 days prior to data collection) and all questions and answer options are internationally de ned and described in a measurement manual. Additionally, the questionnaire was conceived as an online data entry program leading the questionnaire completion. To ensure uniform execution of the measurement and uniform answering of the questions, data collectors were trained. Using the train-thetrainer procedure, the national coordinator trained the responsible person within each hospital (called the institutional coordinator). The institutional coordinator then trained the data collectors (registered nurses) within the hospital.
Additionally, the measurement manual with all the information was made available for the data collectors directly in the data entry program.
On the predetermined measurement day, the patient level data were collected by the trained data collectors on-site at the patient's bedside and/or through patient documentation (retrospective assessment). The questions on the institutional and ward level were answered by the institutional coordinator.

Statistical analysis
The data from the different measurement points and the two countries (Switzerland and Austria) were pooled into one dataset. Descriptive statistics (numbers, percentages, 95% con dence intervals [CI], median, interquartile range [IQR]) were used to describe the organisational factors, the sample and the restraint prevalence rate.
A multilevel modelling approach was used in order to determine how much variance in restraint use can be explained on the hospital level (hospital effect). This means that the analysis took into account that patients are clustered in hospitals with their organisational factors. The baseline (before variable selection) of the multilevel logistic regression model was built as shown in Fig. 1.
For variable selection, we used the Akaike information criterion (AIC) (30) backwards procedure implemented in the R package MASS (31). Here, however, the hospital random effect had to be treated as a xed effect. During development of the analysis, we also considered using similar variable selection procedures for logistic multilevel models, but the few software implementations we found were not practicable for our problem. Since the hospital effect is an explicit part of the question, the AIC procedure was employed such that the hospital variable cannot be unselected. Further, to enhance the stability of the variable selection, i.e. to reduce the number of noisy variables selected due to the large sample size, we used a split-half approach where the AIC procedure was applied on both of two subsets from a random split of the data, and then only the variables included in both selections were used for the nal multilevel model. The model then was built as a generalised linear mixed model t by maximum likelihood (Laplace approximation) implemented in the R package Ime4 (32). Additionally, the ICC (intraclass correlation coe cient) was estimated, and a log-likelihood ratio test was performed to evaluate the relevance of the random effect.
Two ICD-10 diagnosis groups (congenital malformations, deformations and chromosomal abnormalities; certain conditions originating in the perinatal period) and the answer option unknown/no diagnosis were present in less than 1% of patients and would have led to convergence problems of the regression model. Therefore, these variables had to be excluded. For similar reasons, the variable age had to be centred. Multicollinearity was tested using the variance in ation factor (VIF).
There were no missing data as the online data collection program only allowed for nishing the survey when all questions were answered.
The statistical analysis was conducted utilising R Version 4.0.1 (33)

Results
The study included 29,477 patients from a total of 140 hospitals (  Only considering the xed effects (patient characteristics and organisational factors), the model could explain 24% of the variance in restraint use (marginal R 2 = 0.24). By adding the random effect (hospital as cluster variable), the model explains 55% of the variance in restraint use (conditional R 2 = 0.55). The log-likelihood ratio test was statistically signi cant (p-value < 0.000), indicating that adding hospital as a random effect (cluster) does improve the model. Additionally, the ICC (0. 41) shows that the random effect is also relevant from a clinical point of view. This means that a relevant part of the variance in restraint use can be explained at the hospital level.

Discussion
In this secondary data analysis of cross-sectional data on restraint use in Swiss and Austrian hospitals, we analysed the impact of organisational factors on the use of restraints in the somatic acute care hospital setting, as well as the practice variation among hospitals. Overall, the restraint prevalence rate was 8.7%. We found that the availability of guidelines regarding restraint use on the institutional level and refresher courses for at least 80% of ward nursing staff in the last two years are associated with less restraint use. On the contrary, no association was found for the availability of a multidisciplinary expert committee regrading restraint use within the institution and regular audits on the ward level to ensure compliance with the guidelines regarding restraint use. Furthermore, the ndings show that a relevant part of the variance in restraint use is explained at the hospital level (random effect), suggesting that a hospital effect exists regarding restraint use. The difference between hospitals also appears to be greater than that between countries, as might have been expected given the much higher restraint prevalence rate in Switzerland (the country variable was not selected for the model). Thus, there is evidence that, in similar patient situations, restraints are used more frequently in some hospitals than in others. This nding supports assumptions from the literature that, regarding restraint use, local habits, routine and institutional culture seem to play an important role (18)(19)(20)(21)42). Such routine or habitual restraint use, independent of an objective and evidencebased evaluation, violates professional values and fundamental human rights. Therefore, critical interprofessional re ections on the current restrictive practice within hospitals are needed to minimise non-professional, non-legal and nonethical restraint use. However, based on well-known safety models, like the Swiss cheese model, we know that patient safety is not only in uenced by health professionals involved in direct patient care (micro level) (43). The conditions within an institution (meso level) and on a national level (macro level) also have a signi cant impact on patient safety. For this reason, critical re ection on current restraint practices should take place on micro, meso and macro level.
On the micro level, a critical interprofessional re ection of practice is only possible with appropriate knowledge about the topic of interest. Regarding restraint use, it is widely discussed that health professionals in the hospital setting do not have su cient knowledge and expertise (21). As a result, restraints are often applied in situations that are not appropriate (14,19,22,44). For example, restraints are used for fall prevention, even though there is growing evidence that restraints are ineffective in preventing falls (16,17). Also, in this study, indications could be found that knowledge in uences the use of restraints, since attending a refresher course is associated with less restraint use. Thus, in line with the recommendations of a Cochrane review regarding restraint reduction in general hospitals (45), education of health professionals seems to be a relevant component for restraint reduction. In this regard, it seems important that an interprofessional approach is taken, as this is the only way to change the institutional culture, the perception of risk-taking and the work ethic (42). In particular, the results of this study show how important these institution-speci c aspects seem to be (hospital effect).
However, changes in these institution-speci c aspects also require a strong commitment from the meso level. First of all, there is a need for open discussion within an institution, for example to clarify responsibilities for safety (42). Especially in the care of elderly people, the assessment of security issues needs different perspectives (46). For example, functional needs must also be weighed in the decision-making process in terms of using or not using restraints. This is even more important as, like the ndings show, older and more care-dependent patients have an increased risk of being restrained during their hospital stay, and as restraint use is associated with functional decline. In addition, mental and behavioural disorders are associated with a higher use of restraints. This means that a very vulnerable patient group is most affected, i.e. patients who often cannot stand up for themselves; therefore, ethical considerations are even more important. In this regard, the management has the responsibility to support front-line staff by in uencing the structural conditions for example, as also shown in this study, by providing policies/ guidelines that support decision-making or at least restraint management in line with legal and ethical requirements (18, 20-22, 24, 45). In addition, they can adapt the infrastructural conditions, for example by removing restraint equipment from the wards, as it is known that the availability of restraint equipment in uences its use (23). It seems interesting that, in this study, regular audits and the availability of an expert committee were not found to be associated with restraint use. A possible explanation might be that, for both tasks, the individual person (who conducts the audit or is a member of the expert committee) must be able to critically re ect on the situation in which restraints are used and, in particular, to take an outside perspective in order to identify restraint use due to the institutional culture or attitudes. However, as discussed above, the knowledge and expertise of the individual person might be insu cient and therefore no effect of these two organisational factors could be measured.
To support critical re ection on the micro and meso level and thus to support the change in restrictive practice in order to protect human rights of personal freedom and to ensure professional restraint use, interventions should also be taken on the national (macro) level (43,47,48). For example, in both included countries (Switzerland and Austria), clear legal regulations regarding restraint use in the hospital setting are lacking (15). However, clear regulations, professional statements of nurses or medical associations and national guidelines would help institutions to clarify their policies, would support the uniform education of health professionals and would provide a basis for national quality improvement programs in the hospital setting. Such programs often lead to more uniform monitoring of restraint use within institutions and thus enable comparison, which are both important aspects in restraint reduction (24,49).
As restraint use is a very sensitive issue, in this respect, a national quality measurement with a risk-adjusted comparison should be considered. This is the only way to guarantee that the different patient mix of institutions is taken into account and that a fair statistical comparison can be made (43). Moreover, there is otherwise a risk that institutions with a higher restraint prevalence rate will only see their patient mix (e.g. older, more care-dependent) as the reason for the higher rate and will then re ect on the institution-speci c aspects insu ciently. However, as described, this critical re ection seems to be essential for less restrictive practice. In addition, such efforts on the national level could stimulate a more open information policy regarding restraint use in hospitals, more critical thinking about restrictive practice in general and open discussions both within institutions but also in society. These aspects are well-known from similar approaches in the mental health or long-term care setting (50,51).

Limitations
Beside its relevant ndings, this study has some limitations. First of all, some organisational factors expected to be associated with restraint use (e.g. nurse to patient ratio) and health professional-related factors were not assessed with LPZ 2.0. It is, therefore, possible that the impact of the included organisational factors is over-or underestimated as is the relevance of the hospital effect. Secondly, it is possible that a selection bias exists. Patients who could not give informed consent and had no legal representative available had to be excluded. It could be that these patients were at high risk for restraint use and, therefore, the prevalence rate might be underestimated. Also, the impact of the predictors might be slightly different when including these patients in the analysis. Similar consequences could also be caused by a potential recall or documentation bias because restraint use was assessed over a period of 30 days. However, it is known that, regarding restraint use, the documentation is often incomplete (5,15). Thirdly, the cross-sectional design has its limitations; on the one hand, the patient situations under investigation can uctuate strongly within institutions on the measurement day and, on the other hand, no causal correlations can be identi ed using a cross-section design. For example, greater care dependency could lead to restraint use, but could also be a consequence of restraint use.
Despite these limitations, the results are expected to be generalisable due to the sample size and the methodological accuracy. They provide important indications for future quality development efforts. In this context, it seems to be of interest to investigate explanations for the additional 31% of explained variance on the hospital level. The inclusion of further structural characteristics in data collection and a subsequent analysis or a qualitative approach, for example by observing the (interprofessional) processes surrounding restraint use, could be helpful in this regard.

Conclusions
Regarding restraint use, a hospital effect exists. This indicates that restraints are used more frequently in certain hospitals than in others, even when considering the different patient mix. To provide restraint-free care as much as possible requires both speci c knowledge and appropriate structures. Based on these ndings, considerable potential for restraint reduction appears to exist in the interprofessional critical re ection of decision-making processes within a hospital; especially, the identi cation of situations in which restraints were used out of routine or institutional culture. This critical re ection ideally goes along with addressing the knowledge and attitudes towards restraints of the interprofessional team as well as of the Switzerland, only verbal and no written consent was required, since in 2012 Swissethics and the cantonal ethics committees classi ed the annual LPZ data collection as a quality measurement for which no written consent of the patients is required.
The decisive factors were the aim of the data collection (ensuring and further developing the quality of care), the data collection method, the type of data collected (only data of the regular care process) and the fact that no intervention is carried out. The documentation of the verbal consent was in the responsibility of the participating hospitals. It was recommended that consent be recorded in the patient documentation or centrally for all patients in a separate document.
Data were collected pseudonymously so that no conclusions can be made regarding the individual patients. Participation was voluntary. The whole study was performed in accordance with the Declaration of Helsinki.

Consent for publication
Not applicable.

Availability of data and materials
The data that support the ndings of this study are available from Swiss National Association for Quality Development in Hospitals and Clinics (Swiss data) and Department of Nursing Science from the Medical University of Graz (Austrian data) but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available.

Competing interests
The authors declare that they have no competing interests.

Funding
The quality measurement (data collection) in Switzerland was funded by the Swiss National Association for Quality Development in Hospitals and Clinics (ANQ) and in Austria by the participating hospitals themselves. The entire data analysis was nanced by Bern University of Applied Sciences.
Authors' contributions ST contributed to the conceptualization, methodology, data collection, data curation, data analysis, data interpretation and writing of the manuscript. SH supervised the project and contributed to the conceptualization, acquisition, data interpretation as well as reviewing and editing of the manuscript. SB contributed to the data collection, data interpretation and reviewing and editing of the manuscript. DR contributed to the methodology and the validation of the data analysis, the data interpretation as well as reviewing and editing of the manuscript. SZ supervised the project and contributed to the conceptualization, data interpretation as well as reviewing and editing of the manuscript. All authors read and approved the nal manuscript.