4.1 Search strategy

A literature review was conducted to search for literature published in electronic databases PubMed, Ovid MEDLINE, CINAHL, ProQuest, Web of Science, PsycINFO, Academic Search Premier, ERIC, OmniFile Full Text Select, Cochrane Library and Hua Yi Online Library that met the inclusion criteria as of February 2021 (Appendix S2). The key words set were restraint, physical restraint, belt restraint, elderly, reduced physical restraint, multifactorial intervention, education, consultation, nursing home and care provider. In this study, we focused on Chinese and English literature. Systematic reviews were manually checked to distinguish the related studies missed by electronic databases search (Figure 1).

4.2 Inclusion criteria

4.3 Search outcomes

We initially searched 2116 studies and deleted 750 duplicate files. After reviewing the title and abstracts of 1366 articles, we selected 46 articles for full-text review. Finally, we determined that seven studies matched the inclusion criteria. The other 39 trials were excluded with reasons documented in Figure 1.

5.1 Research objectives and subjects

The seven papers included in the analysis were all designed to examine the effectiveness of reducing elderly restraint. For the definition of physical restraint, one study (Koczy et al., 2011) was based on the definition of restraint by Joanna Briggs Institute (JBI). Furthermore, another study (Pellfolk et al., 2010) explicitly excluded bedrails. However, four studies (Abraham et al., 2019; Huizing et al., 2009; Köpke et al., 2012; Testad et al., 2016) considered any restriction on a person's freedom of movement as physical restraint, including the bedrail. Furthermore, one study (Kong et al., 2017) did not define restraint.

In terms of the study population, six studies (Abraham et al., 2019; Huizing et al., 2009; Köpke et al., 2012; Koczy et al., 2011; Pellfolk et al., 2010; Testad et al., 2016) were conducted among residents, all of whom were elderly, and the majority of whom were female. For the sampling criteria, two studies (Köpke et al., 2012; Pellfolk et al., 2010) used physical restraint for 20% or more of the residents in the facilities; however, one study (Abraham et al., 2019) did not consider it necessary to specify the rate of physical restraint. Additionally, in one study (Koczy et al., 2011), at least five residents used physical restraint. The exclusion criteria were Korsakov's disease and mental illness (Huizing et al., 2009) and the absence of residents on the day of evaluation (Abraham et al., 2019; Köpke et al., 2012). Furthermore, three studies examined from the age of care providers, two of which (Huizing et al., 2009; Pellfolk et al., 2010) had an average age of 37–43.5 years, while the sample in Kong et al. (2017) was over 50 years old. These three studies were conducted on female subjects. About the sample criteria, only one (Kong et al., 2017) specified that each nursing home had more than 70 care providers who had experience with physical restraint, had not received any education about restraint in the past 6 months.

5.2 Interventions

Six of the seven studies were based on theory (Abraham et al., 2019; Huizing et al., 2009; Köpke et al., 2012; Koczy et al., 2011; Kong et al., 2017; Testad et al., 2016) to design interventions. All literature interventions were led by nurse specialists or unit leaders, or representatives to give care providers' interventions. Nursing staff includes registered nurses, licensed practical nurses and nurse's aides (Huizing et al., 2009; Kong et al., 2017; Pellfolk et al., 2010), whereas Testad et al. (2016) included all staff working in nursing homes, including non-nursing staff. Interventions include multifactorial interventions (4 papers), education (2 papers), education and consultation (1 paper) (Table 1).

The multifactorial interventions in four studies (Abraham et al., 2019; Köpke et al., 2012; Koczy et al., 2011; Kong et al., 2017) consisted of policy statements on minimum restraints, education (40–90 min/session, 1 to 3 sessions), intensive training (6 h to 1.5 days), consultation (1 hour/session, 2 h/2 weeks), alternatives and provision of support (e.g. providing needs and advice to care providers, residents, relatives and legal guardians), with a duration of 6 weeks to 6 months for the restraint reduction program (Table 1). Education was adopted in one study (Pellfolk et al., 2010): in addition to representatives who participated in a 2-day education seminar, other care providers attended one education session per month, 30 min each time for 6 months, which was effective in increasing the knowledge of care providers or reducing restraints. However, concerning education in a study (Testad et al., 2016), in addition to unit representatives participating in group discussions and 2-day seminars, other care providers attended a 1-h session per month for at least 7 months, which resulted in a significant reduction in restraint rate in the control group compared to the experimental group. In terms of education and consultation in a study (Huizing et al., 2009), the participants attended a 2-h education program (90 min of education and 30 min of consultation) for five sessions for 2 months, and 1–4 h of consultation each time, at least once a week for 8 months, but they were not effective in reducing restraint. Overall, the above RCT literature suggests that educational interventions are a common strategy to reduce physical restraints among older residents in nursing homes.

5.3 Measurement methods

The outcome scores covered both care providers and residents. For care providers, physical restraint knowledge, attitudes (Kong et al., 2017; Pellfolk et al., 2010) and restraint status (Pellfolk et al., 2010) were included. Based on residents, the main results were restraint rate (Abraham et al., 2019; Huizing et al., 2009; Köpke et al., 2012; Koczy et al., 2011; Testad et al., 2016). While the secondary results included restraint type, restraint intensity, multiple restraints (Huizing et al., 2009; Koczy et al., 2011), falls and related injuries (Abraham et al., 2019; Köpke et al., 2012; Koczy et al., 2011; Pellfolk et al., 2010), use of psychotropic drugs (Köpke et al., 2012; Koczy et al., 2011; Pellfolk et al., 2010; Testad et al., 2016), behavioural symptoms (Koczy et al., 2011; Testad et al., 2016) and quality of life (Abraham et al., 2019). Only one of these trials (Pellfolk et al., 2010) evaluated the effectiveness of physical restraint implemented by care providers and physical restraint status among residents.

5.4 Efficacy of physical restraints

First, in terms of care providers, two trials (Kong et al., 2017; Pellfolk et al., 2010) confirmed that a restraint reduction program could increase care providers' knowledge. Contrastingly, the effectiveness of care provider attitudes (Kong et al., 2017) and restraint reduction (Pellfolk et al., 2010) were only confirmed by one trial each. In terms of timing, the effectiveness of the restraint reduction program on care providers' knowledge was confirmed 6 weeks after the intervention (Kong et al., 2017) and 6 months after the intervention (Pellfolk et al., 2010).

Second, in terms of residents, five trials (Abraham et al., 2019; Köpke et al., 2012; Koczy et al., 2011; Pellfolk et al., 2010; Testad et al., 2016) demonstrated effectiveness in reducing the restraint rate, with one of them (Koczy et al., 2011) confirming the effectiveness of the restraint reduction program on the restraint intensity. However, one study (Huizing et al., 2009) showed no effect in reducing the restraint rate. Three trials (Abraham et al., 2019; Köpke et al., 2012; Pellfolk et al., 2010) demonstrated no restraint reduction on falls and related injuries. Four trials (Köpke et al., 2012; Koczy et al., 2011; Pellfolk et al., 2010; Testad et al., 2016) confirmed no significant increase in restraint's adverse effects reduction programs on psychotropic drug use. Two trials (Koczy et al., 2011; Testad et al., 2016) proved that restraint reduction programs had no significant effect on behavioural symptoms (agitation).

In terms of timing, the reduction in restraint rate was achieved at 3 months (Köpke et al., 2012; Koczy et al., 2011), 6 months (Köpke et al., 2012; Pellfolk et al., 2010) and 12 months (Abraham et al., 2019) after the intervention. There were no significant differences between the two groups in terms of falls, fall-related injuries (Abraham et al., 2019; Köpke et al., 2012; Pellfolk et al., 2010), psychotropic drugs (Köpke et al., 2012; Koczy et al., 2011; Pellfolk et al., 2010; Testad et al., 2016) and behavioural symptoms (agitation) (Koczy et al., 2011; Testad et al., 2016). Nevertheless, Testad et al. (2016) found that at 7 months after the intervention, there was a significant reduction in the restraint rate in the control group (18.4% → 8.8%) compared to the experimental group (19.3% → 18.1%) (p < 0.001; p = 0.025).

5.5 Methodological quality and risk of bias

The quality of the research literature was assessed using the Cochrane Risk of Bias Tool, and the results of the quality assessment of seven studies are shown in Table 2. Five trials (Abraham et al., 2019; Köpke et al., 2012; Kong et al., 2017; Pellfolk et al., 2010; Testad et al., 2016) described methods for generating or assigning random sequences. Three RCTs (Abraham et al., 2019; Köpke et al., 2012; Kong et al., 2017) had appropriate control on the confidentiality principle. One trial (Kong et al., 2017) was blinded to participants and researchers. Three trials (Abraham et al., 2019; Huizing et al., 2009; Köpke et al., 2012) presented blinding of outcome assessors and were rated as low risk of bias.

Four studies (Huizing et al., 2009; Koczy et al., 2011; Pellfolk et al., 2010; Testad et al., 2016) had dropout rates or loss to follow-up of over 20%, and one of them performed an intent-to-treat (ITT) analysis and was rated as low risk of bias. Seven RCTs could report results as predetermined in the study and were rated as low risk of bias.

Deeks et al. (2011) suggested that any type of variability among studies in systematic reviews can be heterogeneous. This includes clinical heterogeneity (variability in participants, interventions and study outcomes), methodological heterogeneity (variability in study design and risk of bias) and statistical heterogeneity (differences in intervention effects assessed across studies). When there are fewer than 10 studies, meta-analysis should not usually be considered (Higgins et al., 2022; Ryan et al., 2016). The inclusion of studies in the meta-analysis was considered inappropriate due to incomplete data (participants) or the heterogeneity in presentation (interventions and outcomes) in the literature, especially when only seven trials were included.

6.1 Research objectives and subjects

The purpose of all seven studies was to investigate the effectiveness of reducing elderly restraint. The definition of physical restraint in the literature is that any restriction on individuals' freedom of movement is considered physical restraint, including bedrails (4 articles), which agrees with Möhler et al. (2012). The use of bedrails is common in clinical practice, and many studies (Brugnolli et al., 2020; Lan et al., 2017; Möhler et al., 2012) refer to inconsistencies in the definition of body restraint, with different definitions of the use of bedrails. Lan et al. (2017) still suggest that the definition of physical restraint should be consistent and expressly stated in the article.

Care providers in the study were primarily female, with an average age of 37–43.5 years (2 articles), and residents were mainly female elderly, which accords with several studies (Hofmann & Hahn, 2014; Huang et al., 2014; Zhu et al., 2014) that women live longer and are more likely to be admitted to a long-term care facility when disabled. One article specified the characteristics of care providers in each nursing home, and the inclusion (4 articles) and exclusion (3 articles) criteria for resident selection were consistent with Lan et al. (2017) suggestion. Both resident characteristics and the work environment of care providers affect the use of physical restraints.

6.2 Interventions

The results of this study indicated that six of the seven studies designed interventions based on theory. Several trials used decision-making process and empirical guidelines (4 papers), and one trial each applied the theory of planned behaviour (1 paper) and the theory of change (1 paper) as strategies for implementing restraint reduction programs; however, there was insufficient evidence to evaluate or discuss these theories or to evaluate or discuss in-depth whether the results confirmed the theory (Lan et al., 2017; Möhler et al., 2012). However, “reducing restraint in care” is the desired goal for all long-term care providers (Kong et al., 2017; Kor et al., 2018) and is the goal for the highest quality of care (Möhler et al., 2012); thus, ongoing quality assessment and management is essential to ensure the quality of care.

Three out of the seven trials identified nursing staff as including registered nurses, licensed practical nurses, and nurse's aides; their restraint reduction programs were targeted at the nursing staff, and interventions were provided by nurse specialists or unit leaders, which were consistent with Hubbs (2013) and Möhler et al.'s findings (2012). Three other publications supported education or consultation of care providers by specialist nurses or unit leaders (American Nurses Association, 2012; Gulpers et al., 2013; Möhler et al., 2012) and indicated that only trained care providers should implement restraints (Dikiciyan, 2012).

The interventions in restraint reduction programs include (1) multifactorial interventions (4 articles) such as policy statements on least restraint policies, education, intensive training (workshop training and consultation), consultation and provision of support; (2) education (2 articles); and (3) education and consultation (1 article). Four multifactorial interventions and one educational intervention improved care providers' knowledge and attitudes, or reduced restraints, which agrees with Lan et al. (2017).

In summary, there is evidence to support the use of multifactorial interventions over single interventions, such as policy changes (reducing the use of restraint policies), education and consultation (by specialist nurses or unit leaders for care providers), development and provision of alternative interventions (e.g. sensor pads, balance training, exercise, special pillows and lowering bed height), and the inclusion of residents and family members (Gulpers et al., 2013).

6.3 Measurement methods

In this paper, the outcome evaluations involved both care providers (restraint reduction knowledge, attitudes and behaviours) and residents (restraint rate, restraint type, restraint intensity, multiple restraints, falls and related injuries, psychotropic drug use, behavioural symptoms and quality of life). Only one study (Pellfolk et al., 2010) evaluated the effectiveness of both care providers and residents, and the results of this study are consistent with those of Möhler et al. (2012) and Lan et al. (2017).

Reduced physical restraint can be considered a quality indicator of long-term care (Fashaw et al., 2020; Mashouri et al., 2020) and is, therefore, associated with physical injuries in the older people, such as immobilization leading to functional decline, pressure sores, falls and cognitive decline (Hofmann & Hahn, 2014), depression (Fariña-López et al., 2013), and quality of life (Hubbs, 2013; Möhler et al., 2012). None of the included trials investigated the impact of the restraint reduction program on physical functioning, pressure sores and depression of residents in nursing homes.

Care providers' knowledge of restraint implementation may influence their attitudes and behaviours toward restraint implementation (Azab & Negm, 2013). Nurses are often the first line of decision-makers about restraint, and care providers are the restraint implementers. Therefore, to implement restraint reduction care, it is vital to understand care providers' knowledge and attitudes about restraint reduction (Azab & Negm, 2013; Kor et al., 2018).

6.4 Efficacy of physical restraints

This study's findings highlight a restraint reduction program's effectiveness in improving care providers (restraint reduction knowledge, attitudes and behaviours) and residents (restraint rate, restraint intensity, falls and related injuries, psychotropic drug use and behavioural symptoms) in nursing homes. Such results agree with the findings of Brugnolli et al. (2020) and Lan et al. (2017); however, their systematic review and statistical analysis were effective only in residents (restraint rate). Moreover, the present study results differ from those of Möhler et al. (2012), whose systematic review only demonstrated ineffectiveness in residents (restraint rate). The possible explanation is that the purpose of the present study and the inclusion criteria varied (intervention type and outcome variables), thus causing different results.

Among the indicators included in the trials for residents, one educational intervention (Testad et al., 2016) instead showed a significant reduction in restraint rate in the control group compared to the experimental group, possibly because the Norwegian government at that time had a comprehensive staff education program and legislative policies in place to support nurses' attitudes and behaviours to reduce restraint. Alternatively, the intervention of education and consultation (Huizing et al., 2009) was not effective, which may be related to the need to explain the definition of physical restraint, the characteristics of dropouts (older and impaired) during the study period, and whether the control group was provided with general routine care or optimized routine care.

6.5 Methodological quality and risk of bias

Five of the seven included studies described random assignment procedures, which are a lottery system using identification codes (Pellfolk et al., 2010), computer-generated randomization (Abraham et al., 2019; Köpke et al., 2012), listing nursing homes in random order (Testad et al., 2016), sealed opaque assignment envelopes (Kong et al., 2017), to determine the randomization mode and to achieve a balance between the experimental group and the control group, rated as low risk of bias. However, in one paper (Abraham et al., 2019), due to poor communication, participants who should have been randomized to intervention group 1 were incorrectly assigned to the control group; however, the authors did not consider this to be a risk of bias as they had done sensitivity analyses on the primary outcomes, which showed comparable results. The other two articles (Huizing et al., 2009; Koczy et al., 2011) only mentioned conducting randomized trials, but they did not give details on how the randomization process was conducted, nor did they give details on the participant selection method, possibly suggesting that the selection of subjects was based on non-probability sampling for convenience (Hubbs, 2013).

Three studies described concealed assignment grouping, either by envelope method (Kong et al., 2017) or by non-participants (Abraham et al., 2019; Köpke et al., 2012). One trial presented blinding of participants and researchers. Furthermore, three trials presented blinding of outcome assessors, which would avoid leading to the Hawthorne effect or assessor bias that reduces intrinsic validity (Hubbs, 2013).

Four studies had dropout rates or loss to follow-up of more than 20%, and one of them performed an ITT analysis to overcome attrition bias. Seven RCTs in the literature could present the results in the manner predetermined in the program. Möhler et al. (2012) suggested that reporting bias was unlikely to affect the results because of the extensive literature search of the registered trials in electronic databases (PubMed, Ovid MEDLINE, CINAHL, etc.).

This study used the RCT system to evaluate the impact of restraint reduction programs on older adults' physical restraints in nursing homes. This review's strength is that it analyses only RCTs on the impact of physical restraints on residents implemented by care providers and the impact on residents' physical restraint status. The results demonstrate that a restraint reduction program led by nurse specialists or unit leaders and targeting care providers provides multifactorial interventions and improves restraint reduction more effectively than an education or consultation intervention alone. The program includes policies that promote institutional restraint reduction, educational sessions (40–90 min/session, 1–3 sessions), consultation (1 h/session, 2 sessions), development and provision of alternative interventions, and support. The duration of effective restraint reduction programs is 6 weeks to 6 months: at least 6 weeks significantly improves care providers' knowledge. The duration of at least 3 months significantly improves the restraint rate of residents. Outcome measures should cover both care providers (restraint reduction knowledge, attitudes and behaviours) and residents (restraint rate, restraint type, restraint intensity, multiple restraints, falls and related injuries, psychotropic drugs, behavioural symptoms, quality of life, physical functioning, pressure sores and depression).

6.6 Limitations

This review has several limitations: (1) Our criteria for inclusion of articles were limited to Chinese and English literature only, and we did not include unpublished literature or manuscripts, which may have omitted parts of the literature, resulting in sampling bias and publication bias and affecting the analysis results; (2) the literature was not included in the meta-analysis due to incomplete data or differences in presentation; and, (3) being limited to the term “restraint reduction program” may have led to the omission of some interventions that were not identified as restraint reduction programs due to inconsistent definitions of restraint. Inconsistent definitions of restraint may lead to different restraint reduction outcomes, such as whether the use of bed rails is considered physical restraint or not (Huizing et al., 2009), but this may be a subject for future research.