Conservative management following closed reduction of traumatic anterior dislocation of the shoulder

Summary of findings for the main comparison. Immobilisation in external rotation versus immobilisation in internal rotation following closed reduction of traumatic anterior dislocation of the shoulder

Immobilisation in external rotation versus immobilisation in internal rotation following closed reduction of traumatic anterior dislocation of the shoulder
Patient or population: patients undergoing conservative management after closed reduction of traumatic anterior dislocation of the shoulder Setting: splints or slings applied in emergency departments or clinics Intervention: immobilisation of arm in external rotation Comparison: immobilisation of arm in internal rotation
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	No. of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Immobilisation in internal rotation	Immobilisation in external rotation
Re‐dislocation Follow‐up: at 12 months or longer	Low risk^a		RR 0.67 (0.38 to 1.19)	488 (6 RCTs)	⊕⊝⊝⊝ VERY LOW^d,e
	248 per 1000	167 per 1000 (95 to 296)
	Moderate risk^b
	312 per 1000	209 per 1000 (119 to 372)
	High risk^c
	417 per 1000	280 per 1000 (159 to 497)
Validated patient‐reported outcome measures for shoulder disability^f Follow‐up more than 24 months	See comments	See comments	‐	380 (4 RCTs)	⊕⊝⊝⊝ VERY LOW^e,g	3 of the 4 trials reported no or little difference in scores. 1 trial (97 participants) reported a difference favouring external rotation in the WOSI score^f: MD −43.20, 95% CI −72.38; −14.02. This, however, is unlikely to be clinically important.
Resumption of pre‐injury activities	See comments	See comments	‐	347 (3 RCTs)	⊕⊝⊝⊝ VERY LOW^e,h	1 study (169 participants) found no evidence of a difference between interventions in the return to pre‐injury activity of the affected arm (RR 1.02, 95% CI 0.80 to 1.29). 2 studies (178 participants) reported on return to sports for the subgroup of participants who had been sports active; both results were in favour of external rotation.
Participant satisfaction with the intervention	See comments	See comments	‐	‐	‐	Outcome not reported
Quality of life	See comments	See comments	‐	‐	‐	Outcome not reported
Any instability: re‐dislocation or subluxation, composite outcome Follow‐up at 12 months or longer	419 per 1000ⁱ	352 per 1000 (260 to 478)	RR 0.84 (0.62 to 1.14)	395 (3 RCTs)	⊕⊝⊝⊝ VERY LOW^e,j	2 other studies (135 participants) provided very low certainty evidence on instability defined as re‐dislocation and/or a positive apprehension test. Although favouring external fixation (RR 0.28, 95% CI 0.14 to 0.57), we judged the evidence at very low certainty^k (downgraded for risk of bias, imprecision and indirectness reflecting the suboptimal nature of this outcome).
Adverse events	See comments	See comments	‐	645 (7 RCTs)	⊕⊝⊝⊝ VERY LOW^l	Adverse events were mostly not prespecified as an outcome, i.e. reported ad hoc. We split these into 'transient and resolved adverse events' and 'important' (serious) adverse events. In the first category, there were 9 cases of shoulder stiffness or rigidity in the external rotation group versus 2 cases of axillary rash in the internal fixation group. There were 3 'important' adverse events: hyperaesthesia and moderate hand pain; eighth cervical dermatome paraesthesia; and major movement restriction between 6 and 12 months. It was not clear to what extent the adverse events could be attributed to the treatment.
*The risk in the intervention group (and its 95% CI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; MD: mean difference; RR: risk ratio; WOSI: Western Ontario Shoulder Instability Index
GRADE Working Group grades of evidence High certainty: We are very confident that the true effect lies close to that of the estimate of the effect Moderate certainty: We are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low certainty: Our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect Very low certainty: We have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect
^a Assumed low risk based on the lowest control group (internal rotation group) risk out of the 6 contributing studies ^b Assumed moderate risk based on the median control risk of the 6 contributing studies ^c Assumed high risk based on the highest control group risk out of the 6 contributing studies ^d Downgraded by 1 level for risk of bias (mainly performance, detection and selection biases), 1 level for inconsistency (substantial heterogeneity: I² = 61%, Chi² = 0.002) and 1 level for imprecision (low number of events, CI overlapping no effect) ^e Publication bias was graded as undetected for all outcomes. We identified 5 studies evaluating immobilisation in external versus internal rotation that have been completed but that have yet not been published. While this suggests a risk of publication bias, we judged that the information available to us was insufficient for downgrading ^f 3 trials reported results based on the WOSI (range from 0 (least disability) to 2100 (worst disability)). 1 trial used the OSI (range from 0 (worst impairment) to 48 (least impairment)) ^g Downgraded by 1 level for risk of bias (mainly performance, detection and selection biases), 1 level for inconsistency (marked unexplainable difference of the effect of 1 study to that of the other studies) and 1 level for imprecision (low number of participants; 3 of the 4 studies found either no or only a small effect that was either reportedly non‐significant or had a CI including overlapping no effect) ^h Downgraded by 1 level for risk of bias (mainly performance, detection, selection biases), 1 level for inconsistency (difference in effect of the 3 studies ranging from a large effect favouring immobilisation in external rotation to no effect) and 1 level for imprecision (low number of outcome events; CIs of 3 of the 4 estimates overlapping no effect) ⁱ Assumed risk based on the median control risk of the 3 contributing studies ^j Downgraded by 2 levels for risk of bias (mainly performance, detection, selection and attrition biases) and 1 level for imprecision (low number of events; CIs for 2 of the 3 estimates overlapping no effect) ^k Downgraded by 2 levels for risk of bias (mainly performance, detection, selection and attrition biases), 1 level for imprecision (very low number of events) and 1 level for indirectness (suboptimal outcome measure) ^l Downgraded by 2 levels for risk of bias (mainly performance, detection, selection and attrition biases), 1 level for imprecision (very low number of outcome events and small study sample sizes; no CIs were reported) and 1 level for indirectness (poor definition and reporting of most adverse events)

Background

Description of the condition

Dislocation of the shoulder occurs when the head of the humerus (the top of the upper arm bone that forms the ball of the shoulder joint) is displaced out of the glenohumeral (shoulder) joint. The extent of dislocation varies from subluxation (partial dislocation) to full dislocation (where the joint surfaces completely lose contact). It is usually diagnosed by a combination of history, physical examination findings and imaging — most often radiography (x‐ray), but more rarely some other imaging modality such as magnetic resonance imaging (MRI). The direction of dislocation varies, but in most primary (first‐time) dislocations, the head of the humerus is displaced anteriorly (forwards) in relation to the glenoid fossa (the socket of the shoulder joint). The cause is usually trauma, typically during contact sports in adolescents and younger adults. In older adults, dislocation may result from a fall from standing height.

The nature and extent of damage to the soft tissue surrounding the shoulder joint from a traumatic anterior dislocation vary, and may involve bony, cartilaginous, ligamentous as well as tendinous or muscular structures (Demehri 2017; Forsythe 2015). Common presentations include the Bankart lesion, characterised by damage to the anteroinferior part of the glenoid labrum (the fibrocartilage rim that deepens the joint socket) and the capsule surrounding the joint (Bankart 1938); and the Hill‐Sachs lesion, which involves a compression fracture of the humeral head, as well as damage to its overlying cartilage (Hill 1940).

Estimates of the incidence of traumatic anterior shoulder dislocation vary across the literature. A recent epidemiological overview of estimates of the incidence of shoulder dislocation in various countries reported incidences per 100,000 person years of 12.3 cases in Denmark, 23.1 cases in Canada, 23.9 cases in the USA, 27.5 cases in Sweden and 56.3 cases in Norway (Cameron 2017). Cameron 2017 reported the incidence is highest during the second and third decade of life, with a peak in the late teens and early twenties, and that it decreases with increasing age. Furthermore, the incidence is higher in males and in athletes. The proportion of males was 71.8% in a large US‐based epidemiological study including a total of 8940 shoulder dislocations (Zacchilli 2010).

Once dislocation has occurred, the shoulder is less stable and is more susceptible to re‐dislocation. Estimates of the rate of re‐dislocation vary considerably across the literature. Two recent systematic reviews, with different inclusion criteria and numbers of studies, of prognostic studies investigating risk factors in people after conservative management of a traumatic first‐time anterior shoulder dislocation reported an overall proportion of recurrent instability (re‐dislocation or recurrent subluxation) of 39% (range 4% to 74%) after a minimum follow‐up of one year (Olds 2015); and of 21% (range 19% to 88%) after a minimum follow‐up of two years (in Wasserstein 2016). Olds 2015 reported proportions of recurrent instability of 51% in people aged 15 to 20 years; 36% in people aged 21 to 40 years; 11% in people aged 41 to 60 years; and 10% in people aged 61 or older. Re‐dislocation mainly occurs within the first year; Wasserstein 2016 reported a mean (SD) of 10.8 (0.42) months for the first episode.

Both Olds 2015 and Wasserstein 2016 found sex, age and concomitant fractures of the greater tuberosity to be key prognostic factors for recurrent instability after a primary traumatic anterior shoulder dislocation. The risk of recurrent instability was reported to be 3.18 times (95% CI 1.28 to 7.89) more likely in males than females in Olds 2015; and 2.68 times (95% CI 1.66 to 4.31) more likely in males in Wasserstein 2016. Olds 2015 found people aged 40 years or under were 13.46 times (95% CI 5.25 to 34.49) more likely to suffer recurrent instability than those older than 40 years. In Wasserstein 2016, people under 40 years were 20 times (95% CI 10.0 to 33.3) more likely to suffer recurrent instability than those who were 40 or over. Conversely, people with a concomitant fracture of the greater humeral tuberosity were found to be 7.69 times (95% CI 3.33 to 16.67) less likely to have a recurrent instability in Olds 2015; and 3.85 times (95% CI 3.33 to 10.00) less likely in Wasserstein 2016. There was poorer quality and often inconsistent evidence for other factors across the studies in the reviews and also across the two reviews.

Description of the intervention

Traditionally, a non‐surgical (conservative) approach, comprising closed reduction, three to six weeks' immobilisation in a sling (i.e. in internal rotation) and a subsequent physiotherapy or physical therapy programme has been used for first time dislocations (O'Brien 1987). However, we note recent trends to earlier mobilisation and thus a shorter duration of immobilisation (e.g. Berendes 2015). Moreover, a period of up to one week of immobilisation is proposed in a recent British (BESS/BOA) patient care pathway, which refers to evidence suggesting that the risk of re‐dislocation is not decreased with longer immobilisation (e.g. Paterson 2010).

Regarding the immobilisation position, recent years have seen much interest in an alternative to the traditional (internal rotation) immobilisation method, whereby the shoulder is immobilised in external (outward) rotation using a custom‐made (Itoi 2003) or commercially manufactured (Sullivan 2007) brace. The interest in immobilisation of the arm in external rotation traces back to work by Itoi and colleagues (Itoi 1999; Itoi 2001; Itoi 2003; Itoi 2007), who found that the separation of the labrum from the glenoid (as present with a Bankart lesion) was significantly reduced when the shoulder was positioned in external rotation compared with the traditional internal rotation (sling) position. Itoi postulated that this may enhance the healing of the Bankart lesion and reduce the risk of recurrent instability. He later suggested that the addition of an abduction component may further improve outcomes (Itoi 2015). Limited published data are available on the use of immobilisation in external rotation in clinical practice. However, the findings of two published surveys, conducted among orthopaedic surgeons in the Netherlands and Germany (Berendes 2015 and Balke 2016 respectively), indicate considerable variability. Berendes 2015 found that only 3% of the participating surgeons immobilised the shoulder in external rotation, whereas 97% immobilised it in internal rotation. Balke 2016 found that 15% of the participating surgeons always immobilised the shoulder in external rotation, whereas 46% did not use this position at all and 39% advised on immobilisation in external or internal rotation individually.

Physiotherapy or rehabilitation, typically started after the immobilisation period, usually entails advice, education and an exercise‐based regimen (typically addressing stability, coordination and strength of the shoulder, shoulder girdle, upper spinal muscles, or combinations of these) aimed at restoring normal shoulder function. This may be supplemented by manual therapy, soft tissue mobilisations and physical modalities.

Surgical intervention has generally been reserved for cases of chronic recurrence or instability. However, a Cochrane Review (Handoll 2004; updated in 2009) comparing surgical with non‐surgical treatment found some limited evidence supporting primary surgery for young adults, usually male, engaged in highly demanding physical activities who have sustained their first acute traumatic shoulder dislocation.

Our review considers the various approaches to post‐reduction conservative treatment, such as the duration and position of sling immobilisation, the modalities used, and the timing and extent of physiotherapy and rehabilitation interventions.

How the intervention might work

The aim of treatment for anterior dislocation is to restore a functional, painless and stable shoulder. The choice of treatment approach will be influenced by patient age and previous history of dislocation, occupation, level of activity, general health and ligamentous laxity and by expectations of patient adherence to a prescribed therapeutic regimen.

The aim of immobilisation is to allow healing. In this connection, some MRI and cadaveric studies of Bankart lesions have shown better and firmer repositioning of the peeled‐away capsule when the shoulder is externally rotated than when it is internally (inwardly) rotated — the position naturally imposed by a sling (Dymond 2011; Itoi 2001; Kitamura 2005; Liavaag 2009; Miller 2004; Moxon 2010; Pennekamp 2006; Seybold 2009; Siegler 2010). This has kindled and sustained interest in the possibility that immobilisation in external rotation may improve healing, and consequently outcomes, in comparison with the traditional approach. However any immobilisation has potential disadvantages, and there is an argument for shortening its duration (Paterson 2010) or forgoing it altogether (Hovelius 2008) to allow early restoration of movement, especially in the middle‐aged to elderly, who are susceptible to stiffness and frozen shoulder as a result of immobilisation (Robinson 2012) but are less prone to re‐dislocation than the young (Wasserstein 2016).

Finally, various exercise interventions might theoretically increase functional stability by restoring proprioception (spatial awareness) in the shoulder joint and by retraining muscles to help maintain joint congruency (Karatsolis 2006); while motion‐limiting braces might prevent re‐dislocation by restricting shoulder movement in vulnerable directions (Murray 2013).

Why it is important to do this review

This is an update of a Cochrane Review last updated in 2014 (Hanchard 2014); this included only four trials, all of which compared the immobilisation positions of external and internal rotation. It concluded that "the evidence is insufficient to demonstrate whether immobilisation in external rotation confers any benefit over immobilisation in internal rotation". Moreover, Hanchard 2014 pointed to a number of unpublished and ongoing trials that could inform this comparison, and to the continuing need for evidence to inform other aspects of conservative management for this injury. Since then, further trials evaluating immobilisation in external versus internal rotation have been published and there are also several registered studies. Questions still surround other aspects of immobilisation (including the timing of application and duration, position and whether any immobilisation is better than none at all, and rehabilitation (its general effectiveness, its relative effectiveness across different settings and the relative effectiveness of different packages and modes of delivery) and motion‐limiting braces. These considerations illustrated the need for an updated review.

Objectives

Methods

Criteria for considering studies for this review

Types of studies

We considered any randomised or quasi‐randomised controlled trials (the allocation of the latter by, for example, hospital record number or date of birth) evaluating conservative treatment after anterior dislocation of the shoulder.

Types of participants

Individuals who have undergone closed reduction for traumatic anterior dislocation of the shoulder. Ideally, the acute anterior shoulder dislocation should have been confirmed by physical examination and radiography or another imaging modality such as MRI. We intended to consider the potential for misdiagnosis, such as a missed proximal humeral fracture, in trials in which the method of diagnosis was unspecified or in which diagnosis was based on physical examination alone. We included trials including participants with concomitant injuries that are often associated with anterior shoulder dislocation, such as a fracture of the greater tuberosity of the humerus or a Bankart lesion, as long as treatment focused on the dislocation rather than on the concomitant injury.

Although we stated that we would include individuals of any age, we correctly anticipated finding no trials focused specifically on the management of traumatic anterior dislocation in children.

We excluded trials focusing on the treatment of participants with non‐traumatic or habitual dislocations, or concomitant fractures such as proximal humeral fractures involving the surgical neck, or multiple trauma; and those focusing on management of neurovascular complications or postsurgical management. We stipulated that trials with mixed populations involving any one indication of the above would be considered for inclusion if the proportion of the latter (e.g. atraumatic dislocation) was clearly defined for each treatment group and was clearly small (< 10%), or if separate data for acute traumatic anterior dislocation were provided.

Types of interventions

We planned the following.

To assess whether a difference exists between outcomes of different methods (including arm position) and durations (including none or intermittent) of postreduction immobilisation. However, we planned to exclude trials comparing variants (e.g. duration, position) or supplements to particular immobilisation techniques unless the general effectiveness of the method had been established.
To assess whether a difference exists between outcomes of the provision of rehabilitation intervention (of any kind) versus no intervention. Examples of rehabilitation interventions include advice and education, active and passive mobilisation, proprioception and stabilisation exercises, scapular setting and trunk stability exercises. These may be used in combination or individually and may be applied in various ways and settings. Although these interventions are potentially available to all patients allocated the rehabilitation intervention, their actual application may vary according to the perceived needs of individual patients. We aimed to assess this separately for the provision of any rehabilitation (a) during immobilisation, and (b) after immobilisation.
To assess whether a difference exists between outcomes of different types of rehabilitation interventions. Comparisons would have included different single modalities or different combinations of rehabilitation modalities. However, we planned to exclude trials comparing different techniques, timing (duration, frequency) and intensity of single rehabilitation modalities until the effectiveness of the modality itself had been established. We also would have excluded trials evaluating pharmacological interventions and trials testing interventions aimed solely at pain relief.
To assess whether a difference exists between outcomes of different methods of delivering/providing various rehabilitation interventions. Comparisons would have included supervised therapy versus home exercises, different methods of supervised therapy (e.g. individual versus group instruction) and differences in the frequency and duration of rehabilitation. In the first instance, we did not plan to include comparisons of rehabilitation intervention delivered by individual professionals (e.g. doctors, physiotherapists, occupational therapists) with different levels or backgrounds of expertise or training.

For this review update we selected the following four key comparisons that we considered to reflect priority questions.

Arm position during immobilisation: immobilisation in external rotation versus immobilisation in internal rotation (comparison already established in Hanchard 2014).
Use and duration of immobilisation: no or limited‐duration immobilisation (≤ 1 week) versus ‘standard’ duration (typically 3 weeks) (comparison proposed in Hanchard 2014). The underlying rationale for the comparison is likely to vary according to the participant group and extent of injury. Thus, we considered other comparisons of duration depending on the underlying rationale.
Provision of formal rehabilitation, typically post immobilisation: no formal rehabilitation (e.g. advice, education, home exercise sheet only) versus formal rehabilitation. We defined ‘formal rehabilitation’ as a therapeutic intervention typically provided by a health professional (e.g. a physiotherapist) other than the treating medical doctor, that typically includes demonstration and provision of supervised home exercises with or without adjunctive passive modalities over a certain period of time.
Preferable timing of provision of formal rehabilitation: during and following immobilisation versus following immobilisation only. We defined 'formal rehabilitation' as a therapeutic intervention provided by a health professional other than the treating medical doctor, i.e. typically by a physiotherapist, and typically including supervised exercises with or without adjunctive passive modalities.

Types of outcome measures

We sought the following outcome measures.

Primary outcomes

Re‐dislocation: separation of the joint requiring reduction and, ideally, verified.
Validated patient‐reported outcome measures (PROMs) for shoulder instability (e.g. Oxford Shoulder Instability Score (Dawson 1999), Western Ontario Shoulder Instability Index (WOSI) (Kirkley 1998)).
Resumption of pre‐injury activities (work, sport, recreational activities) (yes or no).

Secondary outcomes

Participant satisfaction with the intervention
Validated health‐related quality of life outcome measures (e.g. EQ‐5D (standardised measure of health outcome), Short Form‐36 (SF‐36)).
Any instability: subluxation (separation of the joint not requiring reduction) or subjective instability, either individually or grouped with dislocation as a composite outcome.
Important adverse events (not including re‐dislocation or instability) that were plausibly attributable to post‐reduction management (e.g. persistent pain, frozen shoulder). Other adverse events were to be reported narratively.

In addition, we intended to take note of any reports of service utilisation or resource use, for instance length of hospital stay, outpatient attendance and the provision and nature of physiotherapy; and participants' adherence to their allocated interventions.

We excluded studies that did not report on patient‐relevant clinical outcomes but instead reported solely on non‐clinical outcomes (e.g. radiological outcomes) as their link with clinical outcomes is largely unclear; i.e. not sufficiently established.

Timing of outcome measurement

Approximately one‐third of re‐dislocations occur within three months of the initial dislocation, and a further third between three and 12 months (Rhee 2009). We therefore proposed organising outcomes into the following time frames, with greatest importance attached to long‐term reporting.

Short‐term: up to and including three months following dislocation
Medium‐term: greater than three months and up to and including 12 months following dislocation
Long‐term: greater than 12 months following dislocation

Search methods for identification of studies

Electronic searches

We searched the Cochrane Bone, Joint and Muscle Trauma Group Specialised Register (May 2018), the Cochrane Central Register of Controlled Trials (CENTRAL; 2018, Issue 4) in the Cochrane Library (searched 21 May 2018), MEDLINE including Ovid MEDLINE(R) Epub Ahead of Print, In‐Process & Other Non‐Indexed Citations, Ovid MEDLINE(R) Daily and Ovid MEDLINE and Versions(R) (1946 to 16 May 2018), Embase (1974 to 2018 Week 21), CINAHL (1982 to 21 May 2018), PEDro (Physiotherapy Evidence Database) (1929 to May 2018) and OTseeker (Occupational Therapy Systematic Evaluation of Evidence Database) (inception to November 2012). We also searched the WHO International Clinical Trials Registry Platform (22 May 2018), ClinicalTrials.gov (22 May 2018) and the UK National Research Register (2005, Issue 3, now archived) for ongoing and recently completed trials. For this update, the searches were limited to 2013 onwards. We applied no language restrictions.

In MEDLINE (Ovid Web), the subject‐specific strategy was combined with the sensitivity‐maximising version of the Cochrane Highly Sensitive Search Strategy for identifying randomised trials, and this was modified for use in other databases (Lefebvre 2011). Search strategies for CENTRAL, MEDLINE, Embase, CINAHL and PEDro can be found in Appendix 1. Details of the search strategies used previously are published in Hanchard 2014 and Handoll 2006.

Searching other resources

We checked reference lists of articles. We searched the conference proceedings of the British Elbow and Shoulder Society (BESS), published in Shoulder & Elbow (2013 to 2017). We handsearched conference proceedings published in Orthopaedic Proceedings, a supplement to The Bone and Joint Journal (January 2013 to May 2018). We also checked the lists of ongoing studies and studies awaiting classification in Hanchard 2014 for any publications of these studies.

Data collection and analysis

Selection of studies

Both review authors (CB, CMR) independently assessed potentially eligible trials for inclusion; we resolved all disagreements through discussion. Titles of journals, names of authors and names of supporting institutions were not masked at any stage.

Data extraction and management

Both review authors independently extracted data. We piloted the date‐extraction form on an excluded study. We resolved any disagreement through discussion.

Assessment of risk of bias in included studies

We assessed risk of bias independently, without masking the source and authorship of trial reports. We piloted the assessment form on one trial. Between‐rater consistency in assessment was checked by one review author (CB) at data entry; and we resolved all disagreements by discussion. We used the Cochrane 'Risk of bias' tool (Higgins 2011): this tool incorporates assessment of randomisation (sequence generation and allocation concealment), blinding (of participants and of treatment providers), blinding of outcome assessment, completeness of outcome data, selection of outcomes reported and other sources of bias. Among these other sources we considered discrepancies in the level of skill or care with which compared interventions were applied (performance bias) and commercial sponsorship (because of the potential for reporting bias). In this update, in line with the stated intention in Hanchard 2014, we initially intended to assess risk of bias for objective outcomes (e.g. re‐dislocation) and subjective outcomes (e.g. PROMs) separately in our assessment of blinding of outcome assessment and completeness of outcome data. However, we changed this approach and assessed these two domains separately for each reported outcome instead.

Measures of treatment effect

When available and appropriate, we present quantitative data for outcomes listed in the inclusion criteria graphically. We calculated risk ratios (RRs) and 95% confidence intervals (CIs) for dichotomous outcomes. We calculated mean differences (MDs) and 95% CIs for continuous outcomes.

Unit of analysis issues

We were alert to the remote possibility of unit of analysis issues in the included studies but in the event we identified none. One participant in Chan 2018 was included twice: they experienced two shoulder dislocations, one in each shoulder, at an interval of three years (additional information provided by contact author). We considered the potential impact of this single case on the results was negligible.

Dealing with missing data

We approached study authors for missing data by email. If we requested answers to more than a few questions, we provided a pro forma to ensure clarity and to minimise the burden on trial authors. We did not impute missing data.

Assessment of heterogeneity

We tested heterogeneity between comparable trials using a standard Chi² test; this we considered statistically significant at a P value of less than 0.1. When we noted some indication of heterogeneity, from visual inspection of the results or based on results of the Chi² test, we also quantified heterogeneity/inconsistency using the I² statistic (Higgins 2003). We interpreted this as follows, according to guidance in Section 9.5.2, Higgins 2011.

0% to 40%: might not be important
30% to 60%: may represent moderate heterogeneity
50% to 90%: may represent substantial heterogeneity
75% to 100%: represents considerable heterogeneity

Assessment of reporting biases

If a meta‐analysis of a key outcome had included more than 10 studies, we would have considered exploring the potential for publication bias by generating a funnel plot. We considered the presence and number of completed but yet unpublished trials as an additional potential source of publication bias, but judged that the available information was too limited to allow for a clear judgement.

Data synthesis

Where appropriate, we pooled results of comparable studies using both fixed‐effect and random‐effects models. We decided the choice of the model by careful consideration of the extent of heterogeneity and whether it could be explained, in addition to other factors, such as the number and size of included studies. We used 95% CIs throughout. We considered not pooling data where there was considerable heterogeneity (I² statistic value ≥ 75%) that could not be explained by the diversity of methodological or clinical features among trials. Where it was inappropriate to pool data, we present trial data in the analyses or tables for illustrative purposes and report these in the text.

Subgroup analysis and investigation of heterogeneity

We proposed, where possible, to undertake subgroup analyses by sex, as males are at much greater risk of re‐dislocation (Olds 2015; Wasserstein 2016). We also proposed to subgroup by age, using two thresholds: 20 years or younger versus 21 years or older; and 39 years or younger versus 40 years or older. We chose the former threshold because patients aged 21 years or younger are at much greater risk of re‐dislocation (Wasserstein 2016), and the latter because of the markedly increased susceptibility of patients older than 40 years to post‐immobilisation stiffness and secondary frozen shoulder (Robinson 2012). We further proposed to subgroup by presence (versus absence) of a concomitant fracture of the greater tuberosity of the humerus, as there is evidence of an association between the presence of a greater tuberosity fracture and a decreased risk of instability or recurrent re‐dislocation (Olds 2015; Wasserstein 2016); or of presence (versus absence) of another specific lesion resulting from the dislocation (e.g. a Bankart lesion). However, there were insufficient data to conduct most of these subgroup analyses.

We also planned but did not carry out separate outcome analyses of (1) participants who were physically active compared with those who were more sedentary; (2) physically active young adults engaged in highly demanding physical activities who have sustained primary anterior dislocation compared with others; and (3) participants with a primary dislocation compared with those with a recurrent dislocation. We anticipated that any subgroup differences would be in terms of size of effect (quantitative interaction) rather than direction of effect (qualitative interaction).

We also considered conducting an exploratory subgroup analysis of trials in which immobilisation in external rotation had an abduction component versus those without (external rotation only).

To test for differences between subgroups, we planned to inspect the overlap of confidence intervals and to perform the test for subgroup differences available in Review Manager 5 software.

Sensitivity analysis

We intended to perform sensitivity analyses, when appropriate, to investigate various aspects of trial and review methodology. We intended to include, when data were available, examinations of the effects of (1) removing trials at high risk of selection bias from inadequate allocation concealment or at high risk of detection bias from lack of blinded outcome assessment; (2) conducting worst‐case analyses for trials with missing data; and (3) using fixed‐effect versus random‐effects models for pooling.

'Summary of findings' table and assessment of the certainty of the evidence

Where data were available, we proposed to produce a 'Summary of findings' table for each of the prespecified priority comparisons (see types of interventions) and all primary and secondary outcomes. We applied the GRADE approach to assess the certainty of the evidence related to each of the key outcomes listed in the Types of outcome measures (see Section 12.2; Schunemann 2011). We used the GRADE approach following guidance from the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). However, as the outcome data for most outcomes could only be pooled to a very limited extent, we further considered recent guidance on the application of the approach to evidence when data have been summarised narratively rather than by meta‐analysis (Murad 2017).

Results

Description of studies

Results of the search

We carried out searches were carried out in May 2018 and covered the period between September 2013 and May 2018 (see Appendix 1). We screened a total of 918 records from the following databases: Cochrane Bone, Joint and Muscle Trauma Group Specialised Register (1); CENTRAL (155), MEDLINE (146), Embase (275), CINAHL (221), PEDro (18), the WHO ICTRP (53) and ClinicalTrials.gov (49). We further found three potentially eligible studies from other sources (one through the searches for conference proceedings, one through handsearching for further information on an ongoing trial and one incidentally through a different search).

After removing duplicates, we reduced 918 records to 665. We examined the titles and abstracts of these records and discarded 650, leaving a total of 15 new records (including trial registrations) to examine in more detail. Where possible, we obtained full‐text copies of these potentially relevant records.

We attempted to contact the investigators of the unpublished trials, either listed as ongoing or awaiting classification in Hanchard 2014 or newly identified, for information on their current status (ACTRN12611001183976; ACTRN12616001241426; Chan 2018 (formerly Kelly 2011); Eshoj 2017; ISRCTN41070054; ISRCTN48254181; Miller 2007; Murray 2016 (formerly NCT01111500); NCT02197819; NCT00707018). We received responses in relation to four studies (ACTRN12611001183976; ACTRN12616001241426; Chan 2018; Murray 2016).

At this stage, we excluded seven records. We also excluded one formerly ongoing study that had been abandoned (ACTRN12611001183976). New reports or information resulted in the inclusion of three more trials (Chan 2018: formerly Kelly 2011, a study awaiting classification; Heidari 2014; and Whelan 2014: formerly Whelan 2008, a study awaiting classification). Two registered studies (ACTRN12616001241426; NCT02197819), one study reported in a published protocol (Eshoj 2017), and ARTISAN, a study reported in the UK National Institute for Health Research Health Technology Assessment Database (NIHR‐HTA), were added as ongoing studies. One previously ongoing study was moved to studies awaiting classification (Murray 2016, formerly NCT01111500). We linked any references pertaining to the same study under a single study ID.

There are now seven included trials (Chan 2018; Finestone 2009; Heidari 2014; Itoi 2007; Liavaag 2011; Taskoparan 2010; Whelan 2014); 17 excluded studies (including nine from the previous searches); five ongoing trials; and six studies awaiting classification.

A flow diagram summarising the study selection process for this update is shown in Figure 1.

Figure 1

Study flow diagram for this update (2019)

Included studies

Full details of the individual studies are given in the Characteristics of included studies tables. These also include information about funding sources and declarations of interests. We attempted to contact the corresponding authors of the three newly included trials for additional information on specific aspects including outcome data (Chan 2018; Heidari 2014; Whelan 2014). We received responses from one of the authors of Chan 2018 (Dr Kieran Bentick, personal communication 1 September 2018). We did not receive responses from the authors of Heidari 2014 and Whelan 2014.

Study design

Of the seven parallel, two‐group included clinical trials, six were RCTs (Chan 2018; Finestone 2009; Heidari 2014; Itoi 2007; Liavaag 2011; Whelan 2014), and one was a quasi‐randomised trial (Taskoparan 2010).

Study setting

The seven trials were conducted in seven different countries. Chan 2018 was conducted in six NHS hospitals across England; Finestone 2009 in the Accident and Emergency Department of an Israeli university hospital; Heidari 2014 in the emergency department of a university‐affiliated hospital in Iran; Itoi 2007 in 12 hospitals across Japan; Liavaag 2011 in 13 hospital emergency departments in Norway; Taskoparan 2010 in an emergency department in Turkey; and Whelan 2014 in three university clinics in Canada.

Study size

The trials included 704 participants. Study size ranged from 33 participants in Taskoparan 2010 to 198 in Itoi 2007. Of note: Chan 2018 had aimed to recruit 160 participants, but stopped at 72 participants as the trial was discontinued early.

Participants

All participants had experienced a primary traumatic anterior dislocation of the shoulder reduced by various closed methods. Regarding the prevalence of concomitant injuries, all but one of the trials specified 'associated fractures of the shoulder' as an exclusion criterion; Taskoparan 2010 did not provide any information. Liavaag 2011 and Whelan 2014 specified the type of fractures and other injuries, including labral lesions, that were excluded (see Characteristics of included studies). None of the trials provided information about the prevalence and characteristics of concurrent injuries for their actual study samples.

Five studies evaluated mixed, general populations, without restrictions of age and sex. In Finestone 2009, all 51 participants were male, and 40 of these were soldiers. Liavaag 2011 limited inclusion to patients aged 16 to 40 years, and Heidari 2014 limited inclusion to patients aged 15 to 55 years. Of the 704 participants, 578 (82%) were male. The mean age of the participants across the trials was 29 years: these ranged from 20 years in Finestone 2009 to 37 years in Itoi 2007. Participant age ranged from 12 to 90 years; both extremes were reported in Itoi 2007.

Interventions

All seven included trials compared post‐reduction immobilisation of the affected arm in external rotation (the arm oriented outwards and the forearm away from the chest) versus immobilisation in internal rotation (the traditional sling arrangement, with the forearm rested across the abdomen). In Table 1, the key characteristics of the immobilisation treatment are summarised for each trial.

Table 1. Key characteristics of the immobilisation treatment

Study	Start of immobilisation (post‐dislocation)	Duration of immobilisation	External rotation position	Type of braces and slings	Providers
Chan 2018	within 5 days	4 weeks	30° + 30° abduction	external rotation brace: Smartsling, Ossur, Reykjavik, Iceland); internal rotation sling: Polysling, Mölnlycke Health Care, Gothenburg, Sweden)	appropriately trained members of staff (surgeons, nurses or healthcare assistants)
Finestone 2009	no information	4 weeks	15° to 20°	external rotation brace: (manufacturer unspecified); internal rotation sling: traditional internal rotation sling	unclear
Heidari 2014	presumably on the same day (patients presented within 6 hours)	3 weeks	10° + 15° abduction	external rotation brace: stabiliser brace with adjustable angle of abduction (body: hard polyethylene); presumably commercially manufactured but purpose‐designed; internal rotation sling: sling and swathe bandage	unclear
Itoi 2007	within 2 days (termed day 1 to 3)	3 weeks	10°	external rotation brace: a) wire‐mesh splint covered with sponge and a stockinette (until October 2003); b) prototype brace, Alcare, Tokyo, Japan (from November 2003); internal rotation sling: sling and swathe	the treating surgeons
Liavaag 2011	within 24 hours	3 weeks	15°	external rotation brace: 15° UltraSling ER; DonJoy, Vista, California); internal rotation sling: normal collar and cuff device or sling and swathe	unclear
Taskoparan 2010	on the same day	3 weeks	10°	external rotation brace: "specific splint fixated in 10 degrees external rotation and adduction" (polyethylene/thermoplastic); internal rotation sling: 1st day: "valpaeu bandaging"; from 2nd day: "waist‐assisted sling"	unclear
Whelan 2014	presumably within 7 days (patients were assessed within 7 days)	4 weeks	0° to 5°	external rotation brace: DonJoy (Vista, California) external rotation shoulder brace; internal rotation sling: traditional internal rotation sling	certified orthopaedic technicians

Timing of immobilisation

There was some variation in the commencement of immobilisation. In Liavaag 2011 and Taskoparan 2010 immobilisation commenced on the day of the dislocation; in Itoi 2007 within two days after dislocation; and in Whelan 2014 within five days after dislocation. Heidari 2014 and Whelan 2014 did not specify when treatment commenced, but presumably treatment commenced shortly after the assessment in these trials, which was within six hours after dislocation in Heidari 2014 and within seven days after dislocation in Whelan 2014.

Duration of immobilisation

The duration of immobilisation, whether internal or external rotation, was three weeks in Heidari 2014, Itoi 2007, Taskoparan 2010 and Liavaag 2011; and four weeks in Chan 2018, Finestone 2009 and Whelan 2014. Participants were mostly instructed to remove their brace or sling only for showering.

Types or brands of braces and slings

The trials used a variety of external rotation braces. Chan 2018,Liavaag 2011 and Whelan 2014 reported using commercial off‐the‐shelf braces and specified the specific brand or manufacturer, or both. Immobilisation in internal rotation was mostly done with a traditional sling or sling and swathe bandage.

Position of immobilisation

The degree of external rotation used varied from 0° to 5° in Whelan 2014; 10° in Heidari 2014, Itoi 2007 and Taskoparan 2010; 15° in Liavaag 2011; 15° to 20° in Finestone 2009; to 30° in Chan 2018. In two trials, the arm was additionally immobilised in abduction: 30° in Chan 2018 and 15° in Heidari 2014.

Provision of immobilisation

The providers of the braces and slings, and of the accompanying instructions, were unclear in four studies (Finestone 2009; Heidari 2014; Liavaag 2011; Taskoparan 2010). In Chan 2018, the braces and slings were provided "by appropriately trained members of staff" (i.e. usually by surgeons, nurses or healthcare assistants; additional information provided by the contact author); in Itoi 2007, by the treating surgeons; and in Whelan 2014 by certified orthopaedic technicians.

Post‐immobilisation treatment

In six of the seven included trials, treatment with immobilisation was followed by some form of rehabilitation for both groups. In Liavaag 2011, there was no mention of rehabilitation. Most of the trials provided only limited information about the post‐immobilisation treatment, and most did not specify parameters such as duration or frequency. Whelan 2014 was alone in reporting an overall duration of 16 weeks. Chan 2018 was alone in providing the physiotherapy protocol for their study as a supplement to their report, in which a staged approach to rehabilitation was outlined, which ranged from an initial four‐week "quiet time" to "late rehabilitation" after six weeks. In all trials reporting post‐immobilisation treatment, exercises were the key component. These were supervised by physiotherapists in four trials (Chan 2018; Finestone 2009; Heidari 2014; Whelan 2014), with no information available for Itoi 2007 and Taskoparan 2010).

Outcomes

Primary outcomes

Only Itoi 2007 did not report on re‐dislocation as a discrete outcome. It was not always clear how this outcome was assessed and confirmed in the other six trials: in two trials, Chan 2018 and Heidari 2014, the assessment was either completely or partly patient‐reported; with either no, or no mention of, verification through reference to medical records or further evaluation.

Four studies reported on one or more validated patient‐reported outcome measures for shoulder instability. Three trials used the Western Ontario Shoulder Instability Index (WOSI) (Heidari 2014; Liavaag 2011 ; Whelan 2014) and Chan 2018 used the Oxford Shoulder Instability Index (OSI).

Three studies reported on return to pre‐injury sport or activities (Heidari 2014; Itoi 2007; Liavaag 2011).

Secondary outcomes

None of the included studies reported on participant satisfaction or on generic health‐related quality of life measures (e.g. EQ‐5D or SF‐36).

Any instability, including subluxation or subjective instability, either individually or grouped with dislocation as a composite outcome, was reported in several ways. Itoi 2007 and Liavaag 2011 prespecified re‐dislocation or subluxation as a composite outcome; Liavaag 2011 also prespecified subluxation as a stand‐alone outcome. Whelan 2014 prespecified 'recurrent instability' as the primary outcome; results were reported separately for 'recurrent dislocation', 'recurrent instability' (recurrent dislocation or subluxation) and 'recurrent instability requiring surgical stabilization'. Both Heidari 2014 and Taskoparan 2010 reported the rate of patients with a positive apprehension test.

Although adverse events were mentioned in all of the reports, the trials did not appear to have a priori strategies for defining or collecting these data.

Other outcomes

Adherence was the only other outcome collected by the included studies. Definitions and measurements of adherence varied across the six trials reporting this outcome.

We made the post‐hoc decision to document two further outcomes with the 'other outcomes', because we considered these as of potential interest both to clinicians and researchers: 'difficulties with wearing the braces or slings' and 'surgery'. Although these outcomes may arguably be viewed as 'adverse events', we considered it more appropriate to document them separately. Chan 2018 provided a detailed account of difficulties with wearing the braces and slings. Four studies reported the rate of patients who underwent surgery during the study period (Chan 2018; Finestone 2009; Heidari 2014; Liavaag 2011) .

Follow‐up time points

Follow‐ups were conducted at various time points, and were mostly defined as post‐dislocation (Table 2). However, not all outcomes were reported for all pre‐specified follow‐up time points in all trials, and some trials reported outcomes across a wide period of time that extended considerably beyond their last set follow‐up time. Notably, Taskoparan 2010 provided a table that listed the results of individual participants at individual follow‐ups, ranging from 6 to 41 months and also did not specify when data were collected on re‐dislocation or on adverse events. For Whelan 2014, results are presented for "minimum 12 months' follow‐up", but not for the different follow‐up points.

See: Summary of findings for the main comparison Immobilisation in external rotation versus immobilisation in internal rotation following closed reduction of traumatic anterior dislocation of the shoulder

Table 2. Lengths of follow‐up in the included studies

Study	Final follow‐up	Comments on follow‐up
Chan 2018	24 months	Set follow‐up times: 3 months; 1 & 2 years post‐dislocation
Finestone 2009	mean 33.4 months (range 24 to 48)	Set follow‐up times: 2 & 6 weeks; 3 & 6 months; 1, 2, 3 & 4 years post‐injury
Heidari 2014	24 months and 33 months	Set follow‐up times; 3 weeks post‐intervention; 24 months post‐dislocation (re‐dislocation); 33 months post‐dislocation (WOSI only)
Itoi 2007	mean 25.6 months (range 24 to 30)	Set follow‐up times: 6 months; 1 & 2 years (presumably post‐initiation of immobilisation)
Liavaag 2011	mean 29.1 months (range 24 to 54)	Set follow‐up times: 3 weeks (adherence data); 2 years post‐dislocation.
Taskoparan 2010	mean 21 months (range 6 to 41)	Set follow‐up times: 6 months (function scores); 1 & 2 years (radiographs and MRI) Not specified for re‐dislocation and adverse events Individual patient data presented with follow‐up ranging from 6 to 41 months
Whelan 2014	mean 25 months (range 12 to 43)	Set follow‐up times: 4 weeks and 3, 6, 12, 18 & 24 months post‐dislocation. However, results were presented for a minimum of 12 months

Funding and conflicts of interest

Five trials reported their sources of funding (Chan 2018; Finestone 2009; Heidari 2014; Itoi 2007; Whelan 2014); one stated that no funding was received (Liavaag 2011); and one did not provide any information about funding (Taskoparan 2010).

None of the trials explicitly declared any conflicts of interest. Six studies declared that there were no or at least no financial conflicts of interest while Taskoparan 2010 did not provide any information about conflicts of interest.

Further details about the funding and conflicts of interest are given in the Characteristics of included studies table.

Excluded studies

Fifteen of the 17 excluded studies or articles were excluded mainly as the result of insufficient information and lack of response from study authors (Harper 2000; Kiviluoto 1980; Staply 2002; Wakefield 2001) or failure to meet our selection criteria (Blanchard 2015; Chutkan 2012; Hovelius 1983; Hutchinson 2013; Itoi 2015; Königshausen 2014; Lacy 2015; McCarty 2014; Momenzadeh 2015; Whelan 2010; Xu 2003). Momenzadeh 2015 was excluded because it did not include any of the pre‐specified outcomes of interest for this review; this study focused exclusively on radiological outcomes at three weeks.

As reported in Hanchard 2014, the relationship between Itoi 2003, which was reported as a preliminary study, and Itoi 2007 was unclear. Furthermore, in light of contradictory information received from the trial investigator, we could not rule out the possibility that there were trial participants in common; therefore we excluded Itoi 2003.

Finally, we excluded one previously ongoing study after the principal investigator informed us that the trial had been abandoned (ACTRN12611001183976).

Further details of these studies are given in the Characteristics of excluded studies tables.

Studies awaiting classification

Six RCTs await classification; see Characteristics of studies awaiting classification. Five are parallel, two‐group RCTs comparing immobilisation in external rotation versus immobilisation in internal rotation (ISRCTN41070054; ISRCTN48254181; Miller 2007; Murray 2016; NCT00707018). We have identified no published full reports related to any of these, and our efforts to contact the corresponding authors for information about the current status of their study and the actual or anticipated availability of a published full report were mostly unsuccessful. ISRCTN41070054 (with an initial target sample size of 50, revised down to 38), ISRCTN48254181 with a target sample size of 150 and NCT00707018 with a target sample size of 50 are all completed, according to the WHO International Clinical Trials Registry Platform; and long intervals have passed since the respective anticipated or actual end dates (between 2008 and 2012). Miller 2007 and Murray 2016 are published abstracts, but neither provides sufficient information to stand alone. Miller 2007 reported interim results for 30 participants, but the total sample aimed for or achieved is unknown. Murray 2016 reported results for 50 participants. Dr Murray informed us that publication of this trial is pending (personal communication, 31 May 2018).

Outcomes of these five studies include re‐dislocation at two years (Murray 2016) or at unspecified time points (ISRCTN41070054; ISRCTN48254181); patient‐reported outcome measures for shoulder instability at one year (Miller 2007; NCT00707018); time taken to resume pre‐injury sport or other activities (NCT00707018); and any instability at two years (NCT00707018).

Also still awaiting classification is Itoi 2013. This parallel, three‐group RCT, which was already included as a study awaiting classification in Hanchard 2014, is published as a full report, but compares supplements to a yet unproven method (treatment with or without a shoulder motion restriction band following immobilisation in external rotation). As such, this trial is still not eligible for inclusion at this time.

Ongoing studies

All five ongoing studies are parallel, two‐group RCTs; see Characteristics of ongoing studies. Of these, two compare immobilisation in external rotation versus immobilisation in internal rotation (NCT01648335; NCT02197819). The recruitment status of NCT01648335, with an unspecified target sample size, is 'unknown' (ClinicalTrials.gov, last update posted in March 2013). NCT02197819, with a target sample size of 75, is still recruiting according to the registration record (accessed at ClinicalTrials.gov; last update posted August 2017); the estimated primary study completion date is specified as February 2018. Our attempts to obtain more information on these two trials were unsuccessful. Outcomes of these two studies included re‐dislocation.

The three other ongoing studies compare different aspects of rehabilitation after shoulder dislocation (ACTRN12616001241426; ARTISAN; Eshoj 2017). Recruitment of a target sample of 48 participants has been delayed until 2019 for ACTRN12616001241426, which is intended to test a purpose‐designed smartphone application (including information and an exercise‐based rehabilitation programme) as an adjunct to a supervised rehabilitation programme provided to all participants. The ARTISAN trial is a multicentre, NIHR‐funded study with a target sample size of 478 participants. It aims to compare two different rehabilitation strategies starting after two weeks of immobilisation: a single session of "advice to aid self‐management" versus the same session followed by a course of individually tailored physiotherapy over four months. Eshoj 2017, which has a published protocol, compared a 12‐week specific neuromuscular exercise programme (the 'SINEX' programme) versus 12 weeks of 'standard care'; i.e. a self‐managed shoulder exercise programme with a single introductory supervised physiotherapy session in 56 patients. According to the trial registration record (NCT02371928), data collection for this study was completed in June 2017. We were unsuccessful in our attempt to obtain information on the current status of this study. The primary outcome measures of these three studies are the OSI (ACTRN12616001241426; ARTISAN 2018) and the WOSI (Eshoj 2017).

Risk of bias in included studies

Risk of bias for the seven domains varied across the included studies and across outcomes (see Figure 2; Figure 3). All studies were at some, either high or unclear, risk for one or more outcomes, and we judged none to be at low risk.

Figure 2

Risk of bias graph: authors' judgements about each risk of bias item presented as percentages across all included studies

Figure 3

Risk of bias summary: authors' judgements about each risk of bias item for each included study

Allocation

We judged three of the seven trials to be at low risk of selection bias (Chan 2018; Finestone 2009; Whelan 2010). We judged two at high risk: Itoi 2007, which probably lacked allocation concealment; and Taskoparan 2010, as the result of quasi‐random sequence generation and lack of allocation concealment. We judged the remaining two trials at unclear risk of selection bias as both used sealed envelopes but provided insufficient details to confirm allocation concealment (Heidari 2014; Liavaag 2011).

Blinding

Blinding of participants and personnel

Due to the nature of the studied interventions, participants could not be blinded. We judged the risk of performance bias as 'high' for all seven trials. Regarding the blinding of the care providers, some trials did not provide any information about who applied the braces or slings and gave the initial instructions (Finestone 2009; Heidari 2014; Taskoparan 2010); and in two trials it was unclear whether those who applied the braces and slings and gave the instructions were otherwise independent of the trial (Chan 2018; Whelan 2014). In Itoi 2007, the participants were instructed by the treating surgeons, and the authors themselves stated that "we and the other surgeons might have made a stronger effort to ensure adherence to the external rotation immobilization". In Liavaag 2011, participants were informed of the preliminary results of Itoi 2007, which had favoured immobilisation in external rotation.

Blinding of outcome assessment

Re‐dislocation

We judged two trials that assessed re‐dislocation to be at low risk of detection bias because it was confirmed either by reference to medical records (Liavaag 2011) or radiography and/or record of manipulative reduction in a controlled hospital or healthcare setting (Whelan 2014). We rated another four studies at unclear risk of detection bias, reflecting either a lack of information how re‐dislocation was assessed and/or verified (Finestone 2009; Taskoparan 2010) or that re‐dislocation was (at least partly) patient‐reported, without further verification (Chan 2018; Heidari 2014). We considered that patients' reports of re‐dislocation, a distinct and clear‐cut event, are less vulnerable to subjectivity than other patient‐reported outcomes.

Validated patient‐reported outcome measures for shoulder instability

We judged all four trials that assessed this outcome to be at high risk of detection bias (Chan 2018; Heidari 2014; Liavaag 2011; Whelan 2014); the reason being that these patient‐reported outcome measures are inherently subjective.

Resumption of pre‐injury activities

We judged all three trials that assessed this outcome at unclear risk of bias (Heidari 2014; Itoi 2007; Liavaag 2011). Although this assessment is subjective we considered this outcome, assessed by a simple statement about whether or not the participants had resumed pre‐injury activities or sports, as less susceptible to bias than other subjective outcomes.

Any instability

We judged the three trials that assessed a composite outcome comprising re‐dislocation or subluxation at high risk of detection bias (Itoi 2007; Liavaag 2011; Whelan 2014). This is because subluxation is inevitably subjective and generally much less amenable than re‐dislocation to a clear‐cut definition. There was no mention of blinding for the two trials assessing instability via a clinician‐reported positive anterior apprehension test; hence we rated the risk of detection bias as unclear for these (Heidari 2014; Taskoparan 2010).

Adverse events

We rated all seven trials at unclear risk of detection bias for this outcome given the lack of information to allow a judgement.

Incomplete outcome data

Re‐dislocation

We judged the risk of attrition bias for re‐dislocation at final follow‐up was low for four trials as there were either no or small losses to follow‐up (Chan 2018; Finestone 2009; Heidari 2014; Liavaag 2011). We judged the risk as unclear for Whelan 2014, where no reason was given for the losses to follow‐up but these were similar in the two groups (13% versus 14%). We rated Taskoparan 2010 at high risk of attrition bias because the flow of participants throughout the study was unclear and there appeared to have been a large unexplained (and also unbalanced) number of losses to the 2‐year follow‐up (63% in the external rotation group and 35.3% in the external rotation group).

We judged the risk of bias was high for the interim scheduled follow‐ups at 3 and 12 months for Chan 2018 because the losses were unbalanced between the groups, with high proportions of losses in the internal rotation group: 22% versus 8% at 3‐month follow‐up; and 33% versus 14% at 12‐month follow‐up).

Validated patient‐reported outcome measures for shoulder instability

We rated two of the four trials that assessed this outcome at low risk of attrition bias because the losses were small (Heidari 2014; Liavaag 2011). We judged the risk as unclear for Whelan 2014, for the same reasons as above for re‐dislocation. We judged Chan 2018 at high risk because the losses to follow‐up were unbalanced at both the final 24‐month follow‐up (31% versus 17%) and 12‐month follow‐up (33% versus 17%). The reason for all losses to follow‐up was the inability to contact these participants. No missing data were replaced.

Resumption of pre‐injury activities

In two of the three studies who assessed this outcome, the reported outcome data related to a subgroup of participants who had sustained their injury during sports (Heidari 2014; Itoi 2007). We judged Heidari 2014 at low risk given the small loss to follow‐up and, conversely, Itoi 2007 at high risk because of the high loss to follow‐up. We judged Liavaag 2011 at unclear risk because the losses to follow‐up were reasonably low and balanced between the two groups (11% versus 10%).

Any instability

We judged Heidari 2014 was at low risk of bias for this outcome in view of the small loss to follow‐up; and Liavaag 2011 and Whelan 2014 at unclear risk, given the fairly modest loss to follow‐up was balanced in the two groups (13% versus 14% in both trials). We judged the other two trials reporting this outcome at high risk of attrition bias for any instability: in Itoi 2007, this was due to the high and unexplained attrition in both groups (18.3% versus 21.3%); and in Taskoparan 2010, there were very considerable unexplained losses to follow‐up, particularly at two years (62.5% versus 35.5%).

Adverse events

We judged five trials at unclear risk of attrition bias because of a lack of information (Chan 2018; Finestone 2009; Heidari 2014; Liavaag 2011; Whelan 2014); and two trials at high risk given their high loss to follow‐up (Itoi 2007; Taskoparan 2010).

Selective reporting

We judged the risk of reporting bias as low in one trial (Liavaag 2011), the only study for which a separately published a priori protocol was available. We judged the risk of reporting bias as unclear in five studies (Chan 2018; Finestone 2009; Heidari 2014; Taskoparan 2010; Whelan 2014), mainly because of insufficient information due to lack of a protocol, registration record or other publications. Additionally, in Whelan 2014 results were reported for a follow‐up of "at minimum 12 months" (mean 25, range 12 to 43 months), but not for each of the prespecified follow‐up points; it is unclear whether this may have introduced reporting bias. We judged Itoi 2007 at high risk of reporting bias because the start point and selection criteria varied across the available reports and may not have been determined prospectively, with related concerns regarding outcome assessment.

Other potential sources of bias

As per protocol we considered discrepancies in the level of skill or care with which the compared interventions were applied and commercial sponsorship (involvement). We also considered if there were any other sources of bias that were additional to those assessed in the other domains. We rated the risk of other bias based on whether we judged there was a risk of bias relating to the individual items. We judged the risk of other bias was low for five trials (Chan 2018; Finestone 2009; Heidari 2014; Liavaag 2011; Whelan 2014); unclear for Taskoparan 2010; and high for Itoi 2007.

Discrepancies in the level of skill or care with which the compared interventions were applied

Only Whelan 2014 identified the involvement of a certified orthopaedic technician for application of the interventions. The other trials provided no or minimal information on this aspect. Care programmes other than immobilisation and related advice, although mostly poorly described, appeared to have been comparable for both groups in all trials. Although we were unable to assess whether there were different skill levels in the provision of the two interventions, we did not consider any trial was at additional risk of performance bias.

Commercial sponsorship

Four trials were at low risk of bias related to commercial interests, either by explicit confirmation of receiving no benefits from a commercial company (Finestone 2009; Heidari 2014; Liavaag 2011) or an absence of commercial sources among the listed funders (Whelan 2014). In Chan 2018, the external rotation braces were provided by the manufacturer. However, the trial report notes that the manufacturer "[was] not involved in the design, data analysis or preparation of this data", which we consider indicates that the trial was conducted and reported independently of the commercial interests. Taskoparan 2010 did not provide any information about funding and was thus rated at unclear risk of other bias. We judged Itoi 2007 at high risk of bias relating to this aspect as they disclosed potentially substantial disbursements by the manufacturer of their immobilisers.

Other bias

Itoi 2007 reported switching from a locally made wire‐mesh splint covered with sponge and stockinette to a commercially manufactured external rotation prototype brace part way through the trial. We are unsure whether this change is a source of bias but unexplained change to protocol is of concern. Hence we rated this trial at unclear risk of bias for this aspect.

Effects of interventions

All seven trials compared immobilisation of the affected arm in external rotation versus internal rotation. We could not organise outcomes within our planned framework of short term (up to and including three months following dislocation), medium term (greater than three months and up to and including 12 months) and long term (longer than 12 months) because the available data did not allow this. The most common follow‐up was '24 months or longer', which is when over 85% of re‐dislocations would be expected to have occurred (within two years after the initial dislocation; Robinson 2006).

We undertook some meta‐analyses. However, given the limitations of the data, a cautious approach is needed in the interpretation of these, which is reflected in our GRADE assessments of either low or very low certainty.

There were no or insufficient data to carry out our proposed subgroup analyses detailed in Subgroup analysis and investigation of heterogeneity. We present the available data for different age groups for re‐dislocation and 'any instability'. Possibilities for sensitivity analyses were also very limited.

The effects of interventions are reported below; see also summary of findings Table for the main comparison.

Primary outcomes

Re‐dislocation

Six of the seven included studies contributed data from 488 participants (69% of the 704 randomised participants) to the results for this outcome; Itoi 2007 reported only on instability as a composite outcome of re‐dislocation or subluxation.

There is very low certainty evidence that external rotation may make little difference to the risk of re‐dislocation after 12 or more months' follow‐up (55/245 versus 73/243; RR 0.67, 95% CI 0.38 to 1.19; 488 participants; 6 studies; I² = 61%; see Analysis 1.1; Figure 4). We rated the quality of this evidence as very low after downgrading by one level for risk of bias, one level for inconsistency and one level for imprecision. We provide further data for re‐dislocation in Analysis 1.2. The mean proportion of participants with re‐dislocation across the six trials that investigated this outcome was 22% (range 4% to 37%) in the external rotation groups and 30% (range 25% to 42%) in the internal rotation groups. Sensitivity analysis shows that the removal of the only quasi‐RCT (Taskoparan 2010), which was at high risk of selection and attrition biases, resulted in little change to the results (RR 0.73, 95% CI 0.41 to 1.29; 455 participants, I² = 63%; analysis not shown). Sensitivity analysis confirmed the visual impression that Heidari 2014, the only trial that found a statistically significant result in favour of external rotation, was the source of the significant heterogeneity. The results after the removal of Heidari 2014 did not show a difference between the two groups (53/194 versus 56/192; RR 0.95, 95% CI 0.69 to 1.32; 386 participants; I² = 1%; analysis not shown).

Figure 4

Forest plot of Re‐dislocation at ≥ 12 months follow‐up (nearest to 24 months accepted)

Chan 2018 provided additional data for prespecified follow‐ups at three months, when only one participant had experienced a re‐dislocation; and 12 months, when there was no evidence of a difference between the two groups (RR 0.88, 95% CI 0.37 to 2.10; 55 participants); Analysis 1.3.

Subgroup analyses

There were insufficient data to conduct subgroup analyses. Moreover, consideration was further complicated by the findings that the removal of Heidari 2014 alone resolved the finding of significant heterogeneity; and moreover resulted in evidence of little or no difference between the two groups. Data that could have contributed to subgroup analysis by age are presented in Table 3; as well as illustrating the insufficiency of the available data, the results for the 39 years or younger versus 40 years or older category are dominated by Heidari 2014.

Table 3. Re‐dislocation: data for different age categories

Subgroup	Study	Age category	External rotation group	Internal rotation group
Data available for planned subgroup analysis age ≤ 20 years versus age ≥ 21 years
age ≤ 20 years	Heidari 2014	≤ 20 years	0/0 (0%)	0/0 (0%)
	Taskoparan 2010	≤ 20 years	0/0 (0%)	0/1 (0%)
age ≥ 21 years	Heidari 2014	≥ 21 years	2/51 (4%)	17/51 (33%)
	Taskoparan 2010	≥ 21 years	1/16 (6%)	5/16 (31%)
Data available for planned subgroup analysis age ≤ 39 years versus age ≥ 40 years
age ≤ 39 years	Heidari 2014	≤ 40 years	2/42 (5%)	16/47 (34%)
	Liavaag 2011	≤ 40 years	28/91 (31%)	23/93 (25%)
	Taskoparan 2010	≤ 40 years	1/12 (8%)	5/14 (36%)
	Whelan 2014	≤ 35 years	6/27 (22%)	8/25 (32%)
age ≥ 40 years	Heidari 2014	≥ 41 years	0/9 (0%)	1/4 (25%)
	Taskoparan 2010	≤ 40 years	0/4 (0%)	0/3 (0%)
Other data, including subgroups
Other	Chan 2018	16 to 44 years	8/33 (24%)	10/33 (30%)
	Finestone 2009	17 to 27 years	19/27 (37%)	10/24 (42%)
	Heidari 2014	21 to 30 years	1/16 (6%)	3/18 (17%)
	Heidari 2014	31 to 40 years	1/26 4%)	13/29 (45%)
	Liavaag 2011	16 to 22 years	19/33 (58%)	13/30 (43%)
	Liavaag 2011	23 to 29 years	6/24 (25%)	7/27 (26%)
	Liavaag 2011	30 to 40 years	3/34 (9%)	3/36 (8%)
	Taskoparan 2010	15 to 75 years	individual participant data in study report

The table shows age‐related subgroup data for the predefined categories of interest (≤ 20 years versus ≥ 21 years; ≤ 39 years versus ≥ 40 years), as well as other subgroups or age ranges as reported in the trials. Actual thresholds varied as shown.

We considered conducting an exploratory subgroup analysis of trials in which immobilisation in external rotation had an abduction component versus those without (external rotation only). However, as well as there being only a few trials available in each subgroup, the heterogeneity between the two trials for the external rotation with abduction groups was considerable (I² = 84%); these results are presented for illustrative purposes only in Analysis 1.4.

Time to re‐dislocation

The majority of re‐dislocations appeared to have occurred within the first year after the initial dislocation. Heidari 2014 and Itoi 2007 reported that most re‐dislocations occurred during the first year after the initial dislocation; however, Itoi 2007 did not provide separate data for this outcome. In Finestone 2009, the mean time to re‐dislocation was 12.4 months (range 4 to 36 months); and in Liavaag 2011, the mean time to re‐dislocation was 11.6 months (range 2 to 24 months). The data provided for the different lengths of follow‐up in Chan 2018 suggest that most dislocations occurred within the first year in this study. Taskoparan 2010 and Whelan 2014 did not provide any data.

Validated patient‐reported outcome measures for shoulder instability

Four of the seven included trials contributed data from 380 participants (54% of 704 randomised participants) to the results for this outcome (Chan 2018; Heidari 2014; Liavaag 2011; Whelan 2014). Three trials reported results based on the WOSI; this score, which comprises 21 items, each of which is scored on a 100 mm visual analogue scale, ranges from 0 (least disability) to 2100 (worst disability) (Angst 2011; Kirkley 1998). Chan 2018 reported results of the OSI; this score, which comprises 12 items (each of which has five response options, scoring 0 to 4), ranges from 0 (worst impairment) to 48 (least impairment) (Dawson 1999; van der Linde 2015). Although Taskoparan 2010 used the Rowe Score, another validated instrument for assessing instability, it was not clear whether this was patient‐assessed. Moreover, the data were not usable since they applied to a large range of follow‐up times (6 to 41 months) and the distribution was very skewed with 22 of the 33 at the top end (95 or 100) of the scale). We did not pool the very limited data that were potentially available for pooling for this outcome. We rated the certainty of the evidence for this outcome as very low after downgrading for risk of bias, imprecision and inconsistency (each by one level).

Heidari 2014 found lower WOSI scores, indicating less disability, at the 33‐months follow‐up in the external rotation group (MD −43.20, 95% CI −72.38 to −14.02; 97 participants; Analysis 1.5). Whelan 2014, which expressed the WOSI result in % of the total (reversed) score (0% = worst outcome, 100% = best outcome), found only a minimal difference between the groups in the WOSI at a mean follow‐up of 25 months: −3.00, 95% CI −12.78 to 6.78; 52 participants; Analysis 1.5). Reporting non‐parametric data, Chan 2018 reported no statistically significant differences between the groups in the OSI at either 24‐month (57 participants) or 12‐month (54 participants) follow‐ups, as did Liavaag 2011 (174 participants) based on WOSI results at a mean of 29 months' follow‐up; see Analysis 1.6.

None of the trials provided estimates of the minimal important difference (MID) to guide interpretation of results for the OSI or WOSI. Also, only very limited MID estimates are yet available from the literature. Based on a cohort study of 105 participants with shoulder instability (primary dislocation or recurrent instability) and using an anchor‐based analysis approach, van der Linde 2017 suggested an estimate of around 6 points for the OSI and an estimate of around 14% for the WOSI (which corresponds to 294 points). These estimates provide some preliminary indication that the results of the trials, including that for Heidari 2014 where the larger 95% confidence limit (72.38) was much smaller than the MID of 294 points, for both the OSI and the WOSI are unlikely to be clinically important.

Resumption of pre‐injury activities

Three of the seven included trials contributed data from 347 participants (49% of 704 randomised participants) to the results for this outcome (Heidari 2014; Itoi 2007; Liavaag 2011). This outcome was poorly defined in each trial, and we considered the information provided was insufficient to justify pooling. We rated the certainty of the evidence for this outcome as very low after downgrading for risk of bias, imprecision and inconsistency (each by one level).

The available results are shown in Analysis 1.7. Both Heidari 2014 and Itoi 2007 reported the return to sports for the subgroup of participants who had sustained their initial injury during sport. Heidari 2014 found a large difference in favour of external rotation in return to pre‐injury sportive activities at 24 months: 26/31 versus 12/38; RR 2.66, 95% CI 1.62 to 4.35; 69 participants). Itoi 2007 found a greater return in the external rotation group at 24 months to any level of sport (43/60 versus 31/49, RR 1.13, 95% CI 0.87 to 1.48; 109 participants) and to pre‐injury sport activity (22/60 versus 10/49; RR 1.80, 95% CI 0.94 to 3.43; 109 participants). However, both 95% CIs for Itoi 2007 included the possibility of a result in favour of internal rotation. Liavaag 2011 found no difference between the two groups in the return to the pre‐injury level of activity with the affected arm at a mean 29 months' follow‐up: 51/83 versus 52/86; RR 1.02, 95% CI 0.80 to 1.29, 169 participants).

Secondary outcomes

Participant satisfaction with the intervention

No studies considered this outcome.

Validated quality of life outcome measures

No studies considered generic health‐related quality of life outcome data (using e.g. EQ‐5D, SF‐36).

Any instability: subluxation or subjective instability (either individually or grouped with dislocation as a composite outcome)

Five of the seven included trials contributing data from 530 participants (75% of 704 randomised participants) reported this outcome in two distinct ways. In three trials, the outcome was defined in terms of subluxation or re‐dislocation (Itoi 2007; Liavaag 2011; Whelan 2014). In the other two trials, we defined it as a positive apprehension text or re‐dislocation (Heidari 2014; Taskoparan 2010). Although we pooled the data for each of these two categories, it was in the knowledge of serious limitations of the available data especially reflecting loss to follow‐up in Itoi 2007 (data available for 159 (80%) of 198 participants); Liavaag 2011 (subluxation data were only available for 163 (89%) of participants but 'recurrent instability' was reported for 184 participants in the trial report); and Taskoparan 2010 (25 (76% of 33) followed up for one year).

There is insufficient evidence of little or no between‐group difference in the number of participants with instability (subluxation or re‐dislocation, or potentially both): 77/205 versus 86/190; RR 0.84, 95% CI 0.62 to 1.14; 395 participants, 3 studies; I² = 31%; very low certainty evidence downgraded two levels for very serious risk of bias and one level for imprecision; Analysis 1.8). The data for our other measure of instability (positive apprehension or re‐dislocation) favoured the external rotation group but were dominated by the re‐dislocation data from Heidari 2014: 8/67 versus 29/68; RR 0.28, 95% CI 0.14 to 0.57; 135 participants, 2 studies; I² = 0%). We rated the certainty of this evidence as very low after downgrading by two levels for risk of bias, one level for imprecision and one level for indirectness (reflecting the suboptimal measurement of the outcome).

Itoi 2007 presented subgroup data related to different age groups for recurrent dislocation or reluxation at a follow‐up of two years or longer, with categories that were similar to our pre‐defined age categories: ≤ 20 years versus ≥ 21 years; ≤ 40 years, ≥ 40. The data are presented in Table 4.

Table 4. Any instability: data for different age categories from Itoi 2007

Subgroup (age)	External rotation group	Internal rotation group
≤ 20 years	11/27 (41%)	13/19 (68%)
≥ 21 years	11/58 (19%)	18/55 (33%)
≤ 40 years	19/62 (31%)	27/50 (52%)
≤ 41 years	3/23 (13%)	4/24 (17%)

The table shows age‐related subgroup data for the predefined categories of interest (≤ 20 years, ≥ 21 years, ≤ 39 years, ≥ 40 years)

Adverse events

With the exception of Chan 2018, trials reported specifically on adverse events, but usually in an ad hoc and incomplete way. It was sometimes unclear whether the reported 'adverse events' were exclusively treatment‐related and we found it hard to judge whether some events merited being considered 'important'. Thus, for example, Heidari 2014 reported three cases in the external fixation group of "transient shoulder rigidity that resolved by the time of the follow‐up at 24 months", without giving an indication of the severity, duration or treatment of the shoulder stiffness. We decided this was probably not an important adverse event and pooled this result with data from Itoi 2007 for six cases of temporary stiffness resolved via the "use of self‐directed range‐of‐motion exercises" and two cases of axillary rash in the internal rotation group in Finestone 2009.

Liavaag 2011 reported two "complications" at unspecified time points: one participant in the external rotation group had hyperaesthesia and moderate hand pain, and one in the internal rotation group had eighth cervical dermatome paraesthesia. Taskoparan 2010 reported that one participant in the internal rotation group had "30° limitation in abduction and 10 degrees in internal rotation in the 6th and 12th months. This patient was 75 years old and had additional rotator cuff problems". Taskoparan 2010 did not clarify whether these limitations were active (as with rotator cuff tear) or active and passive (as with frozen shoulder). We judged the complications from these two trials to be important adverse events. Whelan 2014 reported there had been no treatment‐related complications.

These data, stratified by the two categories of severity, are presented for illustrative purposes in Analysis 1.9: transient and resolved adverse events (9/196 versus 2/181; RR 2.73, 95% CI 0.83 to 9.02; 377 participants, 4 studies; I² = 56%); important adverse events (1/134 versus 2/134; RR 0.61, 95% CI 0.08 to 4.46; 268 participants, 3 studies; I² = 0%). We rated the certainty of the evidence for adverse events as very low after downgrading by two levels for risk of bias, one level for imprecision (very few events) and one level for indirectness reflecting the poor definition and reporting of adverse events.

Other outcomes

We intended to document any reports of service utilisation or resource use (e.g. length of hospital stay), outpatient attendance and the provision and nature of physiotherapy; and participants' adherence to their allocated interventions. Of these, only adherence was addressed in the included studies. We made the post hoc decision to report two further outcomes with the 'other outcomes', as we considered these data of interest to both clinicians and researchers: one was 'difficulties with wearing the braces or slings'; the other one 'surgery'. We did not consider these outcomes as 'adverse events'.

Adherence

Six studies reported on adherence, although using different measures and definitions. The data for complete or top‐level adherence to treatment interventions for six trials are shown in Analysis 1.10, with definitions of the measures used in individual trials provided in the footnotes. The differences in the measures used, the significant statistical heterogeneity (I² = 82%) together with differences in effect direction, meant that we did not pool these data. The extent of 'complete' adherence varied between trials, with almost full adherence reported for Finestone 2009; indeed the only omitted participant had removed his splint two days early. In contrast, only 47% of participants of the internal fixation group used their splints for at least 16 hours for at least 20 days in Liavaag 2011. As reported above, Itoi 2007 admitted the possibility that stronger encouragement to ensure adherence had been given to the external mobilisation group.

Difficulties with wearing the braces or slings

Chan 2018 alone provided a detailed description of reported difficulties with wearing the braces or slings. Overall, 27 of 34 participants (79%) in the external rotation group reported difficulties compared with 13 (46%) of 28 participants in the internal rotation group. The majority of the participants (21/34 (62%)) in the external rotation group complained about difficulties with wearing their braces during the night. Reasons included pain, difficulties with sleeping and loosening of straps. Other complaints included "feeling that the brace did not support the arm (21%), functionally awkward in the daytime (18%), difficulty in application of the sling (15%) and pain when in external rotation (12%)". In the conventional sling group, reported difficulties with wearing the slings included "night symptoms (25%), difficulty in application/loosening of the straps (14%), feeling hot/itchy (11%) and pain around the neck strap (7%)". The "level of comfort", which was assessed on a 4‐point ordinal scale (1 = very uncomfortable; 2 = moderately uncomfortable; 3 = slightly uncomfortable; 4 = comfortable), was reported as significantly higher in the conventional sling group; the median was 3.0 (range 1 to 4) in the external rotation group and 3.5 (range 3 to 4) in the internal rotation group (nonparametric analysis; reported P = 0.02).

Taskoparan 2010 noted that "all patients adapted quite well to the fixation methods".

Surgery

Four trials reported the rates of participants who underwent or were scheduled for surgery (Chan 2018; Finestone 2009; Itoi 2007; Whelan 2014). However, only very limited information was provided regarding the precise indications for surgery, and how the decision for surgery was arrived at. Pooled data showed evidence of little or no difference between the two groups (20/172 versus 24/155; RR 0.76, 95% CI 0.44 to 1.30; 327 participants, 4 studies; I² = 0%; very low certainty data downgraded one level for risk of bias and two levels for imprecision; Analysis 1.11).

Economic outcomes

No data on aspects of service utilisation or resource use (e.g. length of hospital stay), outpatient attendance and the provision and nature of physiotherapy were reported.

Discussion

Summary of main results

This updated review of conservative management following closed reduction of traumatic, anterior shoulder dislocation now includes seven trials, which recruited 704 participants in total. In line with previous versions of this review, all trials made just one comparison: that of immobilisation in external versus internal rotation. The evidence for this comparison is presented in summary of findings Table for the main comparison and summarised below for the seven outcomes.

There is very low certainty evidence that there may be little or no difference between the two interventions in the numbers of participants experiencing a re‐dislocation at 12 months' or longer follow‐up. In summary of findings Table for the main comparison, we identify the lowest, median and highest control group rates of re‐dislocation from the included trials to represent illustrative low‐, medium‐ and high‐risk populations and present the anticipated absolute effects for these three populations. We illustrate these results here for a moderate risk population: based on an illustrative risk of 312 people experiencing a dislocation in the internal rotation group, the pooled results from six studies (RR 0.67, 95% CI 0.38 to 1.19) equates to 103 fewer (95% CI 194 fewer to 60 more) re‐dislocations after immobilisation in external rotation.

There is no evidence of a clinically relevant difference between the two interventions in validated patient‐reported outcome measures for shoulder instability. Individually, the four studies reporting on validated patient‐reported outcome measures for shoulder instability at a minimum of 12 months' follow‐up found no evidence of a clinically important difference between the two interventions (very low certainty evidence).

We are uncertain of the relative effects of the two methods of immobilisation on resumption of pre‐injury activities or sports. One study (169 participants) found no evidence of a difference between interventions in the return to pre‐injury activity of the affected arm (very low certainty evidence). Two studies (135 participants) found greater return to sports in the external rotation group but this finding was limited to subgroups of participants who had sustained their injury during sports activities (very low certainty evidence).

No data were available for participant satisfaction with the intervention or for validated health‐related quality of life outcome measures (e.g. EQ‐5D and SF‐36).

There is no evidence of a difference between the two interventions in the number of participants experiencing instability, defined as either re‐dislocation of subluxation (very low certainty evidence).

There was very low certainty evidence available for adverse events, which were reported on an ad hoc basis in the seven trials. Reported 'transient and resolved adverse events' were nine cases of shoulder stiffness or rigidity in the external rotation group and two cases of axillary rash in the internal fixation group. There were three 'important' adverse events: hyperaesthesia and moderate hand pain; eighth cervical dermatome paraesthesia; and major movement restriction between 6 and 12 months. It was unclear to what extent these three adverse events could be attributed to the treatment.

The main results of this updated review confirm the results of the previous version by Hanchard 2014, and are twofold. Firstly, robust evidence for superiority of immobilisation in external rotation over immobilisation in internal rotation is still lacking, which implies that there is no justification for recommending any change in current clinical practice. Secondly, we found no includable evidence related to our other objectives.

Overall completeness and applicability of evidence

The available evidence from clinical trials on conservative management after closed reduction of traumatic primary anterior shoulder dislocation as yet relates exclusively to the comparison of immobilisation in external versus internal rotation. None of the included studies compared different durations of immobilisation or immobilisation versus no immobilisation. The latter may be of particular interest for older people who are at a much reduced risk of recurrence (Olds 2015; Wasserstein 2016) — but at greater risk of shoulder stiffness (de Boer 2005). Furthermore, none of the included studies addressed interventions following immobilisation, including comparisons of rehabilitation versus no rehabilitation or of different variants of rehabilitation. The lack of evidence on post‐immobilisation interventions is a remarkable finding considering that immobilisation represents only the initial short‐term treatment post dislocation. However, we identified three ongoing trials that were designed to compare different aspects of rehabilitation after traumatic primary shoulder dislocation, with target sample sizes of 48 (ACTRN12616001241426), 80 (Eshoj 2017) and 478 (Kearney 2018). This finding may indicate a timely shift of the focus of interest from initial treatment (with immobilisation) to other interventions (following initial immobilisation).

Overall completeness

Our search for this update led to the inclusion of only three new studies, so the current body of evidence consists of seven studies including 704 participants that compared immobilisation in external versus internal rotation. Pooled data for the main outcomes were available from 488 participants (69% of 704) for re‐dislocation and 395 (56% of 704) for any instability. No data are available for patient satisfaction or quality of life.

This should also be considered in the context of unpublished evidence. There are five completed studies that tested the same comparison, that are awaiting assessment and that may have recruited 318 participants (ISRCTN41070054; ISRCTN48254181; Miller 2007; Murray 2016; NCT00707018). Furthermore, there are two ongoing studies, one of which aims to recruit 75 participants (NCT02197819); the status of the other trial is unknown (NCT01648335). Thus the evidence for this comparison is far from being complete.

Applicability

Population

All participants had experienced a traumatic anterior dislocation of the shoulder that had been reduced by a closed method. All studies except Taskoparan 2010 explicitly excluded patients with associated fractures of the shoulder, and two studies also excluded patients with specific labral injuries (Liavaag 2011; Whelan 2014). However, as none of the trials specified the prevalence and/or characteristics of concurrent injuries within their actual samples, the applicability of the findings to patients with, or without, or with different types of concomitant injuries is unclear.

The evidence from the included studies represents the young and active as well as the relatively sedentary. Four studies evaluated mixed, general populations, with no restrictions on sex or age (Chan 2018; Finestone 2009; Itoi 2007; Whelan 2014). In contrast, all 51 participants were male of which 40 were soldiers in Finestone 2009. The greater proportion of males (82% of 704) in the studies is consistent with the epidemiology for this injury. Although there was insufficient information on the distribution of ages, an assessment of the mean ages and inclusion criteria of the seven studies suggests that the majority of participants were above 20 years and below 40 years old. These indicate that review findings are applicable to the majority of people who sustain this injury.

Interventions

The evidence included in this review relates exclusively to the comparison of immobilisation in external versus internal rotation. Some aspects of these interventions, such as the duration of immobilisation (3 or 4 weeks), were similar across the trials. However, given recent trends to promote immobilisation for shorter periods, such as up to a week (see Background), the longer duration used in the included studies may not reflect current practice or trends. Other aspects differed to some extent across the studies, such as the commencement of the immobilisation (ranging from within a couple of hours until seven days post injury) and the position of immobilisation in external rotation (which ranged from 0° to 5° external rotation in Whelan 2010 to 30° in Chan 2018; and was complemented by an abduction component in Chan 2018 and Heidari 2014 (30° and 15°, respectively)). It is unclear whether these differences could have affected estimates of effects and the applicability of the results. However, the fact that the precise position of the shoulder was mostly neither accurately measured nor monitored, and that the position is likely to have varied to some extent through the handling of the braces by the participants, tends to minimise the differences. The different external rotation angles appeared to be mostly justified by a trade‐off between findings from previous clinical trials or studies investigating structural aspects and the (documented or anticipated) tolerability of maximum external rotation angles. Our exploratory subgroup analysis of data for re‐dislocation related to external rotation with versus without an abduction component does not allow for any firm conclusions. It was not clear who provided the braces and slings as well as the accompanying instructions in four studies (Finestone 2009; Heidari 2014; Liavaag 2011; Taskoparan 2010). Differences in the level of experience and skills may have affected the way the braces and slings were applied, as well as the quality of the instructions that the participants received, and may thereby have affected outcome. Although all studies reported that the immobilisation period was followed by some sort of rehabilitation for both groups, most of the studies provided insufficient information regarding the precise duration, amount and content of rehabilitation treatment.

Outcomes and outcome assessment

Although definitions and reports of follow‐up periods varied across the trials, all trials included participants who were followed up for at least one year and generally two years (Table 2). Chan 2018 was the only study reporting data for interim time points (3 and 12 months). The literature on time to re‐dislocation suggests an average time to re‐dislocation of around 12 months: a recent systematic review by Wasserstein 2016 pooled data from four studies and reported a mean time of 10.8 months (SD 0.42). Regarding the studies included in our review, Finestone 2009 and Liavaag 2011 reported a mean time of 12.4 (range 4 to 36 months) and 11.6 months (2 to 24 months), respectively. Moreover Heidari 2014 and Itoi 2007 reported that the majority of re‐dislocations within their study samples occurred within the first year after the initial dislocation (the rate in Itoi 2007 was 83%). These data show that length of follow‐up was generally acceptable in the included trials. Nonetheless, considering that rates of re‐dislocation can still be expected to vary between follow‐ups of one year and two or more years, assessment at different time points (e.g. one year, two years) would have been helpful to ensure consistency across studies as well as to facilitate judgements about applicability.

All of our primary outcomes were addressed by at least some of the included studies. However, of our secondary outcomes there were no data for patient satisfaction or for quality of life using a validated generic health‐related quality of life outcome measure such as EQ‐5 or SF‐36. Interestingly, the term 'quality of life measure' is also commonly applied for disease specific PROMs, such as the WOSI and ASES (Heidari 2014; Whelan 2014). These outcome measures, though, are disease‐specific and were therefore considered with the validated PROMs for shoulder instability.

All studies included some documentation of adverse events; however, most did not specify adverse events as an a priori outcome, but reported them ad hoc. It was mostly unclear whether or not the reported events were systematically assessed in all participants, and no definitions were provided. Moreover the information provided was insufficient to judge whether the events were, or could be, related to the treatment, or whether they were (part of) distinct medical conditions that occurred sometime during the study periods. This was partly due to a lack of information on when the reported events occurred, but also the severity of the observed events was not sufficiently described. It is notable that all nine cases of short‐term shoulder stiffness occurred in the external rotation group; the possibility of this complication seems something that could be considered when treating people with the more rigid position used for external immobilisation. The a priori definition and specification of adverse events, and the systematic collection of data on adverse events, would have been helpful to determine the type and frequency of adverse events associated with immobilisation in external and internal rotation.

No data on aspects of service utilisation or resource use (e.g. length of hospital stay), outpatient attendance and the provision and nature of physiotherapy were reported.

Although patient satisfaction was not reported, adherence assessed in all seven trials and reported difficulties and complaints relating to brace or sling use measured in Chan 2018 may give some indirect guide to patient satisfaction during treatment. There may, however, be other factors. The high adherence rate in Finestone 2009 may be at least partly due to the fact that most participants were soldiers who, as the authors state, "had a high degree of self‐discipline and were treated within a military framework". Itoi 2007 pointed to the possibility that there had been stronger encouragement to ensure adherence given to the external mobilisation group. Liavaag 2011, whose authors also found higher adherence in the external rotation group, conjectured that participants in this group may have had confidence in the benefit of treatment. In contrast, Heidari 2014 reported greater adherence in the internal rotation group, and itemised in their discussion additional limitations associated with external rotation, such as walking safely through doorways, difficulties finding a comfortable sleeping position and risk of trauma in crowded locations. Data supporting this was presented in Chan 2018, who found a greater proportion of participants in the external rotation group reported difficulties, including with sleeping and feeling functionally awkward in the daytime. This finding is unsurprising, as intuitively one might expect maintenance of external rotation at the shoulder during normal day‐to‐day functioning to be cumbersome.

We introduced a new secondary outcome — 'subsequent surgery' — that helps to give a better picture of what happened to some of the people with recurrent instability. There is no evidence of a difference between the two groups.

Certainty of the evidence

We judged the certainty of the evidence included in this review as very low for all outcomes shown in summary of findings Table for the main comparison. The grading of the quality of evidence means that we are uncertain about the estimates of effect.

We downgraded the certainty of the evidence for all outcomes at least by one level for risk of bias. In aggregate, concerns about risk of bias relate, to some extent, to all risk of bias domains, to all outcomes, and to all trials (Figure 2; Figure 3). Notable aspects were that we judged two of seven trials at high risk of selection bias and, reflecting that blinding to the interventions was not possible for either the participants or personnel, that all studies were at high risk of performance bias. This is demonstrated indirectly in Itoi 2007, where the trial authors raised the possibility that stronger encouragement to ensure adherence had been given to the external mobilisation group. The effect of lack of blinding on the risk of detection bias is likely to be outcome‐dependent, as reflected in our assessments for the different outcomes. The risk of attrition bias also varied with outcome: the presentation of the amount of losses to follow‐up was insufficient and often misleading, even when presented in flow diagrams, as they were not specified for each outcome. There was a high risk of selective reporting bias in Itoi 2007. The key domains that led to our downgrading for risk of bias by one level were selection bias, performance bias and detection bias. The key domain that led to our further downgrading by one level was attrition bias.

We downgraded the certainty of the evidence for several outcomes for inconsistency. For re‐dislocation, we noted that the substantial heterogeneity disappeared on the removal of Heidari 2014, which markedly differed from the other studies with regard to the positive findings in favour of immobilisation in external rotation for other outcomes. There are insufficient data to explore the underlying sources of heterogeneity and, in particular, to identify plausible reasons for the more positive findings in favour of external fixation of Heidari 2014.

We downgraded the certainty of the evidence for all outcomes for imprecision. This reflected wide 95% CIs, often including the possibility of benefit favouring either one of the interventions to various degrees, or no effect; as well as too few events and small sample sizes.

Indirectness was graded as not serious for most outcomes except 'any instability', which reflects the suboptimal measurement of the outcome; and 'adverse events', which reflects the ad hoc approach to collecting data on these and suboptimal reporting. Overall, however, the trials provided direct evidence to the review question despite some variation in the populations, interventions and outcomes.

Publication bias was graded as undetected for all outcomes. We identified five studies evaluating immobilisation in external versus internal rotation that have been completed but that have yet not been published. While this is of concern, as it suggests a risk of publication bias, we judged that the information available to us was insufficient for a clear judgement and thus did not downgrade for this item.

Potential biases in the review process

We strove throughout the development and execution of this updated review to minimise any potential for biases. Thus, when feasible, we adhered throughout to the detailed a priori protocol, which detailed every aspect of the review's aims, objectives and methods, and documented any changes that we considered necessary in Differences between protocol and review. We conducted a reasonably comprehensive search without language restriction, and involving various sources, and the key processes of study selection, data extraction and risk of bias assessment were each independently performed by the two review authors.

We contacted investigators and authors of ongoing studies and studies classified as awaiting classification for clarification of questions related to the status of their studies, and corresponding authors of the included studies for clarification of methodological aspects, missing details or data. Unfortunately we were unsuccessful in obtaining responses in the majority of cases. Regarding the three newly included studies, we received a response only from the authors of Chan 2018. Therefore, we had to mostly rely on the information provided in the study reports for the presentation of the study characteristics and results, and for our judgements of risk of bias, which means that our findings and judgements are affected by the completeness and quality of reporting.

As previously indicated (see Overall completeness and applicability of evidence), we recognise the number of studies awaiting classification as an unavoidable limitation of the review.

Although we did not conduct a formal update of the search after May 2018, we maintained regular checks via Medline (PubMed) auto‐alerts up to December 2018. These yielded Murray 2018, the full‐trial report of Murray 2016, another trial that compared immobilisation in external versus internal rotation; see Characteristics of studies awaiting classification). A brief investigation of the effects of adding the results for re‐dislocation from Murray 2018, which included 50 participants, did not indicate that its inclusion would lead to an important change to the review findings. The study will be included in a future update of this review.

Agreements and disagreements with other studies or reviews

We identified five systematic reviews that evaluated clinical outcomes of conservative care following primary anterior shoulder dislocation and that have been published since 2014 (when the previous version of this review, Hanchard 2014, was completed) (Kavaja 2018; Liu 2014; Longo 2014; Vavken 2014; Whelan 2016 ). Of these, three were limited to the comparison of immobilisation in external versus internal rotation (Liu 2014; Vavken 2014; Whelan 2016); whereas Longo 2014 and Kavaja 2018 had a broader scope. Longo 2014 was designed to compare surgical versus conservative interventions, but included the comparison of immobilisation in external versus internal rotation after closed reduction. Kavaja 2018 included any interventions for treating patients after a traumatic shoulder dislocation or with chronic post‐traumatic shoulder instability. The reviews differed in the numbers of included studies, which is mostly attributable to their publication date. None of the reviews included Chan 2018. Kavaja 2018, the most recently published review, did not include Taskoparan 2010, the reason being that quasi‐randomised trials were not included in this review.

The most commonly assessed outcome in the reviews was re‐dislocation or recurrence, which was assessed by all. Four of the reviews agreed with our findings of no evidence of a difference in the rate of re‐dislocation between immobilisation in external versus internal rotation following closed reduction (Kavaja 2018; Liu 2014; Vavken 2014; Whelan 2016). All four reviews pooled data on re‐dislocation or recurrence, with overall similar estimates despite some variability in the approaches to meta‐analysis. For example, some reviews — Liu 2014,Longo 2014 and Vavken 2014 — included data from Itoi 2007 in their meta‐analysis, hence synthesizing recurrence (re‐dislocation or subluxation) rather than re‐dislocation alone. In contrast to the findings from our review and the four other reviews, Longo 2014 was alone in reporting finding that immobilisation in external rotation resulted in a lower rate of re‐dislocation compared with internal rotation. This may be because Longo 2014 pooled data from just four studies (Finestone 2009; Itoi 2003;Itoi 2007;Taskoparan 2010). Of note: we excluded Itoi 2003 in favour of the reportedly definitive Itoi 2007 to avoid the potential risk of including data from duplicate populations.

Four of the reviews conducted subgroup analyses related to the participants' age, based on subgroups of under 30 years and over 30 years (Kavaja 2018; Liu 2014; Longo 2014; Vavken 2014; Whelan 2016). While we have presented the available data for different age categories, we consider there are insufficient data to justify conducting subgroup analysis. The findings of the other reviews were in agreement with our findings in relation to other outcomes, as far as these were addressed.

All five reviews considered that results were affected by methodological limitations and risk of bias in the included studies. The two reviews using GRADE assessed the evidence for re‐dislocation or recurrence as low quality (Kavaja 2018; Liu 2014); these assessments were based on a different set of trials than those included in our review.

Figure 1

Study flow diagram for this update (2019)

Figure 2

Risk of bias graph: authors' judgements about each risk of bias item presented as percentages across all included studies

Figure 3

Risk of bias summary: authors' judgements about each risk of bias item for each included study

Figure 4

Forest plot of Re‐dislocation at ≥ 12 months follow‐up (nearest to 24 months accepted)

Analysis 1.1

Comparison 1 Immobilisation in external versus internal rotation, Outcome 1 Re‐dislocation at ≥ 12 months follow‐up (nearest to 24 months accepted).

Navigate to figure in Review


Study	Follow‐up	External rotation, n events/n group (%)	Internal rotation,n events/n group (%)
Chan 2018	24 months ("full period")	8/33 (24%)	10/33 (30%)
Finestone 2009	Mean 33.4 (range 24 to 48) months	10/27 (37%)	10/24 (42%)
Heidari 2014	24 months	2/51 (4%)
Liavaag 2011	Minimum 2 years, mean 29.1 (range 24‐54) months	28/91 (31%)	23/93 (25%)
Taskoparan 2010	Mean 21 (range 6 to 41) months	1/16 (6%)	5/17 (29%)
Whelan 2014	Minimum 12 months, mean 25 (range 12‐43) months	6/27 (22%)

Analysis 1.2

Comparison 1 Immobilisation in external versus internal rotation, Outcome 2 Re‐dislocation data at last follow‐up.

Analysis 1.3

Comparison 1 Immobilisation in external versus internal rotation, Outcome 3 Re‐dislocation, interim follow‐ups (3 and 12 months).

Analysis 1.4

Comparison 1 Immobilisation in external versus internal rotation, Outcome 4 Re‐dislocation: stratified according to external rotation with / without abduction.

Analysis 1.5

Comparison 1 Immobilisation in external versus internal rotation, Outcome 5 Validated patient‐reported outcome measures for shoulder disability (OSI, WOSI).

Navigate to figure in Review


Study	Follow‐up	Measure	External rotation	Internal rotation	Reported P
Final follow‐up
Chan 2018	24 months	OSI (0 to 48: least impairment)	median 43.5 (range 12 to 48); n = 30	median 43.5 (range 13 to 48); n = 27	P = 0.50
Liavaag 2011	mean 29 months (24 to 54)	WOSI (0 to 2100: worst disability)	median 238 (interquartile range 101 to 707); n = 86	median 375 (interquartile range 135 to 719); n = 88	P = 0.32
Interim follow‐up
Chan 2018	12 months	OSI (0 to 48: least impairment)	median 43 (range 6 to 48); n = 30	median 42.5 (range 5 to 48); n = 24	P = 0.35

Analysis 1.6

Comparison 1 Immobilisation in external versus internal rotation, Outcome 6 Validated PROMS: non‐parametric results.

Analysis 1.7

Comparison 1 Immobilisation in external versus internal rotation, Outcome 7 Resumption of pre‐injury activities.

Analysis 1.8

Comparison 1 Immobilisation in external versus internal rotation, Outcome 8 Any instability (subluxation or subjective instability, individually or grouped into composite outcome).

Analysis 1.9

Comparison 1 Immobilisation in external versus internal rotation, Outcome 9 Adverse events.

Analysis 1.10

Comparison 1 Immobilisation in external versus internal rotation, Outcome 10 Adherence to treatment.

Analysis 1.11

Comparison 1 Immobilisation in external versus internal rotation, Outcome 11 Subsequent surgery.

Immobilisation in external rotation versus immobilisation in internal rotation following closed reduction of traumatic anterior dislocation of the shoulder
Patient or population: patients undergoing conservative management after closed reduction of traumatic anterior dislocation of the shoulder Setting: splints or slings applied in emergency departments or clinics Intervention: immobilisation of arm in external rotation Comparison: immobilisation of arm in internal rotation
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	No. of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Immobilisation in internal rotation	Immobilisation in external rotation
Re‐dislocation Follow‐up: at 12 months or longer	Low risk^a		RR 0.67 (0.38 to 1.19)	488 (6 RCTs)	⊕⊝⊝⊝ VERY LOW^d,e
	248 per 1000	167 per 1000 (95 to 296)
	Moderate risk^b
	312 per 1000	209 per 1000 (119 to 372)
	High risk^c
	417 per 1000	280 per 1000 (159 to 497)
Validated patient‐reported outcome measures for shoulder disability^f Follow‐up more than 24 months	See comments	See comments	‐	380 (4 RCTs)	⊕⊝⊝⊝ VERY LOW^e,g	3 of the 4 trials reported no or little difference in scores. 1 trial (97 participants) reported a difference favouring external rotation in the WOSI score^f: MD −43.20, 95% CI −72.38; −14.02. This, however, is unlikely to be clinically important.
Resumption of pre‐injury activities	See comments	See comments	‐	347 (3 RCTs)	⊕⊝⊝⊝ VERY LOW^e,h	1 study (169 participants) found no evidence of a difference between interventions in the return to pre‐injury activity of the affected arm (RR 1.02, 95% CI 0.80 to 1.29). 2 studies (178 participants) reported on return to sports for the subgroup of participants who had been sports active; both results were in favour of external rotation.
Participant satisfaction with the intervention	See comments	See comments	‐	‐	‐	Outcome not reported
Quality of life	See comments	See comments	‐	‐	‐	Outcome not reported
Any instability: re‐dislocation or subluxation, composite outcome Follow‐up at 12 months or longer	419 per 1000ⁱ	352 per 1000 (260 to 478)	RR 0.84 (0.62 to 1.14)	395 (3 RCTs)	⊕⊝⊝⊝ VERY LOW^e,j	2 other studies (135 participants) provided very low certainty evidence on instability defined as re‐dislocation and/or a positive apprehension test. Although favouring external fixation (RR 0.28, 95% CI 0.14 to 0.57), we judged the evidence at very low certainty^k (downgraded for risk of bias, imprecision and indirectness reflecting the suboptimal nature of this outcome).
Adverse events	See comments	See comments	‐	645 (7 RCTs)	⊕⊝⊝⊝ VERY LOW^l	Adverse events were mostly not prespecified as an outcome, i.e. reported ad hoc. We split these into 'transient and resolved adverse events' and 'important' (serious) adverse events. In the first category, there were 9 cases of shoulder stiffness or rigidity in the external rotation group versus 2 cases of axillary rash in the internal fixation group. There were 3 'important' adverse events: hyperaesthesia and moderate hand pain; eighth cervical dermatome paraesthesia; and major movement restriction between 6 and 12 months. It was not clear to what extent the adverse events could be attributed to the treatment.
*The risk in the intervention group (and its 95% CI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; MD: mean difference; RR: risk ratio; WOSI: Western Ontario Shoulder Instability Index
GRADE Working Group grades of evidence High certainty: We are very confident that the true effect lies close to that of the estimate of the effect Moderate certainty: We are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low certainty: Our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect Very low certainty: We have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect
^a Assumed low risk based on the lowest control group (internal rotation group) risk out of the 6 contributing studies ^b Assumed moderate risk based on the median control risk of the 6 contributing studies ^c Assumed high risk based on the highest control group risk out of the 6 contributing studies ^d Downgraded by 1 level for risk of bias (mainly performance, detection and selection biases), 1 level for inconsistency (substantial heterogeneity: I² = 61%, Chi² = 0.002) and 1 level for imprecision (low number of events, CI overlapping no effect) ^e Publication bias was graded as undetected for all outcomes. We identified 5 studies evaluating immobilisation in external versus internal rotation that have been completed but that have yet not been published. While this suggests a risk of publication bias, we judged that the information available to us was insufficient for downgrading ^f 3 trials reported results based on the WOSI (range from 0 (least disability) to 2100 (worst disability)). 1 trial used the OSI (range from 0 (worst impairment) to 48 (least impairment)) ^g Downgraded by 1 level for risk of bias (mainly performance, detection and selection biases), 1 level for inconsistency (marked unexplainable difference of the effect of 1 study to that of the other studies) and 1 level for imprecision (low number of participants; 3 of the 4 studies found either no or only a small effect that was either reportedly non‐significant or had a CI including overlapping no effect) ^h Downgraded by 1 level for risk of bias (mainly performance, detection, selection biases), 1 level for inconsistency (difference in effect of the 3 studies ranging from a large effect favouring immobilisation in external rotation to no effect) and 1 level for imprecision (low number of outcome events; CIs of 3 of the 4 estimates overlapping no effect) ⁱ Assumed risk based on the median control risk of the 3 contributing studies ^j Downgraded by 2 levels for risk of bias (mainly performance, detection, selection and attrition biases) and 1 level for imprecision (low number of events; CIs for 2 of the 3 estimates overlapping no effect) ^k Downgraded by 2 levels for risk of bias (mainly performance, detection, selection and attrition biases), 1 level for imprecision (very low number of events) and 1 level for indirectness (suboptimal outcome measure) ^l Downgraded by 2 levels for risk of bias (mainly performance, detection, selection and attrition biases), 1 level for imprecision (very low number of outcome events and small study sample sizes; no CIs were reported) and 1 level for indirectness (poor definition and reporting of most adverse events)

Table 1. Key characteristics of the immobilisation treatment

Study	Start of immobilisation (post‐dislocation)	Duration of immobilisation	External rotation position	Type of braces and slings	Providers
Chan 2018	within 5 days	4 weeks	30° + 30° abduction	external rotation brace: Smartsling, Ossur, Reykjavik, Iceland); internal rotation sling: Polysling, Mölnlycke Health Care, Gothenburg, Sweden)	appropriately trained members of staff (surgeons, nurses or healthcare assistants)
Finestone 2009	no information	4 weeks	15° to 20°	external rotation brace: (manufacturer unspecified); internal rotation sling: traditional internal rotation sling	unclear
Heidari 2014	presumably on the same day (patients presented within 6 hours)	3 weeks	10° + 15° abduction	external rotation brace: stabiliser brace with adjustable angle of abduction (body: hard polyethylene); presumably commercially manufactured but purpose‐designed; internal rotation sling: sling and swathe bandage	unclear
Itoi 2007	within 2 days (termed day 1 to 3)	3 weeks	10°	external rotation brace: a) wire‐mesh splint covered with sponge and a stockinette (until October 2003); b) prototype brace, Alcare, Tokyo, Japan (from November 2003); internal rotation sling: sling and swathe	the treating surgeons
Liavaag 2011	within 24 hours	3 weeks	15°	external rotation brace: 15° UltraSling ER; DonJoy, Vista, California); internal rotation sling: normal collar and cuff device or sling and swathe	unclear
Taskoparan 2010	on the same day	3 weeks	10°	external rotation brace: "specific splint fixated in 10 degrees external rotation and adduction" (polyethylene/thermoplastic); internal rotation sling: 1st day: "valpaeu bandaging"; from 2nd day: "waist‐assisted sling"	unclear
Whelan 2014	presumably within 7 days (patients were assessed within 7 days)	4 weeks	0° to 5°	external rotation brace: DonJoy (Vista, California) external rotation shoulder brace; internal rotation sling: traditional internal rotation sling	certified orthopaedic technicians

Table 1. Key characteristics of the immobilisation treatment

Table 2. Lengths of follow‐up in the included studies

Study	Final follow‐up	Comments on follow‐up
Chan 2018	24 months	Set follow‐up times: 3 months; 1 & 2 years post‐dislocation
Finestone 2009	mean 33.4 months (range 24 to 48)	Set follow‐up times: 2 & 6 weeks; 3 & 6 months; 1, 2, 3 & 4 years post‐injury
Heidari 2014	24 months and 33 months	Set follow‐up times; 3 weeks post‐intervention; 24 months post‐dislocation (re‐dislocation); 33 months post‐dislocation (WOSI only)
Itoi 2007	mean 25.6 months (range 24 to 30)	Set follow‐up times: 6 months; 1 & 2 years (presumably post‐initiation of immobilisation)
Liavaag 2011	mean 29.1 months (range 24 to 54)	Set follow‐up times: 3 weeks (adherence data); 2 years post‐dislocation.
Taskoparan 2010	mean 21 months (range 6 to 41)	Set follow‐up times: 6 months (function scores); 1 & 2 years (radiographs and MRI) Not specified for re‐dislocation and adverse events Individual patient data presented with follow‐up ranging from 6 to 41 months
Whelan 2014	mean 25 months (range 12 to 43)	Set follow‐up times: 4 weeks and 3, 6, 12, 18 & 24 months post‐dislocation. However, results were presented for a minimum of 12 months

Table 2. Lengths of follow‐up in the included studies

Table 3. Re‐dislocation: data for different age categories

Subgroup	Study	Age category	External rotation group	Internal rotation group
Data available for planned subgroup analysis age ≤ 20 years versus age ≥ 21 years
age ≤ 20 years	Heidari 2014	≤ 20 years	0/0 (0%)	0/0 (0%)
	Taskoparan 2010	≤ 20 years	0/0 (0%)	0/1 (0%)
age ≥ 21 years	Heidari 2014	≥ 21 years	2/51 (4%)	17/51 (33%)
	Taskoparan 2010	≥ 21 years	1/16 (6%)	5/16 (31%)
Data available for planned subgroup analysis age ≤ 39 years versus age ≥ 40 years
age ≤ 39 years	Heidari 2014	≤ 40 years	2/42 (5%)	16/47 (34%)
	Liavaag 2011	≤ 40 years	28/91 (31%)	23/93 (25%)
	Taskoparan 2010	≤ 40 years	1/12 (8%)	5/14 (36%)
	Whelan 2014	≤ 35 years	6/27 (22%)	8/25 (32%)
age ≥ 40 years	Heidari 2014	≥ 41 years	0/9 (0%)	1/4 (25%)
	Taskoparan 2010	≤ 40 years	0/4 (0%)	0/3 (0%)
Other data, including subgroups
Other	Chan 2018	16 to 44 years	8/33 (24%)	10/33 (30%)
	Finestone 2009	17 to 27 years	19/27 (37%)	10/24 (42%)
	Heidari 2014	21 to 30 years	1/16 (6%)	3/18 (17%)
	Heidari 2014	31 to 40 years	1/26 4%)	13/29 (45%)
	Liavaag 2011	16 to 22 years	19/33 (58%)	13/30 (43%)
	Liavaag 2011	23 to 29 years	6/24 (25%)	7/27 (26%)
	Liavaag 2011	30 to 40 years	3/34 (9%)	3/36 (8%)
	Taskoparan 2010	15 to 75 years	individual participant data in study report
The table shows age‐related subgroup data for the predefined categories of interest (≤ 20 years versus ≥ 21 years; ≤ 39 years versus ≥ 40 years), as well as other subgroups or age ranges as reported in the trials. Actual thresholds varied as shown.

Table 3. Re‐dislocation: data for different age categories

Table 4. Any instability: data for different age categories from Itoi 2007

Subgroup (age)	External rotation group	Internal rotation group
≤ 20 years	11/27 (41%)	13/19 (68%)
≥ 21 years	11/58 (19%)	18/55 (33%)
≤ 40 years	19/62 (31%)	27/50 (52%)
≤ 41 years	3/23 (13%)	4/24 (17%)
The table shows age‐related subgroup data for the predefined categories of interest (≤ 20 years, ≥ 21 years, ≤ 39 years, ≥ 40 years)

Table 4. Any instability: data for different age categories from Itoi 2007

Comparison 1. Immobilisation in external versus internal rotation

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Re‐dislocation at ≥ 12 months follow‐up (nearest to 24 months accepted) Show forest plot	6	488	Risk Ratio (M‐H, Random, 95% CI)	0.67 [0.38, 1.19]

2 Re‐dislocation data at last follow‐up Show forest plot			Other data	No numeric data

3 Re‐dislocation, interim follow‐ups (3 and 12 months) Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

3.1 Re‐dislocation, 12 months follow‐up	1		Risk Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
3.2 Re‐dislocation, 3 months follow‐up	1		Risk Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
4 Re‐dislocation: stratified according to external rotation with / without abduction Show forest plot	6		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

4.1 External rotation without abduction	4	320	Risk Ratio (M‐H, Random, 95% CI)	0.94 [0.61, 1.44]
4.2 External rotation with abduction	2	168	Risk Ratio (M‐H, Random, 95% CI)	0.33 [0.04, 2.46]
5 Validated patient‐reported outcome measures for shoulder disability (OSI, WOSI) Show forest plot	2		Mean Difference (IV, Fixed, 95% CI)	Totals not selected

5.1 WOSI at 33 months follow‐up (total score)	1		Mean Difference (IV, Fixed, 95% CI)	0.0 [0.0, 0.0]
5.2 WOSI at mean 25 months follow‐up	1		Mean Difference (IV, Fixed, 95% CI)	0.0 [0.0, 0.0]
6 Validated PROMS: non‐parametric results Show forest plot			Other data	No numeric data

6.1 Final follow‐up			Other data	No numeric data
6.2 Interim follow‐up			Other data	No numeric data
7 Resumption of pre‐injury activities Show forest plot	3		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

7.1 Return to pre‐injury sports at 24 months	1		Risk Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
7.2 Return to sports at any level at 24 months	1		Risk Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
7.3 Return to pre‐injury sports activity level at 24 months	1		Risk Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
7.4 Return to pre‐injury level of activity of affected arm at ≥ 24 months	1		Risk Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
8 Any instability (subluxation or subjective instability, individually or grouped into composite outcome) Show forest plot	5		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

8.1 Recurrent instability	3	395	Risk Ratio (M‐H, Random, 95% CI)	0.84 [0.62, 1.14]
8.2 Positive apprehension test or re‐dislocation	2	135	Risk Ratio (M‐H, Random, 95% CI)	0.28 [0.14, 0.57]
9 Adverse events Show forest plot	7		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only

9.1 Transient and resolved adverse events	4	377	Risk Ratio (M‐H, Fixed, 95% CI)	2.73 [0.83, 9.02]
9.2 Important adverse events	3	268	Risk Ratio (M‐H, Fixed, 95% CI)	0.61 [0.08, 4.46]
10 Adherence to treatment Show forest plot	6		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected

11 Subsequent surgery Show forest plot	4	327	Risk Ratio (M‐H, Fixed, 95% CI)	0.76 [0.44, 1.30]

Comparison 1. Immobilisation in external versus internal rotation