The pathophysiology associated with primary (idiopathic) frozen shoulder: A systematic review

Background Frozen shoulder is a common yet poorly understood musculoskeletal condition, which for many, is associated with substantial and protracted morbidity. Understanding the pathology associated with this condition may help to improve management. To date this has not been presented in a systematic fashion. As such, the aim of this review was to summarise the pathological changes associated with this primary frozen shoulder. Methods Databases: Medline, Embase, CINAHL, AMED, BNI and the Cochrane Library, were searched from inception to 2nd May, 2014. To be included participants must not have undergone any prior intervention. Two reviewers independently conducted the; searches, screening, data extraction and assessment of Risk of Bias using the Cochrane Risk of Bias Assessment Tool for non-Randomised Studies of Interventions (ACROBAT-NRSI). Only English language publications reporting findings in humans were included. The findings were summarised in narrative format. Results Thirteen observational studies (involving 417 shoulders) were included in the review. Eight studies reported magnetic resonance imaging or arthrography findings and 5 recorded histological findings. When reported mean ages of the participants ranged from 40.0 to 59.8 years. Duration of symptoms ranged from 0 to 30 months. The majority of studies (n = 7) were assessed to be of moderate risk of bias, two studies at high risk and the remaining four were rated as low risk of bias. Study characteristics were poorly reported and there was widespread variety observed between studies in respect of data collection methods and inclusion criteria employed. Pathological changes in the anterior shoulder joint capsule and related structures were commonly reported. Imaging identified pathological changes occurring in the coracohumeral ligament, axillary fold and rotator interval. Obliteration of the subcoracoid fat triangle also appeared to be pathognomonic. Histological studies were inconclusive but suggested that immune, inflammatory and fibrotic changes where associated with primary frozen shoulder. Conclusions This systematic review presents a summary of what is currently known about the tissue pathophysiology of primary frozen shoulder. Further studies that use standardised inclusion and exclusion criteria and investigate changes in naïve tissue at different stages of the condition are required.


Background
Although frozen shoulder is considered to be a common musculoskeletal condition, with reviews reporting up to 5.3 % of the population being affected [1], definitive prevalence and incidence rates remain unknown [2]. The condition is associated with; (often severe) pain, sleep deprivation, anxiety, and disability that may be hugely disruptive and impacts on nearly every aspect of daily living [3]. The average duration of the condition is 30.1 months (range 1 to 3.5 years) [4] but it may be substantially longer [5,6], and the burden placed upon individuals and health care services may therefore be considered substantial [7].
The term "frozen shoulder" was introduced in 1934 by Codman who described the disorder as "difficult to define, difficult to treat and difficult to explain" [8]; and in many respects this remains true today. Frozen Shoulder (FS) has been classified into primary and secondary conditions [9]. Primary FS (PFS) is characterised by an insidious onset of idiopathic origin whereas secondary FS is associated with a defined event, such as a known intrinsic (such as rotator cuff disease) or extrinsic (such as trauma) cause [10]. FS associated with medical conditions such as diabetes and thyroid disorders are subcategorised as secondary systemic frozen shoulder [11].
Symptoms associated with frozen shoulder include: localised pain, pain with movement, night pain (rendering the patient unable to sleep on the affected side), marked limitation of active and passive range of movement (particularly external rotation) and normal shoulder radiograph findings [8]. However, the absence of definitive diagnostic criteria imposes challenges for clinical diagnosis and management and research [12]. This diagnostic challenge is further complicated by the clinical overlap in signs and symptoms between frozen shoulder and other conditions, such as; rotator cuff tendinopathy, calcific tendonitis or early glenohumeral arthrosis [13,14]. A recent narrative review suggested thickening of the coracohumeral ligament (CHL), joint capsule and synovium to be diagnostic features for frozen shoulder [15]. However no systematic review has yet collated the data from imaging studies to specify the intra and peri-articular changes that are associated with the condition.
Historically, the pathology of FS has been attributed to structures such as the subacromial bursa and joint capsule [16,17]. As arthroscopic and microbiological techniques have advanced other structures have been associated with the pathogenesis of the condition: namely, the rotator interval (RI), long head of biceps (LHB) and the CHL [18]. Contemporary histological analyses have identified the presence of inflammatory markers within the asscoiated tissue [19]. Cytokines, such as Tumour Necrosis Factor (TNF) α, Interleukin (IL) 1 α and β and IL-6 have also been identified [20]. In addition, studies have reported high numbers of fibroblasts and myofibroblasts, suggestive of a fibrotic process [21,22]. An immunological component has also been linked with frozen shoulder; such as the presence of Blymphocytes, mast cells and macrophages [23]. Such studies have led to the suggestion that FS may begin as an immunological response which escalates to an inflammatory synovitis, eventually leading to fibrosis of the capsule and that future research should focus on disease [15].
The purpose of this systematic review was to identify and synthesise the available evidence regarding the intra and peri-articular pathophysiology of primary frozen shoulder. A secondary aim was to identify deficits in our knowledge that may inform future research. The review was designed to include studies that had investigated the pathology, physiology, physiopathology, neurophysiology, histology, histocytochemistry, microbiology, immunochemistry or immunohistochemistry of the glenohumeral joint and its related structures in adults diagnosed with primary frozen shoulder.

Methods
This review is reported in accordance with the PRISMA statement for reporting systematic reviews [24].

Searches
Databases (Medline, Embase, CINAHL, AMED, BNI and the Cochrane Library) were searched from inception until 2nd May, 2014. Searches were performed independently by two researchers (HB and VR). The search strategy was developed using the Population and Intervention component of the PICO formula (Population, Intervention, Comparator and Outcome) [25]. Search terms related to patho-anatomical and pathophysiological changes associated with primary idiopathic frozen shoulder (Table 1). No language restrictions were applied and searches were limited to human studies. In addition to the formal data base searches a reference list search of included publications was also conducted.

Eligibility criteria
Studies were included if the participants were diagnosed as having PFS and had undergone combinations of imaging, histological or biochemical analysis of the glenohumeral joint. Studies were excluded if participants were diagnosed with any form of secondary frozen shoulder, such as diabetes, rotator cuff disease or trauma [12]. To reduce confounding the findings, studies were also excluded if participants had undergone previous interventions directly to the shoulder joint (and were termed non-naïve studies). This was because steroid injections may impact on the structure and biochemistry of the tissue [15,26]. Furthermore, arthrographic distension and capsular release are designed to disrupt the capsule [27] and manipulation under anaesthetic (MUA) may cause intra-articular damage to multiple structures [28]. Translation services were not available thus non English language studies, identified through the search, were subsequently excluded.

Selection of studies
Two reviewers (HB and VR) reviewed the articles for eligibility and inclusion with a third reviewer (JL) available in the event of consensus not being achieved. Article titles were used to identify relevant studies. Following this, eligibility was checked and recorded on a checklist designed for the review that incorporated PICO criteria. A data extraction form was developed for the review based upon the University of York, Centre for Reviews and Dissemination (2009) guidance [29].

Data analyses
Following data extraction, the study characteristics were tabulated and the studies synthesized. The variables synthesized in this review were reported findings from imaging studies of the shoulder joint and its related structures, as well as histological, neural and vascular findings. In addition, studies were assessed and their risk of bias appraised. Whether meta-analyses would be possible or appropriate was considered at this point.

Risk of bias
Although not always included in systematic reviews investigating pathophysiological mechanisms it was decided a priori to include an assessment of the risk of bias of the studies included in the current systematic review to enhance the validity of conclusion reached. The choice of a risk of bias tool for the review proved difficult as no one tool was perfectly compatible with this type of review. As the review question did not explore diagnostic accuracy the QUADAS-2 tool to evaluate the risk of bias and applicability of primary diagnostic accuracy studies was not considered appropriate. The ACROBAT-NRSI (A Cochrane Risk Of Bias Assessment Tool for Non-Randomized Studies of Interventions) is used when appraising the risk of bias in nonrandomized studies that compares the health effects of at least two interventions. Although the current review explored mechanisms rather than interventions, its domains appeared relevant and appropriate to the review and was chosen for use in the current review [30]. Studies were appraised to be at high, moderate or low risk of bias independently by two reviewers (HB and VR) with a third reviewer available in the event of any nonagreement (JL).

Results
Three thousand five hundred fifty-one potentially relevant studies were identified in searches. Title, abstract and reference list screening identified 58 articles meeting the review criteria. Duplicates (n = 16) were removed and the full text of articles read. Thirteen studies met the inclusion criteria for the review and 29 studies were excluded ( Table 2). A summary is provided in the PRISMA flow diagram (Fig. 1). The study characteristics are presented in Table 3. All 13 included studies were observational in design. Nine studies included a control group [19,[31][32][33][34][35][36][37][38], four did not [14,[39][40][41]. Of those using a control group, four included patients with rotator cuff pathology [31,32,35,37], three used asymptomatic controls [34,36,38] and two studies included patients with shoulder instability [19,33]. One study included two control groups [31], one with rotator cuff pathology and the other included people without symptoms. Study characteristics were generally poorly reported and there was widespread variation in diagnosis, methods of sample selection, timing of sample selection and presence of confounding variables such as use of oral medications. Eight out of 13 studies (62 %) based their inclusion criteria on the Codman classification [14,31,32,34,36,38,39,42]. However, it was evident that there were substantial variations in the interpretation of this classification ( Table 3).
The risk of bias data is presented in Table 4. The majority of studies (n = 7) were identified as having a moderate risk of bias, with two studies assessed of being at high risk of bias and the remaining four rated as low risk of bias. In general, sample sizes were small, ranging from one to seventy two (average = 28) participants. All studies used convenience sampling. Despite eleven studies identifying potential confounding factors, only six [14,31,[33][34][35]38] reported how they had taken account of them in their study design and/or in their analysis. The risk of bias data and widespread variation between studies did not permit meta-analyses.
Three studies used contrast enhancement arthrography, with two utilising magnetic resonance angiogram (MRA) [35,36], and the third, radiology [40]. Arthrography findings were contradictory (Table 3). Song et al. [36] reported substantial thickening of the joint capsule in the axillary recess and the RI. Neviaser [40] reported reduced joint capacity secondary to thickening and contracture of the capsule (region unspecified), obliteration of the axillary fold and often complete or near complete abolition of the subscapularis bursa. In contrast, Manton et al. [35] reported a trend for greater capsular thickness in the axillary recess and at the humeral head and increased synovial thickness in the axillary recess in controls, when compared to patients with FS. They also reported that RI abnormalities were more common in control participants, concluding that there are no useful MRA signs of FS.

Histology findings
Extensive histological findings were reported (Table 3). Tissue samples demonstrated the following: a dense collagen matrix and high population of fibroblasts and contractile myofibroblasts [19,21,33,41]; a fibrotic process limited to the anterior part of the capsule [41]; elevated levels of inflammatory cytokines in the SAB and anterior capsule [19] and the presence of mature and regenerating nerve fibres in the anterior capsule [37].
Five studies explored the histological and molecular changes associated with idiopathic FS (Table 5). When the study characteristics were reviewed limitations were evident. As previously identified, symptomology, demographics and the stage of the condition were poorly recorded. Secondly, there was substantial diversity between studies with regards to what was being measured. Furthermore, the techniques used to obtain the data also varied ( Table 3).

Neuronal and vascular findings
Xu et al. [37] investigated neuronal changes within the condition. They reported elevated levels of several immunoreactive neuronal proteins (GAP43, PGP9.5 and P75) in the anterosuperior joint capsule. The distribution of these proteins was either close to small blood vessels or within fibroblastic tissue. Increased vascularity was a common feature identified in the histology studies; particularly located in the anterosuperior structures but absent in the inferior structures (with the exception of the AF).

Summary of main findings
This review identified that the anterior shoulder structures in primary frozen shoulder were the location of greatest pathological change and in the subsequent

Clinical inclusion criteria
The review identified substantial variations in interpretation of the Codman classification. Future research must clearly detail defined and standardised diagnostic guidelines, to allow for more accurate and definitive comparisons between findings in studies.

Imaging
Imaging investigations varied substantially across the included trials and are a potential reason for the variations in findings. Three studies used arthrography, with two using direct arthrography, where contrast material was injected directly into the joint [35,40]. The basis for this is to permit a more precise visualisation of the intraarticular structures [43]. The contrast material was injected until the capsule distended which occurred at approximately 12-14 ml of fluid [44]. Neviaser [17] reported that normal shoulder joint capacity is between 28-35 ml, often reducing to 5-10 ml in cases of FS. Manton et al. [35] reported a tolerance of less than 10 ml in all nine people with FS. Although the significance of reduced joint capacity in the diagnosis of FS remains uncertain [18,[45][46][47], the effect of capsular distension when introducing the contrast material may have confounded the published findings relating to the capsule and synovium [35,47]. Song et al. [36] utilised indirect MRA, where contrast was injected intravenously into an antecubital vein. Indirect MRA requires exercising the joint for 10 to 15 min pre-imaging to increase vascular perfusion to improve flow into the joint [48], and again the influence of this activity on the reported findings is unknown.  There is no definitive guidance as to which imaging modality demonstrates greater diagnostic value in FS, and the heterogeneity of techniques used, and their associated potential confounding factors, limits deriving definitive conclusions relating to the articular and peri-articular changes associated with FS.

Histology
Symptomology, demographics and disease stage were poorly reported in the studies included in this review. The widespread diversity between studies with regards to what was being measured and how data was collected made comparison and synthesis of findings difficult. The main findings with respect to pathology identified in this review are presented in Tables 5 and 6 and are summarised below.

Fibrosis and contracture
Bunker [39] and Uhthoff and Boileau [41] used immunocytochemistry (ICC) and immunohistochemistry (IHC) to review matrix components. Both reported fibroblastic proliferation in the superior capsule and the RI. This is consistent with the imaging findings and with previous histological studies [49,50]. Vimentin is a cytocontractile protein and its presence may be assessed during ICC. Bunker [39] reported that vimentin was strongly expressed and confirmed that the cells were fibroblasts. In addition, when exposed to a smooth-muscle actin, many of the fibroblasts displayed a differentiation into a myofibroblastic phenotype. The myofibroblast, or contractile fibroblast, is the pathognomonic cell of contractile scar tissue and is found in Dupuytren's and the other fibromatoses [51]. Kilian et al. [33] used reverse transcription polymerase chain reaction (RTPCR) to study the mRNA (messenger RNA) transcription rates in the fibrosing stage of FS. They reported decreased levels of fibroblast like cells and α1 (III) chains which was indicative of a low number of myofibroblasts. The differing results may be due to samples being acquired at different stages of disease process; Bunker [39] did not supply information regarding stage of the condition or duration of symptoms since onset so comparison of results is challenging. Discrepancies in data may also relate to the way in which the tissue samples were managed. RTPCR         Carrillon et al. [32] Kilian et al. [33] Lho et al. [19] Li et al. [34] Manton et al. [35] Neviaser [40] Sofka et al. [14] Song et al. [36] Uhthoff & Boileau [41] Xu et al. [37] Zhao et al.  [52]. Although they had a relatively small number of participants (N = 4) Uhthoff and Boileau [41] conducted a comprehensive study to determine if fibroplasia affects all structures equally. Samples were taken anteriorly, posteriorly, superiorly and inferiorly around the shoulder joint. All structures demonstrated fibroplasia, however, vimentin was strongly expressed anteriorly but was absent in the posterior capsule, leading the authors to suggest that fibroplasia and contracture may be different processes. Their cohort consisted of 4 female subjects with no information pertaining to stage of the condition.
To reduce confounding variables the direct local introduction of medication into the joint was an exclusion criteria for the current review. All patients had failed conservative management; but no studies specified what this included. Common conservative management strategies for FS include oral analgesics and NSAIDs [53]. Therefore, the systemic effects of oral medications should be considered. Evidence in both bone and tendon literature suggests that ibuprofen reduces tensile strength, collagen fibre organisation and fibroblastic proliferation [54]. Almekinders et al. [55] conducted an in-vitro study of the effects of indomethacin on injured human tendon tissue. They reported diminished levels of fibroblast DNA synthesis in the groups treated with

Inflammation and immune modulation
Cyclooxygenases play an important role in inflammation and the collagen catabolic process within peripheral tissues [53]. Lho et al. [19] used RTPCR and IHC and reported increased expression of COX1 in the endothelial cells and stroma of the joint capsule and increased expression of COX2 in the capsule and subacromial bursa of the FS group. Furthermore, levels of IL-1α, IL-β, IL6 and TNF-α also differed between the capsule, bursa and joint fluid. Interleukin and TNF-α are pro-inflammatory cytokines; released from immune cells such as macrophages [54]. This may imply that high numbers of these cells may be present in the joint [56]. Bunker [39] reported low numbers of macrophages and leukocytes in the RI. This variation may be reflective of differing pathological processes between structures and/or that the biopsies were taken from different stages of disease, and possibly different diagnostic criteria. Neither Bunker [39] nor Lho et al. [19] provided sufficient background data regarding their participants to explore this. Furthermore, no comparable studies were included in this SR which again reveals a gap in the evidence base that is worthy of exploration.

Neuronal and vascular factors
Pain is associated with FS [3]. Hand et al. [6] conducted a longitudinal study of 223 patients with frozen shoulder with a mean follow up time of 4.4 years (range 2 to 20 years), with 41 % of the patient's reporting mild to moderate pain and 6 % reported severe pain. To date, few studies have investigated the causes of pain experienced by patients with FS [57]. In this review, Xu et al. [37] investigated neuronal components associated with PFS, reporting elevated levels of several immunoreactive neuronal proteins (GAP43, PGP9.5 and P75) in the anterosuperior joint capsule, close to small blood vessels or within fibroblastic tissue. These findings confirm the presence of mature and regenerating nerve fibres in the anterosuperior capsule and may explain the severe pain experienced by sufferers of the condition in the early stages (less than six months). Increased vascularity was a common feature identified in the histology studies; particularly located in the anterosuperior structures but absent in the inferior structures (with the exception of the axillary fold (AF)). This is consistent with the literature where hypervascularity and angiogenesis have been reported as potential sources of pain due to their  association with neovessels [58,59]. Xu et al. [37] reported stronger expression of CD34 (a haematopoietic cell marker) in the superior joint capsule of the FS group compared RC tears, as a control population. Limited conclusions may be drawn from this study because of the small sample size (n = 8). Ryu et al. [58] investigated FS in a diabetic cohort and reported CD34 to be strongly expressed. However, caution must be taken when extrapolating these results as the patient's had received corticosteroid injections. A recent study by Okuno et al. [59] reported that arterial embolization of neovessels in the RI provided rapid relief of pain in their FS group. Limited knowledge exists regarding the pain mechanisms involved with FS [60]. This SR has provided some insight into possible causes. This knowledge has great significance for clinicians as pain is often the dominant complaint in patients with FS. The literature has suggested that the condition may manifest differently between individuals. A greater understanding would greatly assist clinicians to effectively manage this symptom in their patients. It is clear further research is required.

Limitations
It is acknowledged that this systematic review has a number of limitations. These are reviewed in the following section.

Searches
Only English language publications were included in this review so the introduction of language bias cannot be ruled out. Studies where an English translation could not be sourced were identified during abstract analysis. One reviewer (VR) identified eighteen studies where the full text English article could not be sourced. The second reviewer (HB) identified six of those eighteen. All eighteen studies were excluded. The evidence surrounding language bias is conflicting [30], and it is not known how these non-English publications may have influenced the findings of the current review.
No date restriction was applied to the studies so that all available studies could be identified and included in this systematic review, believed to be the first of this type, in this condition. MRI was first introduced into healthcare in the 1980's, and over time image quality has advanced substantially [61]. The implication of this is that the reported imaging findings from the earlier studies [32,35] may lack the sensitivity of those in later studies [31]. This also may have influenced the findings and contributed to reported discrepancies.

Inclusion and exclusion criteria
The aim of the review was to investigate the intra and peri-articular pathophysiology of the glenohumeral joint in people diagnosed with primary idiopathic frozen shoulder. Only studies specifying primary frozen shoulder were included as it was not possible to separate the findings from investigations that included both primary and secondary frozen shoulders. This meant that primary frozen shoulder findings may have been missed by excluding studies that incorporated both. The decision to only include samples from people diagnosed with primary FS hopefully generated more homogenised data.
Direct injection of medication into the joint was an exclusion criteria in this review to reduce the potential confounding influencing this may have had on the findings. However, there may be other sources of confounding which might have affected the findings of this review. All patients included in the review had had failed conservative management but none of the studies specified what this included. Common conservative management strategies for frozen shoulder include oral analgesics and NSAIDs [62] and potentially, the systemic effects of such oral medications may have also influenced findings.

Widespread variation
The main limitation of this review relates to the included studies. Variations in diagnosis, methods of sample selection, timing of sampling, and confounding variables such as use of oral medications, all may have influenced the reported findings and the conclusions of this review. Meta-analysis was not considered due to the considerable and widespread variance within the included studies [30].

Risk of bias
The majority of studies (n = 7) were identified as having a moderate risk of bias, with two studies assessed of being at high risk of bias and the remaining four rated as low risk of bias. Study characteristics were poorly reported. There are three possible concerns for this review. The first is, as previously mentioned, a risk of bias tool specifically for use in pathophysiology reviews was not found, meaning there may be specific domains relevant for this type of review which have not been appraised or discussed. The second is that, with only a minority of studies being assessed as low risk of bias, the findings of this review may contain systematic bias [30]. Metaanalyses were not included in this review: it is accepted that meta-analyses of studies that are at risk of bias may be seriously misleading since meta-analysis may simply compound the errors, thus producing an erroneous results which may be interpreted as having more credibility [30]. The third concern is that the ACROBAT-NRSI tool, as a recent development to meet the need for a tool to assess risk of bias in non-randomised studies, has yet to be widely used or evaluated. Further research on the performance of the tool in the future may influence the findings of this review or enable the findings to be placed more appropriately in context.

Conclusions
This systematic review is the first review to synthesise imaging and histological studies to examine the pathophysiology associated with primary frozen shoulder. The review highlights the role of the anterior shoulder structures in primary frozen shoulder, but there is a lack of available evidence considered at low risk of bias to inform understanding of the pathophysiology of the primary frozen shoulder condition. Consensus regarding inclusion criteria (and the interpretation of the Codman classification criteria) is first required for future research to promote studies providing comparable findings. Following this, further studies that identify findings at clearly defined stages of the condition are required to improve the understanding of the disease continuum. Improved understanding may then inform management specific to each stage of this painful, disabling common condition so that it is no longer difficult to define, treat or explain.

Funding
The Health Foundation (http://www.health.org.uk/) is gratefully acknowledged for providing funding to support this project.

Availability of data and materials
All the data supporting the findings of this Systematic Review are contained within the manuscript.
Authors' contributions HB: Conceived the study, identified the research question, identified the inclusion and exclusion criteria, completed a systematic literature search, identified articles for abstract analysis, completed abstract analysis, completed data extraction and contributed to writing up the findings. VR: Conceived the study, identified the research question, identified the inclusion and exclusion criteria, completed a systematic literature search, identified articles for abstract analysis, completed abstract analysis, completed data extraction and contributed to writing up the findings. CML: Contributed to the design of this review and its processes, and contributed to the writing up, revising and editing of this manuscript. JL: Conceived the study, contributed to the identification of the research question and the inclusion and exclusion criteria, acted as a third reviewer and contributed to write up of the findings and editing the manuscript. All authors read and approved the final manuscript.