Development of a Multidomain Conceptual Framework for Frozen Shoulder: A Systematic Review Integrating Clinical, Biological, Psychological and Lifestyle-Related Contributors Authors: Santiago Navarro-Ledesma, José Javier Pérez-Montilla, Dina Hamed-Hamed, Fabrizio Brindisino, Filip Struyf Published: 2026 ABSTRACT Background/Objectives: Frozen shoulder (FS) is increasingly recognised as a heterogeneous condition in which clinical presentation, metabolic/endocrine burden, imaging phenotype, psychological factors, lifestyle-related contributors, and inflammatory biology may all contribute to disease expression and outcome. However, these contributors have not previously been integrated within a unified multidomain framework for whole-person phenotyping. To develop an evidence-derived multidomain conceptual framework for FS by integrating clinical, biological, psychological, and lifestyle-related contributors into a clinically interpretable structure. Methods: A systematic review with quantitative evidence synthesis was conducted following methodological principles for systematic reviews and reported in accordance with PRISMA 2020. The protocol was registered in PROSPERO (CRD420251158406). Observational studies relevant to predefined FS domains were identified, classified according to analytic contribution, and synthesised using a structured evidence-weighted approach. Where quantitative pooling was methodologically appropriate, meta-analytic synthesis was undertaken; where heterogeneity precluded conventional pooling, evidence was integrated through structured domain-level quantitative synthesis. Within-sphere and cross-sphere weights were derived from directness, consistency, prognostic or phenotyping relevance, methodological robustness, risk-of-bias profile, and domain specificity, and should be interpreted as exploratory evidence-derived estimates. Results: The final multidomain corpus comprised 63 studies, including 39 allocated to the clinical sphere, 5 to the metabolic/endocrine sphere, 10 to the psychological sphere, 8 to the imaging sphere, and 1 to the inflammatory sphere as direct but lower-confidence support. The final exploratory framework retained five spheres: clinical, metabolic/endocrine, imaging, psychological, and inflammatory. Within-sphere weighting identified baseline stiffness/range-of-motion restriction and symptom duration as dominant clinical components, diabetes-related burden and glycaemic control as dominant metabolic/endocrine components, capsular structural biomarkers as dominant imaging components, distress/mood together with sleep-related burden as dominant psychological components, and IL-1β-related biomarker and susceptibility evidence as the principal retained inflammatory signal. Provisional cross-sphere weighting yielded global contributions of 24.0% for the metabolic/endocrine sphere, 22.9% for the imaging sphere, 21.1% for the clinical sphere, 20.0% for the psychological sphere, and 12.0% for the inflammatory sphere. Conclusions: FS appears to be more appropriately represented as a weighted multidomain condition than purely as a clinical stiffness syndrome. The identified domains should not be interpreted as isolated entities, but rather as interacting and overlapping contributors within a whole-person clinical presentation. The proposed framework should be interpreted as an initial conceptual platform for future validation and personalised treatment-allocation research rather than as a clinically validated decision-support tool. 1. Introduction Frozen shoulder (FS) is traditionally characterised by spontaneous or progressive shoulder pain, marked limitation of glenohumeral motion, and substantial disability [1,2]. Although it is often approached clinically as a recognisable stiffness syndrome, its real-world presentation is clinically and biologically heterogeneous [3]. Patients differ not only in pain intensity, irritability, and degree of mobility loss, but also in endocrine–metabolic comorbidity, psychological burden, sleep disturbance, imaging phenotype, lifestyle-related factors, and response to treatment [4]. Therefore, a purely symptom-based, range-of-motion-based, or localised capsular interpretation is unlikely to capture the full complexity of FS, in which biological, metabolic, behavioural, psychological, inflammatory, and lifestyle-related factors may interact in shaping clinical presentation and recovery trajectory [3,4,5]. Epidemiologically, FS represents a relevant clinical burden. Most contemporary summaries estimate its frequency in the general population at approximately 2–5%, with predominance in women and individuals between 40 and 70 years of age [6,7], although lower estimates have been reported in some primary-care datasets depending on case definition and sampled population [8]. The epidemiological signal is stronger in systemic-risk groups, as FS is substantially more frequent in individuals with diabetes mellitus (DM) and thyroid disease, with reported prevalence estimates of approximately 10–38% in these populations [9,10]. This supports the view that FS should not be understood solely as an isolated local shoulder disorder. This multidomain heterogeneity has important practical consequences. Treatment selection is still frequently guided by broad tissue irritability categories rather than by a structured estimate of which clinical or non-clinical domains are most relevant in each patient [5,11]. Consequently, two patients with apparently similar range-of-motion restrictions may differ substantially in the degree to which their presentation is influenced by metabolic burden, tissue-level features, psychological distress, sleep-related symptom amplification, lifestyle-related factors, movement avoidance, or active inflammatory biology [12,13,14,15]. This creates a gap between the multidimensional nature of FS and the relatively non-quantified way in which treatment is commonly prioritised [15]. Several lines of evidence support a multidomain interpretation of FS. Clinical studies have shown that baseline stiffness, symptom duration, pain intensity, and disease stage are associated with clinical trajectory and recovery [16]. Endocrine–metabolic comorbidities, particularly DM and poorer glycaemic control, have been associated with recurrence, poorer treatment response, and less favourable long-term outcome [9,14,15]. MRI and ultrasound studies have identified structural capsular abnormalities, edema-related activity signs, and vascular imaging features with potential relevance for phenotype, stage, or prognosis [17,18]. Psychological distress, pain-related cognitive factors, psychosocial quality of life, and sleep-related burden appear to modulate symptom severity and perceived recovery [19,20,21]. Finally, inflammatory and fibro-inflammatory mechanisms have been implicated in FS pathobiology through cytokine activity, matrix remodelling, and capsular fibrosis [22]. Previous multimodal and biopsychosocial approaches have helped move the field beyond a purely local capsular model, but important methodological gaps remain. Existing models recognise interactions between clinical presentation, systemic biological factors, imaging findings, psychological burden, and inflammatory mechanisms; however, these contributors have rarely been organised into a structured framework that can be applied at patient level. Sleep-related burden remains underrepresented, imaging evidence is often interpreted through single-modality perspectives, and the relative contribution of each domain has not been clearly operationalised within a unified model [4,14,15,23]. Therefore, published work has not yet fully progressed from multidomain description to a framework capable of supporting sphere-level interpretation and future personalised treatment-allocation research [4,16]. The present study was designed to address this gap through two linked but distinct phases. First, we conducted a systematic review with quantitative evidence synthesis to identify and organise evidence related to clinical, metabolic/endocrine, imaging, psychological, inflammatory, and lifestyle-related contributors to FS. Second, we used this evidence base to develop an exploratory multidomain conceptual framework and research-use operational prototype for patient-level phenotyping. The aim was not to provide a validated clinical decision tool, but to create an evidence-derived structure capable of organising heterogeneous research into clinically interpretable domains and estimating their provisional relative contributions. Accordingly, the framework is intended to support whole-person reasoning and to provide a first research platform for future validation, recalibration, and personalised treatment-allocation studies. 2. Methods 2.1. Study Design and Reporting Framework This study was designed as a systematic review with quantitative evidence synthesis followed by exploratory multidomain conceptual framework development. The review was conducted following methodological principles for systematic reviews, reported in accordance with PRISMA 2020 [24], and prospectively registered in PROSPERO (registration ID: CRD420251158406). Methodological principles relevant to observational prognostic research were used to guide evidence interpretation, while STROBE/TRIPOD-aligned logic informed the framework-development and prototype-operationalisation components. The work comprised two linked but distinct phases. Phase 1 consisted of systematic evidence identification, study selection, data extraction, risk-of-bias appraisal, and domain-level evidence synthesis. Phase 2 consisted of conceptual framework development, in which the evidence identified in Phase 1 was organised into clinically interpretable domains and translated into a research-use operational prototype. This separation was introduced to distinguish conclusions directly supported by the reviewed literature from the interpretative framework-development process. Quantitative pooling was undertaken only when the retained studies were sufficiently comparable in design, population, predictor, outcome, and effect metric to permit meta-analytic synthesis. When conventional pooling was not methodologically appropriate, evidence was integrated using structured domain-level quantitative synthesis. Because the primary objective was multidomain framework development rather than estimation of a single pooled effect or validation of a prediction model, all domain weights and prototype scores were treated as exploratory and provisional. No internal validation, external validation, calibration analysis, reproducibility testing, or predictive-performance analysis of the prototype was performed in the present study. 2.2. Search Strategy and Source Selection The literature base was assembled through structured domain-focused searches designed to identify studies relevant to predefined FS domains: clinical presentation and prognosis, metabolic/endocrine burden, imaging biomarkers, psychological constructs, sleep-related burden, and inflammatory biology. Because the project involved framework development, domain-specific searching was refined when conceptual gaps became apparent during synthesis, particularly in relation to sleep-related burden and the expansion of the imaging domain from an initial ultrasound-centred formulation to a combined MRI-plus-ultrasound domain. Comprehensive literature searches were performed in PubMed, Web of Science, and ScienceDirect on 12 January 2026. Searches were limited to studies published within the previous ten years, written in English, involving human adults, and available as open access full text. The search strategy relied primarily on free-text terms combined with Boolean operators and applied to title/abstract or topic fields depending on database functionality. The complete search strategies, fields, and filters used for each database are reported in Table 1. 2.3. Study Selection and Eligibility Criteria Two reviewers (S.N.L. and J.J.P.M.) independently screened titles and abstracts, followed by full-text assessment of potentially eligible reports. Disagreements were resolved through discussion and consensus. Studies were eligible when they investigated frozen shoulder or adhesive capsulitis in adults and reported extractable quantitative data relevant to at least one predefined domain of interest. Eligible designs included prospective and retrospective cohorts, observational prognostic studies, case–control studies, cross-sectional studies, imaging studies, biomarker studies, and non-randomised interventional studies when they provided extractable phenotype-, prognosis-, treatment-response-, or domain-relevant data. Randomised trials, reviews, editorials, conference abstracts without sufficient quantitative detail, single-patient case reports, and purely narrative publications were not included in the core weighting framework. Selected non-core studies could be retained only when they provided contextual biological or conceptual support and were explicitly labelled as non-core evidence. Studies were excluded when they did not investigate FS specifically, lacked extractable quantitative data relevant to the predefined domains, or focused on structurally non-comparable clinical scenarios without sufficient interpretability for the framework. 2.4. Analytic Classification After full-text review, studies were classified according to their analytic contribution. The full corpus comprised all studies mapped to one of the final domains. A smaller core analytic subset was then identified to inform direct weighting when the study design, population, predictor, and outcome were considered sufficiently relevant and interpretable for domain-level synthesis. Studies that were clinically informative but less directly comparable were retained for supportive, sensitivity, or contextual interpretation rather than weighted equivalently with the core analytic subset. This two-step classification was used to reduce structural heterogeneity and avoid pooling or weighting evidence with substantially different analytic meaning. 2.5. Domain Definition and Predictor Reassignment The final framework retained five domains: clinical, metabolic/endocrine, imaging, psychological, and inflammatory. Sleep was not retained as an independent sixth domain; after targeted review of the FS sleep literature, sleep-related burden was incorporated as a formal subdomain within the psychological domain because of its conceptual overlap with distress, pain-related burden, irritability, and psychosocial functioning. MRI and ultrasound were integrated into a single imaging domain because the retained evidence converged on related structural, vascular, and activity-related tissue features rather than on independent modality-specific constructs. To reduce double counting, predictors were assigned according to their principal biological or clinical meaning rather than according to the nominal focus of the source paper. For example, diabetes mellitus, HbA1c, thyroid disease, dyslipidaemia, and liver-associated metabolic markers were assigned to the metabolic/endocrine domain even when reported in clinically oriented studies. Conversely, baseline pain, symptom duration, stage/irritability, disability, range-of-motion restriction, and autonomic symptoms were retained within the clinical domain. Domain assignment and predictor reassignment rules are provided in Supplementary File S2. 2.6. Data Extraction Data extraction was performed at the study and domain levels. Study-level extraction captured design, population characteristics, sample size, FS subtype or treatment context, follow-up duration, outcomes, candidate predictors, and analytic role within the framework. Domain-level extraction captured predictor identity, linked outcome, direction of association, effect metric, adjustment status, prognostic versus cross-sectional relevance, methodological interpretability, and relevance to the target domain. During prototype operationalisation, an additional extraction layer identified candidate patient-level inputs, preferred scales, severity direction, standardisation rules, and missing-data handling strategies. Extraction was performed by J.J.P.M. and independently checked by S.N.L.; discrepancies were resolved by consensus. 2.7. Risk of Bias and Methodological Quality Risk of bias and methodological quality were assessed using design-specific tools. ROBINS-I was used for non-randomised interventional studies where applicable; QUIPS for prognostic studies; NIH Study Quality Assessment Tools and Joanna Briggs Institute tools for observational, cross-sectional, imaging, biomarker, and genetic-association studies; and JBI criteria for case-level contextual evidence when applicable. Because the retained literature was heterogeneous across domains, risk of bias was not applied as a universal exclusion threshold. Instead, it informed confidence in domain interpretation, subdomain weighting, and cross-domain weighting. Detailed risk-of-bias and methodological quality assessments are provided in Supplementary File S1. 2.8. Evidence Synthesis and Weighting Procedure Each domain was synthesised using a structured evidence-weighted approach. Priority was given to longitudinal adjusted prognostic evidence, followed by longitudinal unadjusted evidence, treatment-response or phenotype-relevant observational evidence, and cross-sectional or subgroup evidence. Within each domain, predictors were grouped into clinically meaningful subdomains. The provisional weighting process considered the same criteria across domains: directness of evidence, consistency across retained studies, prognostic or phenotyping relevance, methodological robustness, risk-of-bias profile, specificity of the construct to the domain, and operational interpretability. The weighting procedure was not a data-driven coefficient estimation from a single empirical model. Rather, it was a structured reviewer-judgement process applied consistently across domains. For each domain, subdomains were ranked and assigned provisional internal weights according to the strength and interpretability of the retained evidence, and weights were normalised so that the internal subdomains summed to 100% within each domain. Cross-domain weights were then assigned using the same evidence-appraisal principles and normalised to 100% across the five domains. The internal weighting structure is provided in Supplementary File S2, and the main quantitative signals supporting weighting decisions are summarised in Supplementary File S3. 2.9. Prototype Operational Scoring Model To translate the conceptual framework into a research-use operational prototype, each domain was expressed as a weighted combination of standardised subdomain inputs. The provisional multidomain equation was defined as: FS multidomain burden score = 0.211 (C) + 0.240 (M) + 0.229 (Im) + 0.200 (P) + 0.120 (Inf), where C represents the clinical domain, M the metabolic/endocrine domain, Im the imaging domain, P the psychological domain, and Inf the inflammatory domain. These coefficients correspond to the provisional cross-domain weights derived from the structured evidence-weighting procedure. One principal operational input was selected for each construct whenever possible to preserve interpretability. In the clinical domain, movement restriction was represented by external rotation, forward flexion, and abduction deficits; disability by SPADI total; and pain by a 0–10 VAS/NRS measure. In the metabolic/endocrine domain, diabetes mellitus and HbA1c were selected as principal drivers. In the imaging domain, structural capsular biomarkers and activity-related imaging signs were prioritised. In the psychological domain, distress/mood and sleep-related burden were operationalised using HADS- and PSQI-based constructs. In the inflammatory domain, IL-1β-related biomarker activity and inflammatory susceptibility were retained as provisional inputs. The complete operational matrix is provided in Supplementary File S4. All operational inputs were standardised to a 0–100 scale before weighting. Binary variables were coded as 0 or 100, continuous variables were linearly rescaled to predefined ranges, and ordinal variables were mapped to equally spaced 0–100 categories. For range-of-motion variables, severity was defined using movement deficit rather than raw mobility values so that higher scores consistently represented greater burden. This standardisation was used to permit prototype calculation and should not be interpreted as evidence that all relationships between predictors and disease burden are linear. 2.10. Handling of Missing Data and Sensitivity Specification When one input was missing within a subdomain, available inputs in that subdomain were proportionally reweighted. If an entire subdomain was unavailable, the remaining subdomains within the corresponding domain were rescaled to sum to 1, with reduced-confidence interpretation. Because the inflammatory domain was supported by substantially less direct retained evidence and is less routinely measurable in clinical settings, a four-domain sensitivity specification was also calculated by omitting the inflammatory domain and renormalising the remaining clinical, metabolic/endocrine, imaging, and psychological weights to 100%. 2.11. Worked Example A realistic fictional patient profile was used to demonstrate implementation of the prototype. This worked example was included only to illustrate the calculation, operational feasibility, and presentation of a domain-specific profile. It was not used as validation evidence, and the resulting numerical values were not interpreted as having established diagnostic, prognostic, or treatment-allocation meaning. 3. Results 3.1. Search and Study Selection The study selection process and reasons for exclusion are shown in the PRISMA flow diagram (Figure 1). The database search identified 1262 records across PubMed, Web of Science, and ScienceDirect. After filters were applied, 316 records remained: PubMed (n = 74), Web of Science (n = 209), and ScienceDirect (n = 33). Following the removal of 72 duplicates, 244 records were screened by title and abstract. Of these, 181 were excluded, leaving 63 studies for inclusion in the review. The final multidomain corpus comprised 63 studies including 8648 participants. Thirty-nine studies were allocated to the clinical domain, five to the metabolic/endocrine domain, ten to the psychological domain, eight to the imaging domain, and one to the inflammatory domain. The core analytic subset directly informing weighting was smaller and comprised 18 clinical studies, four metabolic/endocrine studies, six psychological studies, and seven imaging studies. The inflammatory domain was retained as a distinct but lower-confidence domain, supported by one direct case-control biomarker/genetic-susceptibility study and contextual biological evidence rather than by a comparably broad observational core. This imbalance in evidence volume and certainty was considered when interpreting the weighting structure. 3.2. Study Characteristics Study characteristics by domain are summarised in Table 2. The clinical domain was the largest and most heterogeneous, including prospective and retrospective cohorts, comparative observational studies, non-randomised intervention cohorts, and prognostic studies evaluating pain, range of motion, disability, symptom duration, treatment response, recurrence, or recovery trajectory. The metabolic/endocrine domain included studies addressing diabetes mellitus, glycaemic control, thyroid/endocrine variables, dyslipidaemia, and liver-associated metabolic markers. The imaging domain included MRI- and ultrasound-based studies evaluating capsular thickness, coracohumeral ligament abnormalities, axillary recess findings, edoema or hyperintensity, enhancement, and vascular flow. The psychological domain included studies addressing anxiety, depression, psychosocial quality of life, self-efficacy, kinesiophobia, and sleep-related burden. The inflammatory domain included limited direct biomarker and susceptibility evidence. 3.3. Risk of Bias and Methodological Quality Risk-of-bias assessment showed that the retained evidence was methodologically heterogeneous and largely observational, with most studies showing at least moderate concerns. In non-randomised interventional studies assessed with ROBINS-I, common sources of bias included confounding, absence of randomisation, selection of specific patient populations, retrospective design, and incomplete use of appropriate control groups. Prognostic studies assessed with QUIPS frequently showed moderate-to-high concerns related to insufficient adjustment for confounders, heterogeneity in outcome definitions, and variability in statistical analysis and reporting. Studies evaluated with NIH and JBI tools generally showed moderate methodological quality, with recurrent limitations including limited confounding control, lack of longitudinal follow-up, and restricted ability to support causal inference. Detailed assessments are provided in Supplementary File S1. At domain level, the clinical evidence base was the largest but also the most heterogeneous. The metabolic/endocrine and imaging domains were supported by smaller but more conceptually coherent evidence sets, although limitations remained in external generalisability, confounding control, imaging protocol standardisation, and biomarker assessment. The psychological domain relied substantially on self-reported constructs, which may increase measurement overlap between pain, disability, sleep disturbance, and psychological burden. The inflammatory domain carried the greatest interpretive uncertainty because the retained direct evidence was limited to a single case–control study, despite biological plausibility and relevance to FS pathophysiology. 3.4. Final Multidomain Architecture The final multidomain architecture retained five domains: clinical, metabolic/endocrine, imaging, psychological, and inflammatory (Figure 2). Sleep-related burden was nested within the psychological domain because the retained evidence indicated overlap with distress, irritability, pain-related burden, and psychosocial functioning. MRI and ultrasound evidence were integrated into a single imaging domain because the studies converged on related structural and activity-related tissue features rather than on independent modality-specific constructs. Domain assignment and predictor reassignment rules are summarised in Supplementary File S2. 3.5. Domain-Level Evidence Synthesis Within the clinical domain, the strongest recurrent signals were baseline stiffness/range-of-motion restriction and symptom duration. Baseline pain, stage/irritability, disability/function, autonomic symptoms, and contextual variables provided additional but smaller signals. The clinical evidence suggested that pain and mobility may follow different temporal trajectories, with pain often improving earlier and range-of-motion gains continuing over later follow-up windows. Within the metabolic/endocrine domain, diabetes-related burden and glycaemic control were the dominant signals. Diabetes mellitus repeatedly emerged as an adverse factor associated with poorer treatment response, recurrence or reintervention, and less favourable long-term outcome. HbA1c and glucose-related measures suggested that glycaemic severity may provide information beyond binary diabetes status. Thyroid/endocrine-axis variables, lipid metabolism, and liver-associated metabolic markers contributed secondary exploratory signals. Within the imaging domain, capsular structural biomarkers were the strongest internal component, including coracohumeral ligament thickening, inferior glenohumeral ligament-related findings, axillary recess capsule thickness, and ultrasound-defined capsular thickening patterns. Edoema-, hyperintensity-, and enhancement-related activity signs contributed a second layer, while microvascular imaging and compensatory sonographic features provided additional supportive signals. Within the psychological domain, distress/mood was the strongest internal component, followed by sleep-related burden. Psychosocial quality of life and recovery-interference constructs, including self-efficacy and kinesiophobia, provided additional support. This structure indicates that the psychological domain extends beyond anxiety and depression alone and includes sleep-related and recovery-related constructs relevant to FS burden. The inflammatory domain was retained because the included case–control evidence reported higher serum IL-1beta levels in patients with primary FS than in controls and an association between IL-1beta polymorphism and susceptibility. However, the breadth of direct retained evidence was substantially weaker than for the other domains. Therefore, the inflammatory domain was treated as biologically relevant but quantitatively provisional and lower confidence. The provisional within-domain weighting structure is provided in Supplementary File S2. The main quantitative signals supporting domain interpretation and weighting are summarised in Supplementary File S3. 3.6. Provisional Cross-Domain Weighting Cross-domain normalisation yielded provisional global weights of 24.0% for the metabolic/endocrine domain, 22.9% for the imaging domain, 21.1% for the clinical domain, 20.0% for the psychological domain, and 12.0% for the inflammatory domain. These values should be interpreted as evidence-derived exploratory weights rather than validated predictive coefficients. The relatively high clinical evidence volume did not translate into a dominant clinical weight because methodological heterogeneity, predictor overlap, and reassignment of endocrine–metabolic variables reduced its relative specificity. Conversely, the inflammatory domain was weighted more conservatively because its direct retained clinical evidence was limited. 3.7. Sensitivity Specification Excluding the Inflammatory Domain Because the inflammatory domain was lower confidence and less operationally mature, a four-domain sensitivity specification was calculated by omitting the inflammatory domain and renormalising the remaining weights. In this sensitivity specification, the renormalised weights were 27.3% for the metabolic/endocrine domain, 26.0% for the imaging domain, 24.0% for the clinical domain, and 22.7% for the psychological domain. Applied to the worked example, the total multidomain score changed from 59.8/100 in the five-domain model to 60.7/100 in the four-domain sensitivity specification. This analysis was included to illustrate numerical stability of the worked example under omission of the least certain domain, not to validate the model. 3.8. Prototype Operationalisation A prototype operational scoring matrix was developed to convert the conceptual framework into measurable patient-level inputs. For each domain and subdomain, the matrix defined the preferred operational variable, preferred scale or format, direction of severity, standardisation rule to a 0–100 metric, and missing-data handling strategy. The principal operational inputs are summarised in Supplementary File S4. Within the clinical domain, movement restriction was operationalised using external rotation, forward flexion, and abduction deficits. Symptom duration, baseline pain, SPADI total, stage/irritability, autonomic symptoms, and contextual variables were retained as complementary clinical inputs. Within the metabolic/endocrine domain, diabetes mellitus and HbA1c were selected as principal operational drivers, with additional modifiers for thyroid dysfunction, lipid burden, and liver-associated metabolic abnormalities. Within the imaging domain, structural capsular biomarkers and activity-related imaging signs were prioritised. Within the psychological domain, HADS and PSQI were selected as the principal operational measures for distress and sleep, respectively. Within the inflammatory domain, IL-1beta-related biomarker activity and inflammatory susceptibility were retained as provisional inputs, although this pathway was considered lower confidence and less operationally mature because the retained direct evidence was limited and biomarker availability is not routine in most clinical FS settings. 3.9. Worked Example of Prototype Application A realistic fictional patient profile was constructed to illustrate prototype implementation using movement deficits, symptom duration, pain intensity, disability, diabetes status, glycaemic control, imaging biomarkers, psychological scores, sleep-related burden, and provisional inflammatory inputs. This example was intended only to demonstrate calculation and presentation of a domain-specific profile. It was not used as validation evidence, and its numerical output should not be interpreted as having established diagnostic, prognostic, or treatment-allocation meaning. In the five-domain model, the resulting domain-specific scores were 52.7 for the clinical domain, 69.4 for the metabolic/endocrine domain, 68.2 for the imaging domain, 49.9 for the psychological domain, and 53.7 for the inflammatory domain. Application of the provisional multidomain equation yielded a total FS burden score of 59.8/100. When expressed as relative weighted contributions within this example, the profile was approximately 27.9% metabolic/endocrine, 26.1% imaging, 18.6% clinical, 16.7% psychological, and 10.8% inflammatory. These distributions represent conceptual approximations derived from the current literature and should be interpreted as illustrative research-use outputs. To provide an integrated visual synthesis of the evidence-derived framework, Figure 3 summarises the five multidomain contributors retained in the model, their provisional relative contribution, and the intended research-use interpretation of the framework. 4. Discussion The present study developed an evidence-derived multidomain conceptual framework for FS and translated it into a research-use operational prototype for patient-level phenotyping. The main finding is that FS is not adequately represented by clinical presentation alone. Current evidence supports a multidomain structure in which clinical, metabolic/endocrine, imaging, psychological, and inflammatory contributors are all relevant, but with different degrees of evidential maturity and different levels of provisional weight. This interpretation is aligned with recent integrative research framing FS as a condition shaped by the interaction of local tissue pathology with metabolic, inflammatory, pain-processing, behavioural, and psychosocial mechanisms [4,9,15]. This framework extends previous multidomain and biopsychosocial interpretations of FS. Existing models have been valuable in moving the field away from a purely local capsular view of the condition, but they have generally remained descriptive and have not provided a transparent structure for organising different contributors into a patient-level profile. The present framework addresses this gap by separating two linked phases: first, systematic identification and synthesis of evidence across clinically meaningful domains; and second, exploratory framework development and operationalisation. In this sense, the model should be interpreted as a structured conceptual and research-use framework rather than as a validated clinical prediction tool. The prominence of the metabolic/endocrine sphere is clinically relevant. DM-related burden and glycaemic control emerged as the strongest internal components of this sphere, consistent with evidence showing that FS is more frequent in people with DM and may follow a more severe or prolonged course [9]. Other endocrine–metabolic factors, including obesity, thyroid dysfunction, altered lipid metabolism, and liver-associated metabolic markers, contributed additional but smaller layers of information [7,8,9,10]. These variables may define a systemic biological context in which advanced glycation, chronic low-grade inflammation, oxidative stress, endocrine dysfunction, and pro-fibrotic signalling influence capsular fibrosis and recovery potential [4,14,15,87]. This supports the relevance of metabolic profiling as part of FS phenotyping rather than as a secondary comorbidity screen alone. The imaging sphere also emerged as a major component of the framework. MRI and ultrasound findings were integrated within a single imaging domain because the retained evidence converged on related structural and activity-related tissue features rather than on clearly independent modality-specific constructs. Capsular structural biomarkers, edoema- or hyperintensity-related activity signs, vascular imaging features, and compensatory sonographic findings may provide information about tissue phenotype, disease phase, and prognosis. However, these findings should not be interpreted as direct equivalents of clinical severity. Instead, imaging appears to provide a complementary tissue-level layer that may help explain why patients with similar symptoms can differ in structural or activity-related disease expression. The psychological sphere underwent important refinement during framework development. Earlier versions of the model captured distress/mood and psychosocial quality of life, but targeted inclusion of sleep-focused research showed that sleep-related burden was too relevant to omit. At the same time, current evidence did not support sleep as a fully independent sixth sphere because sleep disturbance overlaps substantially with distress, pain-related burden, irritability, and psychosocial functioning [21,80]. Therefore, sleep-related burden was nested within the psychological sphere while retaining its status as a major subdomain. This structure recognises that the psychological contribution to FS extends beyond anxiety and depression and includes distress/mood, sleep-related burden, psychosocial quality of life, and psychological interference with perceived recovery [19,20,21,88,89,90,91]. Future studies should determine whether sleep adds independent explanatory value beyond the broader psychological domain, ideally using longitudinal designs, validated sleep instruments, and, where feasible, objective sleep measures. The inflammatory sphere remains the least secure domain in the framework and should be interpreted with particular caution. This does not imply that inflammation is biologically unimportant in FS. Mechanistic and translational research supports a role for inflammatory activation, cytokine imbalance, matrix remodelling, and failed fibrosis resolution [42]. In the retained case–control study, patients with FS showed higher serum IL-1β levels than controls, and an IL-1β polymorphism was associated with susceptibility, whereas MMP-3, TGF-β1, and GDF5 were not significantly associated within the same cohort. This supports retention of inflammation as a distinct domain, but also indicates that the direct retained signal is narrow and centred mainly on IL-1β-related biomarker and susceptibility evidence. Broader inflammatory-fibrotic mechanisms—including IL-10- and IL-17A-related pathways and repair/resolution cell populations—provide important contextual support but were not part of the retained quantitative evidence base used for weighting. Therefore, the lower weighting assigned to the inflammatory sphere reflects the limited maturity of the retained quantitative evidence rather than a judgement that inflammation is biologically unimportant [4,19,21,87,92]. Another implication of the framework concerns pain processing. Recent integrative research suggests that FS pain is not always reducible to local nociceptive input alone, but may involve interactions between peripheral inflammation, autonomic dysfunction, central and peripheral sensitisation, sleep disruption, and psychological amplification [93]. This view is consistent with maintaining clinical, psychological, and inflammatory spheres as analytically distinct components of the framework. However, analytical separation should not be interpreted clinically as evidence that these domains operate as independent compartments. Metabolic dysfunction, inflammatory activity, pain sensitisation, sleep disturbance, psychological distress, physical inactivity, and movement avoidance are likely to interact dynamically within individual patients. Accordingly, the framework should be interpreted as a tool for clinical orientation and research prioritisation rather than as a rigid phenotypic categorisation system [23]. A strength of the present study is that it moves beyond conceptual description and into operationalisation. The prototype demonstrates that the multidomain framework can be translated into measurable patient-level inputs and can generate both a total multidomain burden score and a sphere-specific contribution profile. Before validation, however, clinicians should not interpret these scores as diagnostic thresholds, prognostic probabilities, or treatment-allocation rules. The sphere-specific outputs are best understood as structured descriptive profiles that may help organise clinical reasoning and research stratification. The worked example was therefore included only to illustrate how the model can be operationalised; the value of 59.8/100 should not be interpreted as having established clinical meaning. This distinction is particularly important for future clinical implementation. Some inputs included in the complete framework, especially inflammatory biomarkers and advanced imaging variables, are not routinely available in many rehabilitation settings. Therefore, the full prototype may be most appropriate at this stage for research cohorts or specialist settings. Future work may need to develop simplified versions based on routinely obtainable variables, while preserving the multidomain logic of the framework. Such simplified models should be derived and validated prospectively rather than assumed from the present exploratory structure. Several limitations must be considered. First, the weighting procedure was evidence-derived and structured, but it was not estimated from individual-patient data, a single empirical prediction model, or an empirically estimated treatment-response model. Although explicit domain definitions, predefined assignment rules, independent reviewer procedures, risk-of-bias appraisal, and supplementary weighting tables improve transparency, traceability, and internal consistency, they do not eliminate investigator judgement. Another independent reviewer group could reasonably assign different weights when interpreting the same heterogeneous evidence base, particularly in domains supported by sparse, indirect, or methodologically variable data. Accordingly, the present weighting structure should be interpreted as an auditable first-step synthesis rather than as a fully reproducible or empirically derived coefficient system. Future iterations should complement structured reviewer judgement with formal Delphi or expert-consensus procedures, independent external replication of the weighting matrix, prospective individual-patient-data validation, and empirical derivation or recalibration in multicentre cohorts. Second, the search strategy may have introduced selection bias. The review was limited to three databases, English-language publications, open-access full texts, human adult studies, and studies published within the previous ten years. EMBASE was not searched. These criteria improved feasibility and ensured that all retained studies could be fully inspected and extracted, but they may have excluded relevant high-quality evidence published before the selected time window, in other languages, behind subscription barriers, or in additional databases. Because the present work was intended as a first exploratory framework rather than a definitive evidence map, the search was not repeated during revision. Future updates should remove the open access restriction, include additional databases such as EMBASE, consider broader language and date coverage where feasible, and test whether the resulting evidence base modifies the relative weighting of the domains. Third, the retained evidence was methodologically heterogeneous. Risk-of-bias assessment showed recurring concerns regarding confounding, selection bias, outcome measurement, statistical adjustment, and reporting. These concerns were not evenly distributed across domains. The clinical sphere was the largest but also the most heterogeneous; imaging and biomarker studies provided biologically relevant signals but were constrained by protocol variability and limited longitudinal follow-up; psychological studies relied substantially on self-reported constructs that may overlap with pain, disability, and sleep-related burden; and the inflammatory sphere was supported only by limited direct observational evidence. Accordingly, the apparent balance of the framework should be interpreted in light of differences in both the quantity and confidence of evidence across domains. Fourth, several assumptions were required to operationalise the prototype. Standardising all variables onto a 0–100 metric improves interpretability, but it may oversimplify nonlinear relationships or threshold effects for variables such as range-of-motion deficit, HbA1c, psychological distress, sleep disturbance, and imaging abnormalities. Similarly, proportional redistribution of weights in the presence of missing variables is pragmatic, but its impact on score stability has not yet been fully evaluated. The four-domain sensitivity specification excluding the inflammatory sphere suggested that the global worked-example score remained broadly stable, but this should be interpreted only as a preliminary robustness check, not as formal validation. These limitations define the next phase of research. Future work should move beyond structured evidence-derived weighting toward independent replication and empirical refinement. Formal Delphi consensus procedures, external expert reassessment of the weighting matrix, prospective individual-patient-data validation, and statistical recalibration in multicentre cohorts will be necessary to determine whether the proposed domain weights are stable, reproducible, and clinically useful. Prospective validation in real FS cohorts is also necessary to assess the calibration, stability, feasibility, and clinical usefulness of the framework. Future studies should test alternative weighting scenarios, evaluate nonlinear and threshold-based scoring functions, quantify the impact of missing data, and determine whether sphere-specific profiles are associated with treatment response, recurrence, symptom persistence, and patient-centred outcomes [94]. Particular attention should be paid to whether sleep-related burden should remain nested within the psychological sphere or be modelled independently, and whether inflammatory biomarkers add stable phenotypic or prognostic value beyond clinical, metabolic/endocrine, imaging, and psychological data. Finally, FS may represent a clinically meaningful opportunity for broader lifestyle-oriented intervention. Many patients present with sleep disturbance, reduced physical activity, metabolic dysfunction, stress-related burden, and altered recovery capacity. In this context, the shoulder complaint itself may function as an entry point for integrated health optimisation extending beyond local tissue management alone. The present framework does not prescribe treatment allocation, but it supports the rationale for future studies evaluating whether multidomain and lifestyle-informed rehabilitation strategies improve outcomes compared with approaches centred mainly on local shoulder impairment. 5. Conclusions FS is more appropriately understood as a multidomain condition than as a purely local stiffness syndrome. In the present evidence-derived framework, clinical presentation, metabolic/endocrine burden, imaging phenotype, psychological factors, sleep-related burden, lifestyle-related contributors, and inflammatory biology were organised into an integrated whole-person model. Metabolic/endocrine, imaging, and clinical factors emerged as major contributors, psychological factors added a substantial complementary layer, and inflammatory biology remained biologically relevant but quantitatively more provisional. This study provides a first conceptual and operational platform for multidomain phenotyping in FS. The framework is exploratory and should not be interpreted as a validated clinical decision-support tool, a definitive disease-partitioning system, or a treatment-allocation algorithm. Instead, it offers a structured basis for future validation, recalibration, and refinement of patient-level phenotyping strategies. Future research should determine whether these interacting domains can improve prognostic stratification, guide comprehensive rehabilitation planning, and support more personalised but integrated management of FS.