We conducted a cross-sectional study using baseline data from a prospective study of shoulder patients in Central Denmark Region [18]. The study population has previously been described [13]. In brief, the population included patients, referred to one of six public departments of orthopaedic surgery on suspicion of SIS in the period 1 January 2011 to 28 February 2012. Inclusion criteria was age 18–63 years, residence in 18 out of 19 municipalities in Central Denmark Region (an island municipality was left out), at least one visit to a public department of orthopaedic surgery, response to a questionnaire at first visit or before surgery for SIS, and at least one available radiograph of the shoulder [13]. We excluded patients that did not fulfil the questionnaire.

The study was authorised by the Danish Data Protection Agency (journal number 2010–41–4316) and the Danish National Board of Health permitted the evaluation of radiographic examinations (reference number 3–3013-192/1/). In Denmark, questionnaire and register studies do not require approval by committees on health research ethics.

Oxford Shoulder Score (OSS) is a 12-item outcome measure of patient-reported shoulder pain and disability ranging from 0 (worst) to 48 (best) points [19, 20]. Data on OSS were extracted from patients’ response to a questionnaire at first visit at a public department of orthopaedic surgery or before any surgery for SIS. Participants were asked to consider their most painful shoulder when answering the OSS, or, in case of same level of pain, the right shoulder. We defined a low OSS as < 25 points, thus focusing on the third of the patients with more severe symptoms.

Radiographic examination was routinely performed before the first visit to a public department of orthopaedic surgery. We have previously described the radiographic examinations. In brief, the examinations included up to three radiographs i.e. anterior–posterior (AP) views in external and internal rotation and an outlet view (i.e. lateral projection with 10°–20° cranio-caudal angulation of the ray) (Fig. 1). The evaluation of the radiographs was performed by two medical doctors at residential level of orthopaedic education (LCA and KS), supervised by an experienced musculoskeletal radiologist (JG). The evaluation was done using a detailed manual with extensive illustrations, and an initial procedure of calibration between the two evaluators was performed [13].

Standard projections. Anterior–posterior projections with humerus in external rotation (A), internal rotation (B), and outlet view (C)

The radiographic findings comprised subacromial calcifications (Fig. 2), acromial morphological characteristics, acromioclavicular OA, signs of previous GH dislocation (Bankart and/or Hill-Sachs lesions), and architectural measures (i.e. acromial tilt, acromion index, and lateral acromial angle). For subacromial calcifications, we included presence (no/yes), area (no calcification, ≤ 0.2 cm2, > 0.2 to ≤ 0.6 cm2, and > 0.6 cm2), and characteristics according to Molés classification (no calcification, types A and B, type C, and type D) of any calcifications. No differentiation of the observed calcifications, based on the suspected underlying pathology of calcifications, was done (e.g. calcific tendinitis or CPPD). For acromial morphological characteristics, we included the Bigliani classification of acromion shape (type I “flat”, type II “curved”, and type III “hooked”) (Fig. 3); the presence (no/yes); types of lateral spurs (i.e. no lateral spur, heel type, traction type, and birdpeak type) (Fig. 4); and presence of medial acromial spurs (no/yes) (Fig. 5). For acromioclavicular OA, we included the presence (yes/no) (Fig. 6). For signs of previous GH dislocation, we focused on the presence of either Bankart or Hills Sachs lesion (yes/no). For the architectural measures, we included acromial tilt (angle between undersurface of acromion and line from tip of coracoid process to posterior aspect of acromion), acromion index (relationship between distance from glenoid fossa to lateral aspect of acromion resp. humerus), and lateral acromial angle (angle between acromion undersurface and glenoid fossa) (Fig. 7) [13]. The architectural measures were categorised into 3 groups (i.e. acromial tilt: 13.6° to 30°, > 30°to 35°, and > 35° to 50.7°, acromion index: 0.23 to 0.6, > 0.6 to 0.7, and > 0.7 to 0.96, and lateral acromial angle: 50.3° to 80°, > 80° to 90°, and > 90° to 121°). We used IMPAX version 6.5 to evaluate the radiographs, including in-programme tools to measure calcification areas as well as angles and distances for architectural measures.

Subacromial calcification

Acromion types: type I, flat (A); type II, curved (B); type III, hooked (C)

Lateral spurs in three different patients: bird beak type (A + B) and heel type (C)

Osteoarthritis of the acromioclavicular joint with narrowing of joint and spurring of medial acromion

Acromial tilt (A). Lateral acromial angle (B). Acromion index (C)

Covariates included sex, age, duration of patient reported shoulder problems, and evaluator of the radiograph. Sex and age (18–49 and 50–63 years) were extracted from personal identification numbers [21], while information on duration of shoulder problems was extracted from the questionnaire before first visit at the department. Duration of shoulder problems was grouped into 3 categories i.e. < 6 months, ≥ 6 to 24 months, and ≥ 24 months. We included evaluator of the radiograph as a covariate to adjust for any systematic difference in the radiographic findings between the two doctors.

If one or two questions from the OSS were left unanswered, we entered the mean value of all the patient’s other answers. If more than two answers were missing, we did not calculate the score [22]. Associations between a low OSS and radiographic findings were analysed using logistic regression [23]. The associations were estimated using crude, partly, and fully adjusted analyses. In partly adjusted analyses, we included sex, age, medical doctor, and duration of patient reported shoulder problems. A priori, we decided that the full model should include only one variable to represent calcification (present of calcification), spurs (lateral spurs), and acromion shape (acromion tilt). The full model therefore included sex, age, medical doctor, duration of patient reported shoulder problems, presence of calcification, lateral spur, acromioclavicular OA, and acromial tilt. Due to collinearity with lateral spur, Bankart/Hill Sachs lesion was excluded from the fully adjusted model. We did not include any of the calcification classifications in the fully adjusted model due to insufficient reliability of the Molés classification and low prevalence of the subtypes in the other classifications [13]. Association between the radiographic findings and low OSS reported several months later can be doubted if OSS changes with time. Therefore, we performed two sensitivity analyses. We excluded patients with radiographs performed more than 1 month before their completion of the questionnaire, and, in a separate analysis, patients with radiographs performed more than 3 months before their completion of the questionnaire. We used STATA 17 (StataCorp LP, College Station, TX, USA).