Bone loss on the humerus side occurs in 67% of primary shoulder dislocations and up to 100% of recurrent shoulder dislocations. The size, direction and location of the humeral bone loss contribute to the instability of the shoulder joint. As a result, the presence of humeral bone loss is becoming increasingly important in determining the risk of recurrent dislocations and in determining the indication and technique of a possible operation. There are various measurement methods available for which a universally optimal measurement method and cut-off value are lacking. Furthermore, it is unclear whether and which measurements on different imaging modalities (CT/MRI) correspond with each other.

Description of studies

Breighner (2018) investigated a cross-sectional study in a single institution in the USA. The inclusion criteria were: Patients underwent MR imaging assessment of the shoulder, had or scheduled CT examination within 6 months of MR imaging without intervening shoulder surgery or trauma, and provided informed consent for the inclusion of this study. The exclusion criteria were excessive metal in the shoulder, as this produces artifacts in both modalities. There were 14 out of 34 patients included in the study. The mean age was 40 ±22, with 76% males. The index tests were CT and ZTE MRI. The average time between CT and MR imaging was 10 days ±32 (standard deviation); in most patients’ (29/34) CT and MR imaging examinations performed on the same day. All patients had an Hill-Sachs lesion.

Chalmers (2020) conducted a retrospective study in a single institution in the setting of orthopedic surgical department in the USA. The inclusion criteria were: patients who underwent the surgical treatment for glenohumeral instability as coded using the Common Procedure Terminology codes 29806, 23455, 23466, 23462, 23460, and 23465 at the University of Utah, and who underwent both a CT and an MRI performed within 1 year of each other. The exclusion criteria were: Patients in whom an intervening surgical procedure was performed on the shoulder. There were 53 out of 55 patients included in the study. The mean age was 31 ± 11, with 49% males. The index tests were CT and MRI (sequences not specified). All prevalence of Hill-Sachs lesion was not reported.

Cui (2023) investigated a cross-sectional study in a single institution in the setting of radiology and orthopedics departments in China. The inclusion criteria were: Patients with shoulder dislocation between July 2022 and June 2023, age of 18 years or older, shoulder anterior dislocation, with completion of both MRI and CT of the shoulder joint, and the interval between MRI and CT was 1 week or less. Both patients with primary dislocation and patients with recurrent dislocations were included. Exclusion criterium was a history of shoulder osseous surgery. 21 out of 56 patients were included in the study. The mean age was 27.5 ± 9.5, with 70% males. The index tests were 3D CT and 3D MRI 3D (FRACTURE). All patients had a Hill-Sachs lesion, 16 patients had a bipolar bone defect.

Feuerriegel (2023) conducted a cross-sectional study in a single institution in the setting of emergency department in Germany. The inclusion criteria were: Patients admitted to the emergency department with suspected traumatic dislocation of the shoulder, underwent 3-T MRI of the shoulder within 2 days after trauma, with a CT examination as part of the diagnostic workup in clinical routine. No exclusion criteria were reported. 25 Out of 46 patients were included in the study. The mean age was 40 ± 14.5, with 59% males. The index tests were CT and MRI (T1 GRE, Fracture, UTE). The interval between MRI and CT was unclear but seemed both in clinical routine. All patients had a Hill-Sachs lesion.

Lander (2022) performed a comparative study in a single institution in the setting of orthopedic surgical department in the USA. The inclusion criteria were: Consecutive patients with recurrent glenohumeral instability dislocations, older than 16 years, without prior surgery, and underwent both CT and MRI imaging modalities as well as 3D osseous reformats, which were ordered by their orthopedic surgeon. No exclusion criteria were reported. 18 patients were included in the study. The mean age was 34 (sd not reported), with 66% males. The index tests were CT and MRI or MRA (VIBE), which were executed in no particular order and the interval between MRI and CT was unclear. All patients had a Hill-Sachs lesion.

Sgroi (2021) conducted a cross-sectional study in a single institution in the setting of orthopedic surgical department in Germany. The inclusion criteria were: Consecutive patients with anterior shoulder instability scheduled for arthroscopy were retrospectively enrolled postoperatively with arthroscopic or open shoulder stabilization and available CT and MRI scans of the affected shoulder. The exclusion criteria were: concomitant rotator cuff tear, incomplete imaging diagnostics, and insufficient CT or MRI scans quality. 50 out of 80 patients were included in the study. The mean age was 26.4 ± 11.8, with 74% males. The index tests were CT and MRI (sequences not specified), which were performed as part of their preoperative diagnostic screening according to the routine clinical setup of the research setting. All patients had an Hill-Sachs lesion.

Stillwater (2017) performed a cross-sectional study in a single institution in Canada, without reporting the detailed setting of the study. The inclusion criteria were: All patients with glenohumeral instability or recurrent dislocations (either the first or recurrent), >18 years of age, imaged with both CT and MRI (VIBE) as requested by their orthopedic surgeons. One patient was excluded because of not presenting for the CT component of the study. There were 10 patients (12 shoulders) included in the study. The mean age was 29 (sd not reported), with 70% males. The index tests were CT and MRI (VIBE), which were performed within 24 h of one another without a particular order. All patients had an Hill-Sachs lesion.

Results

Results on linear based measurement methods for bone loss

Reference circle: [(circle diameter – residual humeral head width)/ circle diameter)*100]

Lander (2022) reported the percentage humeral defect for 2DCT (6.04%), 2D MRI (5.98%), 3D CT (8.29%), and 3D MRI (8.17%). There were no statistically significant differences between 3D CT and 3D MRI (p=0.769). Other comparisons do not seem to be tested and reported.

Hall method: (width of articular Hill Sachs lesion in degrees on arc / 180 degrees) *100

Sgroi (2021) reported a percentage humeral bone loss of 21.6% (SD 11.4) using CT compared to 21.0% (SD 10.2) using MRI. There were no statistically significant differences.

Flatow method using a reference circle: the quotient of the humeral head diameter without taking the Hill-Sachs lesion into account and the humeral head diameter taking the Hill-Sachs lesion into account

Using this method, Sgroi (2021) measured the humeral head diameter parallel to the articular surface of glenoid (without taking HS lesion into account) and the humeral head diameter parallel to the first diameter now taking HS lesion into account. The quotient of two diameters is calculated: for CT a mean of 17.4% (SD 8.3) was found, while for MRI 15.4% (SD 9.3) was found. There were no statistically significant differences.

Measurement without reference circle: [(humeral head height parallel to Hill-Sachs lesion – residual humeral head width)/ humeral head height parallel to Hill-Sachs lesion)*100]

Stillwater (2017) found a mean of 12.7% (SD 4.1) humeral head bone loss when measuring with CT, compared to 12.6% (SD 4.1) with MRI. There were no statistically significant differences.

Results on area based measurement methods for bone loss

No studies could be included that reported on area-based measurement methods to determine humeral head bone loss.

Hill-Sachs measurements

Length

Beighner (2018) measured the Hill-Sachs length (width on axial view) with 2 raters using CT and MRI. Rater 1 had an intraclass correlation coefficient (ICC) of 0.77 for measurements between modalities and 95% limits of agreement (LoAs) of -5.57 to 8.03. For rater 2, the ICC was 0.66 and the 95%LoAs ranged from -7.53 to 4.91.

Lander (2022) measured the maximal Hill-Sachs defect on 2D CT (18.19mm), 2D MRI (18.65mm), 3D CT (14.14mm), 3D MRI (12.39mm), and 2D MRI VIBE (19.28mm). No statistically significant differences were found for the measurements on 2D CT compared to 2D MRI and 2D MRI VIBE. Furthermore, the means of 3D CT and 3D MRI measurements also did not differ statistically significantly from each other.

Sgroi (2021) measured the length of the Hill-Sachs lesion in centimetres and found a mean of 1.4cm (SD 0.7) with CT measurements and 1.3cm (SD 0.7) with MRI measurements (not statistically significant). Sgroi (2021) also used Richards arc to determine the Hill-Sachs lesion size in degrees. A reference circle is drawn, and its center point is determined by the cross section of diameter lines. A line from the circle to its centre along the anterior edge of the articular surface is defined as 0 degrees. Size of the Hill Sachs lesion was determined by establishing its location on the 360-degree frame. It was observed that the mean arc size was 37.4 degrees (SD 34.6) on CT images and 34.9 degrees (SD 19.8) on MRI images (not statistically significant).

Depth

Beighner (2018) measured the Hill-Sachs depth on an axial view with 2 raters using CT and MRI. Rater 1 had an ICC of 0.85 for measurements between modalities and 95% limits of agreement (LoAs) of -2.09 to 2.40. For rater 2, the ICC was 0.90 and the 95%LoAs ranged from -2.27 to 1.50.

Sgroi (2021) measured the depth of the Hill-Sachs lesion in centimetres using both CT (mean: 0.7cm, SD 0.3) and MRI (mean: 0.7cm, SD 0.4). There were no statistically significant differences.

Interval

Cui (2023) drew parallel lines at both the medial edge of the lesion to the medial margin of the rotator cuff insertion on 3D humeral models. Distances between those lines in millimetres were Hill-Sachs interval measurements. A mean of 14.29mm (SD 1.93) was found for CT, compared to 14.35mm (SD 2.07) for MRI. This difference was not statistically significant. From a Bland-Altman plot, the mean difference was -0.06mm with 95%LoAs from -1.24 to 1.12.

Feuerriegel (2023) compared the Hill-Sachs interval on the axial view as measured with CT (mean: 17.4mm, SD 4.1), T1 GRE MRI (mean: 17.4, SD 4.2), FRACTURE MRI (mean: 17.3 mm, SD 4.1), UTE MRI (mean: 17.4mm, SD 4.2) and found no statistically significant differences between CT and MRI. The 95%LoAs could be approximated from a figure reported in the manuscript. For the agreement between CT and T1 GRE MRI, the 95%LoA seems to range from approximately -1.20 to 0.99. For CT versus FRACTURE MRI, this range was -1.05 to 0.8 approximately and, finally, for CT vs. UTE MRI the range was approximately -0.85 to 0.85.

Sgroi (2021) found a statistically significant difference between the measurements with CT (mean: 16.6mm, SD 0.5) and MRI (mean: 14.3mm, SD 0.5).

On/off-track classification

In all studies, the glenoid track was calculated using (0.83 * circle diameter) – diameter bone loss.

Lander (2022) reported that the imaging modalities were identical for classifying on/off track.

Sgroi (2021) observed how CT classified 33.3% of all lesions as being off-track, compared to 17.1% using MRI, although no statistically significant differences were found.

Fuerriegel (2023) found that CT and MRI all classified on-track (n=14) and off-track (n=6) in agreement, resulting in kappa=1.00.

Chalmers (2020) reported the proportion classified as on-track as measured by CT and MRI by two raters. Rater one classified 28.3% as on-track using CT and 41.5% on MRI. Rater two classified 37.7% and 39.6% on-track for CT and MRI, respectively.

Table 2. Overview of the results

Level of evidence of the literature

Linear based measurement methods

The level of evidence regarding linear measurement methods was downgraded by 2 levels because of the study limitations (1 level for risk of bias: there are multiple studies of doubtful quality); the number of included patients (1 level for imprecision: total sample size <100). We did not downgrade for inconsistency (reason: all studies report similar results) and indirectness (reason: there seem to be no deviations). Publication bias was not assessed.

Area based measurement methods

No studies could be included that reported on area-based measurement methods to determine humeral head bone loss.

Hill Sachs measurements

Length

The level of evidence regarding linear measurement methods was downgraded by 2 levels because of the study limitations (1 level for risk of bias: there is one study of adequate quality and there are multiple studies of doubtful quality); the number of included patients (1 level for imprecision: total sample size <100). We did not downgrade for inconsistency (reason: studies report similar results for mean differences; However, it is difficult to compare outcomes between studies that test for mean differences vs reporting agreement parameters) and indirectness (reason: there seem to be no deviations). Publication bias was not assessed.

Depth

The level of evidence regarding linear measurement methods was downgraded by 2 levels because of the study limitations (1 level for risk of bias: there is one study of adequate quality and one of doubtful quality); the number of included patients (1 level for imprecision: total sample size <100). We did not downgrade for indirectness (reason: there seem to be no deviations). Inconsistency could not be assessed (reason: the outcomes of both studies are difficult to compare [i.e. mean differences vs. agreement]). Publication bias was not assessed.

Interval

The level of evidence regarding linear measurement methods was downgraded by 2 levels because of the number of included patients (1 level for imprecision: total sample size <100), heterogeneity (1 level for inconsistency: the largest study contributing >50% of the total sample size found differences between mean CT and MRI measurements [>2mm], where other studies found mean CT and MRI measurements close to each other. Not enough information was provided on the MRI procedures to assess if this heterogeneity could be explained by different MRI procedures, and thus it remains unexplained. When focusing on agreement, two studies reporting agreement parameters seem to report similar 95% limits of agreement. However, if Sgroi 2021 would have reported 95%LoAs, it would probably differ from the 95%LoAs reported by Cui 2023 and Feuerriegel 2023. Conclusively, we decided to downgrade by 1 level). We did not downgrade for the study limitations (reason: there are two studies of very good quality available), indirectness (reason: there seem to be no deviations). Publication bias was not assessed. (NB: if we would ignore the study of doubtful quality and only assess both studies of very good quality, then the imprecision assessment would result in downgrading of 2 levels as the total sample size is then n<50, also resulting in ‘low GRADE’).

On/off-track classification

The level of evidence regarding linear measurement methods was downgraded by 3 levels because of the study limitations (1 level for risk of bias: there is one study of adequate quality, however it provided only 17.1% of the sample size. The other studies providing 82.9% of the sample size were assessed as being either doubtful or inadequate. Therefore, we decided to downgrade one level); heterogeneity (2 levels for inconsistency: there seem to be studies reporting perfect agreement and studies reporting potentially relevant disagreement between CT and MRI; some narratively, some proportionally, one statistically). We did not downgrade for imprecision (reason: total sample size >100) and indirectness (reason: there seem to be no deviations). Publication bias was not assessed.

A systematic review of the literature was performed to answer the following question:

‘’Which measurement method is the most accurate to measure humeral bone loss in patients diagnosed with posttraumatic shoulder instability?’’

Table 1. PICO

Relevant outcome measures

The guideline development group considered agreement parameters as a critical outcome measure for decision making; and tests for differences and correlations as an important outcome measure for decision making.

A priori, the guideline development group did not define the outcome measures listed above but used the definitions used in the studies. Outcomes were rated as ‘sufficient’ (correlation ≥0.70, AUC ≥0.70, Kappa ≥7.0), ‘insufficient’ (correlation <0.70, AUC <0.7, Kappa <7.0), or ‘indeterminate’ (correlation, AUC or Kappa not reported) based on the criteria for good measurement properties (Prinsen, 2018).

The guideline development group defined a clinically relevant difference of more than 5% compared to CT assessment as a clinically important disagreement of humeral bone loss measurements (both linear and area based), and 2mm for measurement of length/width/interval. This is acknowledged to be an arbitrary choice since evidence regarding the clinical importance of these differences are lacking. Bland-Altman plots showing 95% limits of agreement within these intervals (±2mm, ±5% from 0 [i.e. no difference]) will be rated as ‘sufficient’, even when a correlation coefficient is absent.

Search and select (Methods)

The databases Medline (via OVID) and Embase (via Embase.com) were searched with relevant search terms (after 2000) until February 6, 2024. The detailed search strategy is depicted under the tab Methods. The systematic literature search resulted in 960 hits. Studies were selected based on the following criteria: the population were patients with post-traumatic shoulder instability, the measurement of bone loss was performed by both MRI and CT tests, the agreement of the measurements by MRI and CT was reported. Cadaveric studies were excluded as these studies were not performed in the clinical setting.

Sixty-eight studies were initially selected based on title and abstract screening. The guideline development group checked the methods of the full-text studies to determine whether different measurement methods of humeral bone loss by MRI and CT was reported. After reading the full text, 61 studies were excluded (see the table with reasons for exclusion under the tab Methods), and 7 studies were included.

Results

Seven primary studies were included in the analysis of the literature. Important study characteristics and results were extracted in the evidence tables. Results are summarized in Table 1. The assessment of the COSMIN risk of bias is summarized in the risk of bias tables.