Healthy shoulders do not all move in the same way. An analysis of an open-source dataset found recurring, person-specific paths for the centre of the upper-arm bone during similar arm motions. Across complete trials, the average excursion was 4.9 millimetres in three dimensions.
The source dataset contained 20 healthy participants who performed five motion types. The analysis retained 19 people aged 22 to 66, with a mean age of 41.5 years. Nine participants were female and 10 were male.
Here, translation means the maximum shift of the humeral head centre relative to the glenoid, the shoulder socket, along the socket's axes during a complete trial. The researchers calculated a three-dimensional distance and separate front-to-back, up-and-down and side-to-side measures, while also retaining up-and-down and front-to-back paths over time.
A few millimetres, but no single pattern
Average movement was 3.9 millimetres from front to back, 3.1 millimetres up and down and 2.5 millimetres from side to side. Relative to socket size, the averages were 14.7% in three dimensions, 15.4% from front to back, 9.5% up and down and 7.4% from side to side.
Those averages conceal substantial variation. Translation trajectories showed recurring subject-specific patterns in comparisons such as scapular versus coronal abduction and forward elevation with and without a handheld load. A group average, in this setting, does not describe a single typical path for every participant.
To test possible links, the researchers used log-transformed linear mixed-effects models, a statistical method that included motion, anatomy, posture and load as fixed predictors and participant identity as a random factor. The analysis used a p-value of 0.05 or less as its significance threshold. These are associations within the dataset, not evidence that one feature caused another.
Socket shape and posture showed modest links
Socket size showed an anatomical link with translation. Each 1% increase in glenoid height was associated with 0.78% higher three-dimensional translation and 1.02% higher up-and-down translation. The three-dimensional result was right at the significance cutoff, while the reported p-value for up-and-down movement was 0.01. When movement was normalised to socket size, height was not significantly associated with translation, and socket width had no significant front-to-back association.
Other aspects of socket shape also showed statistical links. Asphericity, a measure of how much the socket departs from a simple spherical shape, was associated with 0.42% higher three-dimensional translation and 0.66% higher front-to-back translation for each 1% increase in the measure. A one-degree increase in inclination, the socket's angle, was associated with 3.75% higher three-dimensional and 5.28% higher front-to-back translation. The reported effects were small and their uncertainty was broad.
Scapular tilt, the angle of the shoulder blade, was associated with 1.01% higher three-dimensional and 1.26% higher front-to-back translation for each one-degree increase. But tilt was correlated with upward rotation, which was omitted from the final model. The estimate may therefore reflect both measures rather than tilt alone.
Motion mattered more clearly than the added load
The arm movement itself was associated with translation in three dimensions and in the front-to-back and up-and-down directions. The global tests were statistically significant below the study's 1% threshold. High internal and external rotation had lower up-and-down translation than every other motion tested.
In pairwise comparisons, high internal and external rotation had 50.5% lower up-and-down translation than coronal abduction, 39.5% lower than scapular abduction, 46.0% lower than forward elevation and 48.2% lower than low internal and external rotation. It also had 26.8% lower three-dimensional translation than forward elevation. Forward elevation had 26.3% and 24.6% higher front-to-back translation than coronal and scapular abduction, respectively, while low internal and external rotation had 29.6% lower front-to-back translation than forward elevation.
Some three-dimensional and front-to-back comparisons had uncertainty ranges that approached or fell within the study's two-view imaging error, while up-and-down differences were clearer. When coronal abduction, scapular abduction and forward elevation were performed with and without the tested handheld load, no significant difference in translation magnitude was detected. That result applies to the tested load and selected movements only.
The choice of centre changed the measurement
The researchers also compared two possible reference points for the centre of the glenoid: its geometric mean, an average point calculated from the socket's shape, and the centre of a circle fitted to its lower portion. Across 150 pooled trials, the starting position was closer to the geometric mean in 89.3% of trials. The mean starting distance was 1.6 millimetres, or 4.8% of socket height, from the geometric mean, compared with 3.2 millimetres, or 9.6%, from the inferior circle centre.
The end-point comparison was similar: the geometric mean was closer in 88.7% of trials. In this dataset, it was usually the closer reference at both ends of the trial.
A reference, not a clinical rule
These findings have important limits. Joint parameters were measured from bony geometry without cartilage because high-resolution scans needed for cartilage segmentation were unavailable. The analysis included only 19 healthy participants, so it describes a small study group rather than the full range of shoulder anatomy.
The results are associations within this dataset. They do not establish that socket geometry, scapular posture, motion type or load causes a particular translation pattern, and they do not show clinical outcomes or how pathological shoulders would behave. The analysis also does not identify the muscular or soft-tissue features behind the person-specific paths.
The analysis used de-identified kinematic data and bone geometries from an open-access dataset; informed consent was obtained and research use was approved under IRB 71782. Statistical code and analysis data are available on Zenodo, with supplementary material online. The work was funded by the Georg and Bertha Schwyzer-Winiker Foundation, while source data collection was supported by NIAMS of the NIH. The authors reported no competing interests.
Paper data and sources
Original title: Determination of inter-individual differences in humeral head centre translations in subjects with healthy shoulders
Authors: Herbst E, Henninger H, Stettler F et al.
Journal/Repository: Royal Society open science
Status: Peer-reviewed
First online: 2026-10-07
DOI: Not available
Original paper · Full text