Use of dual-energy X-ray absorptiometry to evaluate variation in bone shape and alignment associated with radiographic knee osteoarthritis: Findings from a study of 19,053 individuals in UK Biobank

Knee osteoarthritis (KOA) is a common and debilitating condition characterised by the progressive degeneration of articular cartilage, alongside changes in the subchondral bone, synovial membrane, and surrounding soft tissues (1). Malalignment of the limb alignment, such as varus (bow-legged) and valgus (knock-kneed), often worsens with the progression of KOA and frequently exacerbates symptoms associated with mechanical overload in the affected knee compartment(s) (2). Varus alignment places increased stress on the medial compartment of the knee, while valgus alignment places greater stress on the lateral compartment (3). Lower limb alignment is a significant predictor of KOA severity and progression, including radiographic worsening (e.g. increased joint space narrowing and/or higher Kellgren-Lawrence grades), functional decline, and increased pain (2-4). Variation in lower limb alignment has also been associated with disease onset in some (3-5), but not all studies (6).
While varus and valgus deformities have been widely studied, other aspects of knee shape may also influence the development and progression of KOA. Statistical shape modelling has emerged as a valuable tool for quantifying complex, multidimensional variations in bone shape by identifying modes of variation—key patterns of shape differences among individuals (7). Statistical shape modelling has been used in various studies to evaluate overall hip shape and examine how its variation relates to pathologies such as hip osteoarthritis (8). For example, in a recent study using UK Biobank (UKB) data, we used machine learning techniques to analyse the relationships between radiographic hip OA (rHOA), derived from Dual-energy X-ray Absorptiometry (DXA) images, and joint shape, as evaluated by statistical shape modelling. Moderate rHOA was associated with femoral neck widening and increased acetabular coverage, while severe rHOA exhibited cam morphology and reduced acetabular coverage (9). Some of this shape variation may reflect a hypertrophic sub-type of hip osteoarthritis characterised by excessive bone formation (10); however, whether the same occurs at the knee is currently unknown.
Applying a similar approach to the knee could uncover shape variations beyond coronal plane lower limb alignment, providing insights into how variations in knee shape contribute to KOA progression and outcomes such as total knee replacement. We recently derived radiographic knee osteoarthritis (rKOA) measures from approximately 20,000 right knee DXA images in the UKB and confirmed expected relationships with subsequent risk of joint replacement (11). We also applied statistical shape modelling to examine knee joint shape in individuals who later underwent joint replacement, identifying shape patterns consistent with varus deformity (12). However, it remains unclear whether the observed association between knee shape and KOA risk is due to knee alignment alone or if other aspects of knee shape also contribute to KOA severity.
This study aims to examine the relationship between knee shape, as determined by statistical shape modelling applied to DXA images from UK Biobank, and varying severities of KOA. Additionally, it seeks to determine whether these associations remain after adjusting for hip-knee-ankle (HKA) angle, a commonly used radiographic measure of knee alignment (13), which we have recently derived using a novel machine learning approach applied to total body DXA scans. A better understanding of how knee shape influences KOA progression may support the development of targeted interventions focused on specific morphological features, potentially slowing disease progression.