Can we improve the prediction of hip fracture by assessing bone structure using shape and appearance modelling?

There is a continuing need to improve the prediction of hip fractures to identify those at highest risk, enabling cost-effective use of preventative therapies.

The aim of this work was to validate an innovative imaging biomarker for hip fracture by modelling the shape and texture of the proximal femur assessed from dual energy X-ray absorptiometry (DXA) scans. Scans used were acquired at baseline from elderly patients participating in a prospective, placebo-controlled fracture prevention study of the bisphosphonate, clodronate. 182 subjects who subsequently suffered a hip fracture were age, weight and height matched with two controls who did not suffer a fracture during a median 4-year follow-up period. Logistic regression was used to test if variables were good predictors of fracture and adjust for bone mineral density (BMD).

Shape mode 2, reflecting variability in neck-shaft angle, neck width and the size of both trochanters (0.81 (OR), 0.68-0.97 (CI), 0.024 (P)), and appearance mode 6, recording grey-level contrast (1.33, 1.11-1.59, 0.002), were significant predictors of hip fracture and remained so after adjustment for BMD (shape mode 2 (0.77, 0.64-0.93, 0.006), appearance mode 6 (1.32, 1.10-1.59, 0.003)). Receiver Operating Curve analysis showed the combination of shape mode 2, appearance mode 6 and BMD was 3% better than any single predictor.

Variables derived from shape and appearance models gave a prediction of fracture comparable to BMD and in combination with BMD gave an improvement in the prediction of hip fracture that could predict an additional 2000 hip fracture cases per year in the UK, potentially saving more than £20 million per year and 10,000 cases in the US.