Background: The purpose of this study was to investigate whether statistical parameters describing the 3D shape of the proximal femur and its local BMD distribution can discriminate subjects with and without acute osteoporotic hip fractures.
Methods: QCT datasets of the femur from 98 postmenopausal women (46 with acute hip fractures) (EFFECT study) were used. Statistical shape models were built by non-rigid registration of segmented femurs to a reference. The shape (appearance) model consisted of principal components, PCs, of the registration displacement vectors (respectively, BMD) for the whole sample set. PCs that discriminated fracture significantly after adjustment for age, height, and weight were included into binary logistic regression to obtain odds ratios and AUC of the ROC curve. A stratified validation (proportional number of ctrl and fx randomly chosen datasets) was performed by means of random partitioning the whole sample set into k partitions with k=1, 2, and 5. The case k=1 corresponds to the discrimination test on the whole population. The shape/appearance model discrimination was compared with that of total proximal femur integral BMD for the same partitions. Finally, we compared a combined statistical model (both shape and appearance included) with trochanteric trabecular BMD (TrTrabBMD) + cortical thickness of the neck (NeckCortTh).
Results: Both for the appearance and the shape model only one significant parameter was found: third and tenth PC, respectively. The discriminative performance of the appearance model was comparable with that of integral BMD, the shape model was not superior. Performance of the combined shape-appearance model was significantly worse (AUC 0.71 [0.61,0.81]) than TrTrabBMD+NeckCortTh (AUC 0.79 [0.70,0.88]).
Conclusions: The performance of the statistical models was not superior to that of BMD, despite the fact that the model was built from the same dataset (biased optimal conditions). The hypothesised BMD and geometry model with equal number of parameters was much better at discrimination.
Disclosure: The authors declared no competing interests. This project was partly supported by the German Federal Ministry of Education and Research (BMBF), BioAsset 01EC1005D.