Abstract
Summary Statement
In a retrospective study, FE-based bone strength from CT data showed a greater ability than aBMD to discriminate proximal femur fractures versus controls.
Introduction
Personalised Finite Element (FE) models from Computed Tomography (CT) data are superior to bone mineral density (BMD) in predicting proximal femoral strength in vitro [Cody, 1999]. However, results similar to BMD were obtained in vivo, in retrospective classification of generic prevalent fractures [Amin, 2011] and in prospective classification of femoral fractures [Orwoll, 2009]. The aim of this work is to test, in a case-control retrospective study, the ability of a different, validated FE modelling procedure [Schileo, 2008] to: (i) discriminate between groups of proximal femoral fractures and controls; (ii) individually classify fractures and controls.
Patients & Methods
55 women (22 incident low-trauma proximal femur fractures and 33 controls) were enrolled in 3 clinical centres in Emilia Romagna region, Italy. All received a full femoral CT and DXA exams (in acute conditions for fractured cases) with a standardised protocol. Femoral neck aBMD was measured from DXA. FE models were built from CT (right femur for controls, intact for fractured) [Schileo, 2008]. Differently from existing works, FE strength was calculated for a range of 12 physiological directions of hip joint reactions [Bergmann, 2001] and 10 fall directions [Grassi, 2012]. Bone strength (in stance and fall) was the minimum load inducing on the femoral neck surface an elastic principal strain value greater than the yield limit [Bayraktar, 2004]. Fracture classification was analysed through logistic regressions and AUC of ROC curves.
Results
Mean FE strength and aBMD were significantly lower in the fractured than in the control group (33%, p<0.0001 for strength; 12% p=0.01 for aBMD).
Logistic regression on single variables
All classifiers were significant (p<0.001, AUC=0.88 for both stance and fall FE strength, p=0.02, AUC=0.72 for aBMD). The statistical power of the logistic regressions [Vaeth, 2004] was >0.9 for FE strength, 0.86 for aBMD.
Logistic regressions on multiple variables
Only FE strength was retained significant (p<0.001, AUC=0.88) when including aBMD in the regression.
Adding age to the logistic regression, FE strength and age (but not aBMD) remained significant, with AUC=0.95.
Discussion
FE strength could discriminate the fractured group better than aBMD and than [Keyak, 2011]. FE strength was a better fracture classifier than aBMD, and obtained AUC values slightly higher than [Amin, 2011; Orwoll, 2009]. The high statistical power mildens the small sample numerosity. Cases and controls were not age matched, but FE strength and age were found to be independent classifiers. In conclusion the proposed FE method was superior to aBMD in the classification of proximal femoral fractures.
