David Neilly1, Sameer K Khan2, Jennifer S Gregory3, Richard M Aspden3, James D Hutchison3, David J Deehan4. 1. North of Scotland Specialty Training Programme, United Kingdom; Division of Applied Medicine, University of Aberdeen, United Kingdom. 2. Newcastle upon Tyne Hospitals NHS Foundation Trust, United Kingdom. Electronic address: Sameer.khan@doctors.net.uk. 3. Division of Applied Medicine, University of Aberdeen, United Kingdom. 4. Newcastle upon Tyne Hospitals NHS Foundation Trust, United Kingdom; Institute of Cellular Medicine, Newcastle University, United Kingdom.
Abstract
INTRODUCTION: The geometrical shape of the proximal femur has previously been shown to predict primary hip fractures. Hip fractures are routinely diagnosed on plain radiographs of the pelvis, and these have both hips viewable. We have investigated if statistical shape modelling of the uninvolved hip on plain radiographs, at the time of the first hip fracture episode, could predict a subsequent 'second fracture' on that (uninvolved) side. MATERIALS AND METHODS: 60 radiographs taken at the time of the index hip fracture were blinded and separated into two arms; patients sustaining one hip fracture only (n=30), and those who went on to sustain a second fracture (n=30), over the three-year follow-up period. Two separate shape models were used for these groups and compared using t-tests or Mann-Whitney U-tests, along with Cohen's d to measure the effect size of each measure. RESULTS: We found no statistically significant difference in the shape of the femur between the first fracture and second fracture group (p>0.05) and no results reached a "medium" effect size (Cohen's d <0.5). CONCLUSIONS: Shape modelling is feasible and can be applied in the routine clinical setting. However, we were unable to elucidate any predictive value in this relatively small sample. A reliable radiograph-based method of identifying patients at risk of second fracture would be of value in planning prevention, service provision, and cost analysis. Further work is required and a study with more patients might exclude the type 2 error in our work.
INTRODUCTION: The geometrical shape of the proximal femur has previously been shown to predict primary hip fractures. Hip fractures are routinely diagnosed on plain radiographs of the pelvis, and these have both hips viewable. We have investigated if statistical shape modelling of the uninvolved hip on plain radiographs, at the time of the first hip fracture episode, could predict a subsequent 'second fracture' on that (uninvolved) side. MATERIALS AND METHODS: 60 radiographs taken at the time of the index hip fracture were blinded and separated into two arms; patients sustaining one hip fracture only (n=30), and those who went on to sustain a second fracture (n=30), over the three-year follow-up period. Two separate shape models were used for these groups and compared using t-tests or Mann-Whitney U-tests, along with Cohen's d to measure the effect size of each measure. RESULTS: We found no statistically significant difference in the shape of the femur between the first fracture and second fracture group (p>0.05) and no results reached a "medium" effect size (Cohen's d <0.5). CONCLUSIONS: Shape modelling is feasible and can be applied in the routine clinical setting. However, we were unable to elucidate any predictive value in this relatively small sample. A reliable radiograph-based method of identifying patients at risk of second fracture would be of value in planning prevention, service provision, and cost analysis. Further work is required and a study with more patients might exclude the type 2 error in our work.