Osteoporosis is a major public health problem. Women are at a particularly high risk for osteoporosis and 50% of women age 50 or older may suffer from a fragility fracture in their remaining lifetime. Hip fractures are the most detrimental type of fractures. Research has been conducted to assess hip fracture risk so prevention methods could be used to reduce this risk in the growing number of older women. However, previous risk assessment approaches are limited to a few variables and linear combinations of these factors. Also, there is an increasing number of available measures, such as bone structures and skeletal muscle mass, that can be extracted, for instance, from dual-energy X-ray absorptiometry (DXA), and no reliable risk prediction model exist based on this wealth of information. The overall goal of this study is to develop a comprehensive and flexible model to assess the risk of hip fracture for a specific woman. This will be achieved by constructing a novel predictor that classifies data that include hip structural geometry, sarcopenia measurements as well as risk factors identified in previous studies. The construction of the predictive model will be partly based on a study conducted among a large (n = 11,432) multi-ethnic bone cohort from the nationwide Women's Health Initiative (WHI). In addition, to enhance the quality of the risk prediction, computational data from finite element simulations will be used. There are three specific aims.
The first aim i s to generate a risk model, based on clinical data that accounts for the coupling effects of the factors involved in hip fracture. This research introduces a new approach in the field of hip fracture, Support Vector Machines (SVM), which explicitly identifies the configurations of factors that are likely to lead to hip fracture.
The second aim i s to refine the prediction/decision model from the first aim using both the SVM classifier and finite element modeling. A scheme has been developed to select, in a high dimensional space, data points that would improve the accuracy of the SVM-based risk prediction model. These data points would be evaluated (fracture or not) using a finite element model. The novelty of the proposed finite element model stems from its full parameterization so that the variability of the bone response can be studied with respect to variations (even small) of structural geometry and material parameters.
The third aim i s to validate and compare the SVM-based risk with and without the use of finite element analysis and develop a hip fracture risk calculator for the web. A cross validation will be performed using data sets from the WHI as well as other cohorts. The flexibility of the SVM classification approach makes it easily deployable on the Internet. This study will be carried out using existing cohorts by an interdisciplinary team with experience in epidemiology of osteoporosis research, DXA measurements including hip structures and sarcopenia, fracture assessments, biostatistics approaches for large datasets, high dimensional analysis and finite element modeling, thus making this study highly feasible. The study results will have an extremely significant public health impact by providing an innovative tool for hip fracture risk assessments.

Public Health Relevance

This study will use innovative approaches, existing cohort resources, and interdisciplinary expertise to address a significant public health challenge: assessing the risk of hip fracture, the most detrimental type of fragility fractures. The study aims for a better risk assessment tool on the web that can be used by researchers and clinicians to assess an individual's hip fracture risk. This research will test new predictors and use the assumption free modeling approach to capture complex and non-linear relationships of predictors with fracture risk. This research is significant for reducing fracture burdens in the large and growing U.S. older women population.

Agency
National Institute of Health (NIH)
Institute
National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS)
Type
Exploratory/Developmental Grants (R21)
Project #
5R21AR060811-02
Application #
8327065
Study Section
Neurological, Aging and Musculoskeletal Epidemiology (NAME)
Program Officer
Chen, Faye H
Project Start
2011-09-01
Project End
2014-08-31
Budget Start
2012-09-01
Budget End
2014-08-31
Support Year
2
Fiscal Year
2012
Total Cost
$192,359
Indirect Cost
$57,359
Name
University of Arizona
Department
Public Health & Prev Medicine
Type
Schools of Public Health
DUNS #
806345617
City
Tucson
State
AZ
Country
United States
Zip Code
85721
Jiang, Peng; Missoum, Samy; Chen, Zhao (2014) Optimal SVM parameter selection for non-separable and unbalanced datasets. Struct Multidiscipl Optim 50:523-535