Computer Model to Predict Hip Fracture Risk

Samy Missoum, left, and Zhao Chen will lead the University of Arizona team developing a hip fracture computer model. (University of Arizona)

Researchers at University of Arizona in Tucson are developing a computer model to predict which people are most at risk of a hip fracture. The study, led by engineering professor Samy Missoum and epidemiology and biostatistics professor Zhao Chen (pictured left), is funded by a two year, $357,982 grant from the National Institute of Arthritis and Musculoskeletal and Skin Diseases, part of National Institutes of Health.

According to the Centers for Disease Control and Prevention, there were 281,000 hospital admissions for hip fractures in 2007 and the rate for women was three times the rate for men. One out of five hip fracture patients dies within a year of their injury. In a 1990 study, researchers estimated that by the year 2040, the number of hip fractures would exceed 500,000.

The research by Missoum and Chen aims to build on current predictive models. “We are trying to improve the prediction accuracy by using new methodology from a mechanical engineering perspective,” says Chen. “We are also adding new variables for predicting hip fracture that don’t exist in current models….”

Chen notes that today’s models do not take into account the hip geometric structure, such as width of the bone, as well as nonlinear interaction of various risk factors, which limits their precision. One nonlinear interaction, for example, occurs when a slight change in femur (thigh bone) radius results in a large increase in risk. A similar increased risk might occur when bone mineral density drops below a particular threshold. In turn, these two factors — femur radius and bone mineral density – also may interact in a nonlinear fashion, further exacerbating the problem.

While medical researchers gather data and compute risk factors, mechanical engineers look at hip fractures as a structural problem. “If there is a slight change of geometry, then your chance of a stress concentration and fracture can be much higher or much less,” says Missoum. “It is an insight that is basic to our discipline.”

Chen and Missoum plan to account for these complexities in an accurate and easily manageable computer model that can navigate the maze of nonlinear interactions. The biomedical data for the model will come from a study of 11,432 women from three NIH Women’s Health Initiative clinical centers, who suffered 270 hip fractures.

Missoum and his students are analyzing these data using a statistical method called support vector machine (SVM) that can identify which combinations of risk factors will most likely result in fracture. The algorithms in SVM can uncover relationships between these risk factors, isolating those that are most significant and determining how they interact.

SVM is expected to produce a basic model, but to be robust enough to make predictions, it will need to be refined, and without the clinical data often available for this purpose. Missoum and his students plan to use a technique known as finite element analysis to gather such data from a virtual patient population. Finite element analysis is a numerical method that can simulate different types of patients with a range of hip geometries and bone material properties.

When the researchers generate a fracture risk model, they plan to validate that model against other calculators including FRAX, a new predictive model developed by the World Health Organization, as well as with data from new clinical studies.

One of the goals of the Arizona team is to provide a simple tool that can be widely used, reducing the current list of 40 factors to a core group of 5-8 variables. “We’re going to simplify the model to the minimum number of predictors that will give a reasonable prediction,” says Chen. “People will just need to enter the five to eight variables, and they can calculate their probability of a hip fracture in the next five to 10 years.”

Read more: Stem Cell Method Developed to Increase Bone Strength

*     *     *