Predictive models and their application to one of the most disabling diseases in adults: osteoarthritis of the knee

Carlos Andrés Soto Salamanca, Master's student in Engineering with Emphasis in Mechanical Engineering. Credit: Édgar Bejarano, Communications Office, Faculty of Engineering.

Osteoarthritis (OA) of the knee is a degenerative disease that usually causes disability and in severe cases requires costly surgery and long recovery times. An investigation allows predicting the onset of this disease based on demographic data and biomechanical analysis. The model developed lays the foundation for the development of a clinical tool to reduce the prevalence of this disease.  

Lea este artículo en español aquí.

Osteoarthritis of the knee and its impact on people's quality of life 

OA is characterized by the loss of the mechanical properties of the articular cartilage of the knee and its subsequent erosion. It results in increased friction between the femur and tibia bones involved in movement during walking, leading to pain and joint stiffness. 

Given its high incidence in adults, and due to the seriousness of the effects on the quality of life and functional capacity of people, the mechanical engineer and researcher Carlos Andrés Soto Salamanca worked on the creation of a predictive model of this disease. He used demographic and anatomical data combined with results from biomechanical models. 

This research was presented within the framework of the Master's Degree in Engineering with Emphasis in Mechanical Engineering, under the direction of Professor José Jaime García Álvarez and researcher Alexander Paz Carvajal from the School of Civil Engineering and Geomatics. This work seeks to become an option to make personalized OA predictions, combining academic and scientific advances made to date in two areas: finite element models and anatomical and demographic information of each patient. 

Research: finite element modeling and multinomial logistic regression 

The research process carried out by Soto Salamanca started with the search for anatomical information related to the object of study, for which Magnetic Resonance Imaging (MRI) was crucial. For this purpose, the database The Osteoarthritis Initiative, freely accessible, was used, which yielded more than 2TB of information related to MRIs of more than 1000 subjects. The process of filtering and organizing the data was facilitated by the development of an algorithm implemented in MATLAB. 

Once the data were obtained and organized, the finite element meshes adjusted to the anthropometry of each patient were made using a template-based method. According to the researcher, the algorithm of this method takes femur width and depth and cartilage thickness as reference to scale a base model, thus representing the geometry of each subject's knee cartilage. 

"Each of us has a unique anatomy, which makes us different. In the research we used an algorithm that allowed us to take a base mesh and scale it according to the measurements of the knee," says researcher Soto Salamanca. The researcher adds that based on this information, a simulation of a person's gait pattern was made, in which the level of stress produced by the contact between the cartilages was observed, showing concentrations of stress due to movement. This made it possible to identify regions susceptible to future damage, which over time can lead to osteoarthritis. 

In addition, multinomial logistic regression was used in the investigation. This is a statistical method that allowed the researcher Soto Salamanca to predict the development of the disease using risk factors usually associated with RAO, such as body mass index and age, together with biomechanical variables. 

The analysis was established within a time span of eight years, which is a prudent time to observe the development of osteoarthritis in adult life. The target subjects of this study were people over the age of 45 years. If OA can be predicted up to eight years in advance, habits related to diet and posture in daily living could be corrected. 

Usually research in this area uses the binomial logistic regression model to predict whether or not the patient will develop the disease. In this case, the researcher Soto Salamanca opted for the multinomial model, since this model allowed him to deduce, depending on the information provided, whether the joint presented severe deterioration, moderate deterioration, or remained healthy during the eight-year observation window. 

"Normally it is a disease that occurs in older people, and once it develops, it progresses rapidly. The problem is that you don't realize it until the moment when pain and crackles appear in the joint," says the researcher.  He adds that "this time is crucial for the progression of the disease, because while it takes about two to four years to develop moderate osteoarthritis (possible narrowing of the space between the joints), it takes less time to go from grade two to grade three or four (where it is evident that the bones are in contact). 

"What we did was try to anticipate that. It turns out that many times, when symptoms such as pain or others related to osteoarthritis appear, it is already late, the knee is already affected". According to the researcher, in extreme cases, when the disease reaches serious deterioration, there is no other alternative but to completely replace the knee by means of surgery; otherwise, the sufferer risks suffering from such a limitation that their motor autonomy are endangered. 

Some of the evaluations carried out within the framework of the research, according to the proposed methodology. Credit: researcher's courtesy.

Application and future use by medical professionals 

In the research, a significant improvement was evidenced with respect to previous models that did not combine the sources of information used by the researcher Soto Salamanca. Based on the area under the receiver operating characteristic (ROC) curve (AUC), metric used to evaluate these models, an average of 0.78 was obtained for the cases evaluated in the research (the ideal metric is 1). This represents a significant improvement when compared to the average of the cases that evaluated biomechanical variables and demographic and anatomical factors separately, with an AUC around 0.68. 

Given that the results of the research have supported the usefulness of the model, it is now proposed to become a reference for health professionals to put it into practice with patients in the coming years. Researcher Soto Salamanca proposes to join efforts with health professionals from the Universidad del Valle, such as physiotherapists with whom he has been working on the research, to validate the results of the model developed and design the medical application to prevent the disease. 

In the future, it is expected that the automation of the process will be improved to make it easier to use, and that the predictions will be more accurate and clinically relevant. In this way, more and more healthcare professionals will be able to use it with greater confidence and certainty in the results it produces. 

"The next step is to include more data to the model, as this is paramount, and also to see what other variables we can evaluate. In the future we want to make customized predictions further in advance. Not just 8 years but maybe 10, 12 years. Because nowadays osteoarthritis is occurring a lot in younger people. The idea is to implement it in a wider range of ages", says engineer Soto Salamanca. 

If interested in being in touch with the Master's student or any further information about the investigation, please write the Faculty of Engineering Communications Office: comunicaingenieria@correounivalle.edu.co.