Multi-classifier prediction of knee osteoarthritis progression from incomplete imbalanced longitudinal data

Abstract

Conventional inclusion criteria used in osteoarthritis clinical trials are not very effective in selecting patients who would benefit from a therapy being tested. Typically majority of selected patients show no or limited disease progression during a trial period. As a consequence, the effect of the tested treatment cannot be observed, and the efforts and resources invested in running the trial are not rewarded. This could be avoided, if selection criteria were more predictive of the future disease progression. In this article, we formulated the patient selection problem as a multi-class classification task, with classes based on clinically relevant measures of progression (over a time scale typical for clinical trials). Using data from two long-term knee osteoarthritis studies OAI and CHECK, we tested multiple algorithms and learning process configurations (including multi-classifier approaches, cost-sensitive learning, and feature selection), to identify the best performing machine learning models. We examined the behaviour of the best models, with respect to prediction errors and the impact of used features, to confirm their clinical relevance. We found that the model-based selection outperforms the conventional inclusion criteria, reducing by 20-25% the number of patients who show no progression. This result might lead to more efficient clinical trials. Knee osteoarthritis (OA) is a chronic degenerative joint disease characterised by cartilage loss and changes in bones underneath it, causing pain and functional disability. The main clinical symptoms of knee OA are pain and stiffness, particularly after activity 1 , leading to reduced mobility and quality of life, and eventually resulting in knee replacement surgery. OA is one of the leading causes of global disability in people aged 65 and older, and its burden is likely to increase in the future with the ageing of the population and rise in obesity worldwide 2. OA is a heterogeneous disease where progression spreads over several years with periods of fast changes and periods of stability 3. A major challenge in OA drug development is effective selection of patients to the clinical trials. In an ideal case, all selected patients would show disease progression within the trial period, and their response to the drug in trial would be properly assessed. However, identification of patients in need of treatment, that is those with a high probability of progression, is an open problem.

Identification of Risk Factors and Machine Learning-Based Prediction Models for Knee Osteoarthritis Patients

“…X-ray combined with pain scores have been utilized by Halilaj et al to predict the progression of joint space narrowing (AUC = 0.86 using data from two visits spanning a year) and pain (AUC = 0.95 using data from a single visit) [6]. Similarly, another two studies (Tiulpin et al [10] and Widera et al [11]) made use of Xray images along with clinical data to predict KOA progression using either CNN or ML approaches achieving less accurate results. The current paper is the only one employing exclusively clinical non-imaging data and also contributes to the identification of important risk factors from a big pool of available features.…”

“…They demonstrated that a knee X-ray image alone is already a very powerful source of data to predict whether a particular knee will have OA progression or not [10]. Futhermore, in the same year, Widera et al used several ML models (e.g., logistic regression, K-nearest neighbor, SVC (linear kernel), SVC (RBF kernel) and RF) in combination with clinical data and X-ray image assessment metrics to develop predictive models for patient selection that outperform the conventional inclusion criteria used in clinical trials [11]. However, few studies have tried to apply ML models for the prediction of KOA.…”

“…. Specifically (i) 'gamma': [0, 0.4, 0.5, 0.6], 'maximal depth':[1,2,3, 4, 5, 6, 7, 8], 'minimum child and weight':[1,3,4,5,6,8] were optimized for XGboost, (ii) 'criterion': ['gini', 'entropy'], 'minimum samples leaf':[1,2,3], 'minimum samples split':[3,4,5,6,7] and 'number of estimators':[10,15,20,25,30] for random forest, (iii) 'maximal features': ['auto', 'sqrt', 'log2'], 'minimum samples leafs':[1,2,3,4,5,6,7,8,9,10,11] and 'minimum number of decision splits':[2,3,4,5,6,7,8,9,10,11,12,13,14,15] for decision trees, (iv) 'C': [0.001, 0.01, 0.1, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10] and 'kernel': ['linear','sigmoid','rbf','poly'] for SVMs, (v) 'k-parameter':…”

Precision medicine in osteoarthritis: not yet ready for prime time

“…Imaging features seen on conventional radiography and magnetic resonance imaging (MRI) are another type of biomarker. Artificial intelligence and machine learning may be essential for identifying imaging biomarkers [11,17,18]. Widera and colleagues developed several machine learning-based models to predict disease progression using observational data from the European APPROACH consortium [17,19].…”

“…Artificial intelligence and machine learning may be essential for identifying imaging biomarkers [11,17,18]. Widera and colleagues developed several machine learning-based models to predict disease progression using observational data from the European APPROACH consortium [17,19]. Models were validated using data from the Cohort Hip and Cohort Knee (CHECK) and Osteoarthritis Initiative (OAI) study populations.…”