Machine learning models may predict survival of TKR in patients with OA
Click Here to Manage Email Alerts
Published results showed it is possible to predict when patients with knee osteoarthritis may undergo a total knee replacement and which patients would likely experience fast progression toward TKR.
Using data from the Osteoarthritis Initiative (OAI) cohort, researchers used Lasso’s Cox to identify the 10 most important predictive features among 1,107 features associated with accelerated knee OA leading to TKR. Researchers assessed the prognostic power of the selected features for risk and time to TKR with Kaplan-Meier analysis and applied seven machine learning methods, including Cox proportional hazards model, deepsurv/nonlinear model, linear multi-task logistic regression model, neural multi-task logistic regression model, random survival forest model, linear support vector machines model and Kernel support vector machines model. Researchers also assessed prediction performance using the concordance index, Brier score and time-dependent area under the curve.
Results identified X-rays, the MRI-assessed bone marrow lesions in medial condyle, hyaluronic acid injection, performance measure, medical history and knee symptoms as the most important TKR survival-based features. Researchers found the Cox proportional hazards model, deepsurv model and linear support vector machines model demonstrated the highest accuracy. Researchers chose the deepsurv model to build the prediction model to estimate the time to TKA for a given knee, and used bone marrow lesion, Kellgren Lawrence grade and knee symptoms to predict risk and time of a TKR event.
“As at present the time estimate to TKR is arbitrary to clinicians, this developed survival prediction model built with the OAI cohort could, in the future, better guide clinicians to the best therapeutic strategy to improve the survival of a given knee,” the authors wrote.
Collapse
Read more about
Click Here to Manage Email Alerts