# Prognostic model for osteoarthritis combining imaging and clinical biomarkers

## TL;DR

This study creates a model to predict knee osteoarthritis deterioration using imaging and clinical data, helping identify high-risk patients early.

## Contribution

A novel prognostic model for knee osteoarthritis combining imaging and clinical biomarkers is developed and validated using machine learning.

## Key findings

- The Random Forest model achieved the highest predictive accuracy (AUC of 0.910) for clinical deterioration in knee OA.
- Key predictors include BMI, total bone marrow lesion volume, and urinary C-terminal telopeptide of type II collagen.

## Abstract

This study aims to construct and validate a prognostic prediction model for knee osteoarthritis (OA) based on baseline characteristics, imaging manifestations and clinical indicators, in order to effectively assess the risk of adverse outcomes of individual patients, and to provide scientific basis for early identification of high-risk groups, formulation of precise intervention strategies and individualized management plans.

This retrospective study enrolled 345 knee OA patients. The patients were randomly divided into a training set (n = 241) and a validation set (n = 104) in a 7:3 ratio. Imaging features and clinical indicators were collected for all patients. In the training set, univariate analysis, least absolute shrinkage and selection operator (LASSO) regression, and multivariate logistic regression were applied to identify independent predictors. Furthermore, three machine learning models—Random Forest (RF), Support Vector Machine (SVM), and Gradient Boosting Machine (GBM)—were constructed. The predictive performance of each model was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curve and decision curve analysis, and the optimal model was selected for final prediction, with the importance of key predictive variables analyzed.

No significant differences were observed in baseline characteristics between the training and validation sets (p > 0.05). Univariate analysis showed that body mass index (BMI), medial joint space width (mJSW), total bone marrow lesion volume (TBLV), tibiofemoral angle (TFA), Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) function subscore, serum high-sensitivity C-reactive protein (hs-CRP), and urinary C-terminal telopeptide of type II collagen (uCTX-II) were associated with treatment failure (all p < 0.05). Multivariate Logistic regression identified BMI, TBLV, TFA, WOMAC function subscore, serum hs-CRP, and uCTX-II as independent risk factors for sustained clinical deterioration in knee OA patients (p < 0.05), while mJSW was an independent protective factor (p < 0.05). The RF model exhibited the highest AUC (0.910), significantly outperforming the SVM (0.885) and GBM (0.824), thus being selected as the optimal model.

The RF model, constructed based on imaging features and clinical indicators, effectively predicts the risk of sustained clinical deterioration in knee OA patients, with BMI, TBLV, and serum uCTX-II serving as key predictive markers.