M. Raza , T. Laffaye , R. Stein , H. Ragati-Haghi , R. Amesbury , A. Mathiessen , C.K. Kwoh , J.E. Collins , J. Duryea
{"title":"PREDICTING KNEE OSTEOARTHRITIS PROGRESSION USING STRUCTURAL BIOMARKERS FROM MULTIPLE JOINTS: DATA FROM THE OSTEOARTHRITIS INITIATIVE","authors":"M. Raza , T. Laffaye , R. Stein , H. Ragati-Haghi , R. Amesbury , A. Mathiessen , C.K. Kwoh , J.E. Collins , J. Duryea","doi":"10.1016/j.ostima.2025.100310","DOIUrl":null,"url":null,"abstract":"<div><h3>INTRODUCTION</h3><div>Clinical risk prediction models have been developed to predict knee OA progression with the goal of targeted treatment and clinical trial enrichment. It remains unclear whether, or how, OA in other joints affects knee OA progression.</div></div><div><h3>OBJECTIVE</h3><div>To evaluate whether imaging biomarkers from non-index joints add predictive value for knee OA progression beyond those from the index knee alone.</div></div><div><h3>METHODS</h3><div>We included 648 participants from the Osteoarthritis Initiative (OAI), randomly selected with baseline KL grade of 1, 2, or 3. OAI obtained bilateral knee and hip XR and index knee MRI. Baseline imaging biomarkers included quantitative measures of index and non-index knee and hip fixed location joint space width and femorotibial angle (FTA) from XR and quantitative measures of cartilage thickness from index knee MRI. Clinical covariates were age, sex, BMI, injury history, surgery history, family history of knee replacement, and clinical hand OA (based on presence of Heberden’s nodes at the baseline clinical examination). Outcomes were knee OA progression over 48 months defined as (1) decrease in medial minimum joint space width (JSW) of ≥ 0.7mm and (2) any increase in KL grade.</div><div>We used random forests to determine the combination of predictors that maximize AUC. Random forests can model complex non-linear associations, interactions among predictors, and work well in the setting of correlated data. We examined each set of biomarkers alone and in combination: clinical covariates, index knee XR, contralateral knee XR, index hip XR, contralateral hip XR, index knee MRI. Models were tuned with 5-fold cross-validation and AUCs were computed over 1000 bootstrap samples. We used permutation-based variable importance to rank the most important variables for prediction.</div></div><div><h3>RESULTS</h3><div>The 648 OAI participants were 23% KLG 1, 48% KLG 2, and 28% KLG 3. Average age was 61 (SD 9) and average BMI 29 (SD 5). 152 (23%) had a decrease in JSW ≥0.7mm and 119 (18%) had an increase in KL grade.</div><div>In considering sets of covariates on their own, models with index knee MRI had the highest AUC for both outcomes (model 8), followed by models with index knee XR (model 3, Table). Adding contralateral hip XR to models with index knee XR improved AUC. For example, in predicting JSW≥0.7mm, the AUC increased from 0.627 (model 9) to 0.648 (model 10). Adding hip XR biomarkers did not seem to improve model discrimination (model 10 to model 11). AUCs from models from hip XR biomarkers alone were modest, though higher than for models with only clinical covariates.</div><div>Variable importance for the 10 most important biomarkers for the model with all XR biomarkers (model 12) is shown in the Figure for JSW ≥0.7mm (panel A) and KLG increase (panel B). Baseline medial minimum JSW was the most important predictor for both models. Various measures of fixed location JSW in the contralateral knee were among the top 10 most important predictors for both outcomes.</div></div><div><h3>CONCLUSION</h3><div>Multi-joint structural biomarkers improve predictive performance for knee OA progression, beyond index-knee imaging alone. These findings support broader imaging strategies to enhance RCT enrichment and guide targeted interventions in knee OA.</div></div>","PeriodicalId":74378,"journal":{"name":"Osteoarthritis imaging","volume":"5 ","pages":"Article 100310"},"PeriodicalIF":0.0000,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Osteoarthritis imaging","FirstCategoryId":"1085","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S2772654125000509","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
INTRODUCTION
Clinical risk prediction models have been developed to predict knee OA progression with the goal of targeted treatment and clinical trial enrichment. It remains unclear whether, or how, OA in other joints affects knee OA progression.
OBJECTIVE
To evaluate whether imaging biomarkers from non-index joints add predictive value for knee OA progression beyond those from the index knee alone.
METHODS
We included 648 participants from the Osteoarthritis Initiative (OAI), randomly selected with baseline KL grade of 1, 2, or 3. OAI obtained bilateral knee and hip XR and index knee MRI. Baseline imaging biomarkers included quantitative measures of index and non-index knee and hip fixed location joint space width and femorotibial angle (FTA) from XR and quantitative measures of cartilage thickness from index knee MRI. Clinical covariates were age, sex, BMI, injury history, surgery history, family history of knee replacement, and clinical hand OA (based on presence of Heberden’s nodes at the baseline clinical examination). Outcomes were knee OA progression over 48 months defined as (1) decrease in medial minimum joint space width (JSW) of ≥ 0.7mm and (2) any increase in KL grade.
We used random forests to determine the combination of predictors that maximize AUC. Random forests can model complex non-linear associations, interactions among predictors, and work well in the setting of correlated data. We examined each set of biomarkers alone and in combination: clinical covariates, index knee XR, contralateral knee XR, index hip XR, contralateral hip XR, index knee MRI. Models were tuned with 5-fold cross-validation and AUCs were computed over 1000 bootstrap samples. We used permutation-based variable importance to rank the most important variables for prediction.
RESULTS
The 648 OAI participants were 23% KLG 1, 48% KLG 2, and 28% KLG 3. Average age was 61 (SD 9) and average BMI 29 (SD 5). 152 (23%) had a decrease in JSW ≥0.7mm and 119 (18%) had an increase in KL grade.
In considering sets of covariates on their own, models with index knee MRI had the highest AUC for both outcomes (model 8), followed by models with index knee XR (model 3, Table). Adding contralateral hip XR to models with index knee XR improved AUC. For example, in predicting JSW≥0.7mm, the AUC increased from 0.627 (model 9) to 0.648 (model 10). Adding hip XR biomarkers did not seem to improve model discrimination (model 10 to model 11). AUCs from models from hip XR biomarkers alone were modest, though higher than for models with only clinical covariates.
Variable importance for the 10 most important biomarkers for the model with all XR biomarkers (model 12) is shown in the Figure for JSW ≥0.7mm (panel A) and KLG increase (panel B). Baseline medial minimum JSW was the most important predictor for both models. Various measures of fixed location JSW in the contralateral knee were among the top 10 most important predictors for both outcomes.
CONCLUSION
Multi-joint structural biomarkers improve predictive performance for knee OA progression, beyond index-knee imaging alone. These findings support broader imaging strategies to enhance RCT enrichment and guide targeted interventions in knee OA.