Evaluating hip morphology based radiographic hip osteoarthritis risk prediction models on new populations: data of the world coach consortium
Myrthe A. van den Berg, Fleur Boel, Michiel M. van Buuren, Noortje S. Riedstra, Jinchi Tang, Harbeer Ahedi, Vahid Arbabi, Nigel Arden, Sita Bierma-Zeinstra, Cindy Boer, Flavia Cicuttini, Timothy F. Cootes, Kay M. Crossley, David Felson, Willem Paul Giellis, Josh Heerey, Graeme Jones, Stefan Kluzek, Nancy E. Lane, Claudia Lindner, John A. Lynch, Joyce v. Meurs, Andrea Mosler, Amanda E. Nelson, Michael Nevitt, Edwin Oei, Jos Runhaar, Harrie Weinans, Jesse Krijthe, Rintje Agricola
DOI: https://doi.org/10.1016/j.joca.2024.02.105
Purpose:
Despite the growing burden of hip osteoarthritis (HOA), primary prevention methods are slowly emerging. Early identification of HOA is crucial in enhancing our understanding of HOA development and treatment options. Several hip morphology risk factors play a role during the development of radiographic HOA (RHOA), but the exact contribution to RHOA risk in a broad population remains unclear. By combining individual participant data (IPD) of various studies while considering study heterogeneity, novel modeling techniques could be explored to work towards individualized prediction models. This study aimed to evaluate the predictive performance of several hip morphology based RHOA risk prediction models built on multi-cohort datasets.
Design:
We utilized IPD from nine prospective cohort studies collected within the Worldwide Collaboration on OsteoArthritis prediCtion for the Hip (World COACH consortium). These studies all had standardized anteroposterior (AP) pelvic, long-limb, and/or hip radiographs taken and graded for RHOA at baseline and 4-8 years follow-up. The risk of incident RHOA was defined as hips with no signs of RHOA at baseline (any RHOA grade <2) which developed RHOA within this follow-up period (any RHOA grade ≥ 2). The lateral center edge angle (LCEA) and alpha angle (AA) were calculated automatically and relied on automated landmark placements on the outline of the hip using BoneFinder® (Figure 1). Risk prediction models that considered cohort variability were built with generalized linear mixed effects models (GLMM) and random forest models (RF). The discriminative performance (AUC) of models including the LCEA and/or AA was compared to models based on hip side, sex, age, and BMI alone. Stratified 5-fold cross-validation was performed to investigate the effect of a cohort label on predicted risk. With leave-one-cohort-out cross-validation, the generalizability of the models was evaluated on a new population. The mean AUC over the resulting test sets was compared in both settings.
Results:
In total, 35,922 hips without definite RHOA at baseline were included of which 4.7% developed RHOA within 4-8 years (Table 1). Performance differences between the model configurations and between GLMM and RF models were small (Table 2). Removing the risk contribution of a specific cohort label from the predictions made by the models evaluated in stratified 5-fold cross-validation caused a decrease (~0.2 in AUC) in performance. The leave-one-cohort-out validation showed mean AUC values between 0.54-0.61.
Conclusion:
This study showed poor performance of risk prediction models based on only two hip morphology measurements on new populations. In this multi-cohort dataset, the cohort label has more predictive information on RHOA risk than the considered risk factors. To use this large dataset to build individualized RHOA risk prediction models, additional efforts are recommended to examine the heterogeneity between the cohorts.
