A retrospective predictive model for worsening knee osteoarthritis pain by integrating baseline gait biomechanics and wearable gait parameters.

objectiveTo investigate the performance of a predictive model for worsening knee osteoarthritis (KOA) pain that integrates baseline clinical characteristics, gait biomechanical parameters and wearable device-derived daily activity monitoring data and to evaluate the incremental predictive value of each module.

methodsA total of 984 KOA patients who visited the Department of Orthopedics, Pudong New Area People's Hospital between January 2018 and December 2023 were retrospectively included and randomly split into a training set (n = 689) and a test set (n = 295) at a 7:3 ratio. The primary outcome was a composite endpoint defined as an increase of ≥ 2 points in the WOMAC pain subscale score within 12 months after enrollment, accompanied by subjective patient-reported pain worsening (incidence, 28.0%). In the training set, candidate variables were initially screened by univariate analysis (P < 0.05) and final predictors were selected using LASSO regression with 10-fold cross-validation (lambda.1se = 0.012253). Four logistic regression models (clinical model, clinical + gait model, clinical + wearable model and fusion model) were constructed using a stepwise modular incremental strategy. Model discrimination was assessed using the area under the receiver operating characteristic curve (AUC), calibration using the Brier Score and calibration curves, incremental predictive value using integrated discrimination improvement (IDI) and net reclassification improvement (NRI) and clinical utility using decision curve analysis (DCA). Variable importance was assessed using multivariable logistic regression and SHAP method.

resultsA total of 17 variables were selected by LASSO (7 clinical variables, 4 gait-related variables and 6 wearable-derived variables). In the test set, the AUCs of the clinical model, clinical + gait model, clinical + wearable model and fusion model were 0.673 (95% CI: 0.605-0.740), 0.753 (95% CI: 0.695-0.811), 0.756 (95% CI: 0.695-0.816) and 0.815 (95% CI: 0.761-0.869), respectively. The fusion model achieved a Brier score of 0.144, a calibration intercept of -0.261 and a calibration slope of 0.856. Compared with the clinical model, the IDIs of the clinical + gait, clinical + wearable and fusion models were 0.056, 0.123 and 0.217, respectively and the continuous NRIs were 0.606, 0.614 and 0.697, respectively (all P < 0.05). Multivariable logistic regression identified baseline WOMAC stiffness score (OR = 1.912, 95% CI: 1.583-2.331, P < 0.001), analgesic use (OR = 1.753, P = 0.009) and daytime activity fragmentation index (per 0.1-unit increase, OR = 1.948, 95% CI: 1.597-2.402, P < 0.001) as independent variables risk factors for worsening pain. Regular exercise habits (OR = 0.630, P = 0.033), 20-m walk speed (per 0.1 m/s increase, OR = 0.748, 95% CI: 0.647-0.859, P < 0.001) and sedentary break frequency (OR = 0.780 per break/hour, P = 0.023) were independently associated with decreased odds of pain worsening. SHAP analysis showed that baseline WOMAC stiffness score (mean |SHAP| = 0.101) and daytime activity fragmentation index (mean |SHAP| = 0.098) were the two most important predictors in terms of global contribution. DCA indicated that the fusion model yielded the highest net benefit within a risk threshold probability range of 0.05-0.35. Subgroup sensitivity analyses confirmed the robustness of the fusion model (AUC > 0.79 across all subgroups).

conclusionAn integrated predictive model incorporating baseline clinical characteristics, gait biomechanics and wearable device-derived activity data can markedly enhance discrimination and risk classification of worsening KOA pain. These findings suggest that assessment of the biomechanical gait and monitoring of activity through wearable devices may provide additional predictive value beyond traditional clinical variables in identifying KOA patients at higher risk of pain worsening.