An XGBoost-based integrated prediction model for short-term response to platelet-rich plasma therapy in osteoarthritic joint disease: A retrospective cohort study

Abstract

Objective Response to platelet-rich plasma (PRP) therapy varies among patients with osteoarthritic joint disease, underscoring the need for interpretable tools to support individualized prediction. Methods This single-center retrospective study screened adults with osteoarthritic joint disease receiving intra-articular PRP therapy at a tertiary Grade A hospital in China. Treatment response at 3 months was defined as both a ≥50% reduction in VAS pain score and a ≥20% reduction in WOMAC total score from baseline. Baseline VAS and WOMAC were excluded from the primary model to minimize outcome-definition-related bias. Candidate predictors were selected in the training cohort using LASSO regression and bootstrap stability selection. Logistic regression, random forest, support vector machine, XGBoost, and LightGBM models were developed and assessed for discrimination, calibration, clinical utility, internal validation, and interpretability. Results Among 312 screened patients, 226 were included; 136 were responders and 90 were non-responders. The training and validation cohorts included 158 and 68 patients, respectively. Five predictors were retained: KL grade IV, platelet concentration fold-change, BMI, disease duration, and IL-6. KL grade IV was associated with lower response probability (OR 0.312, 95% CI 0.136–0.716), whereas platelet concentration fold-change was associated with higher response probability (OR 1.636, 95% CI 1.202–2.226). Higher BMI, longer disease duration, and higher IL-6 were inversely associated with response. In the validation cohort, XGBoost achieved the numerically highest AUC of 0.896 (95% CI 0.817–0.958), with accuracy of 88.2%, sensitivity of 85.4%, specificity of 92.6%, and Brier score of 0.098. The optimism-corrected XGBoost AUC was 0.883. Conclusion An integrated post-preparation model combining patient characteristics and PRP preparation parameters showed favorable performance in predicting short-term response after PRP therapy in patients with osteoarthritic joint disease. External validation is required before clinical use.