Incremental predictive value of longitudinal serum biomarker trajectories over baseline levels for rapid structural progression in knee osteoarthritis: a group-based trajectory modeling and least absolute shrinkage and selection operator Cox model

To develop a predictive model for rapid structural progression in knee osteoarthritis (KOA) using dynamic biomarker trajectory subtypes derived from longitudinal serum biomarker measurements, and to quantify the incremental prognostic value of longitudinal trajectory features beyond traditional single-timepoint baseline measurements.

This single-center retrospective cohort study included patients diagnosed with KOA at our institution between January 2019 and March 2023. Group-based trajectory modeling (GBTM) was used to identify longitudinal trajectories of four biomarkers (interleukin-6 [IL-6], tumor necrosis factor-α [TNF-α], cartilage oligomeric matrix protein [COMP], and hyaluronic acid [HA]). Least Absolute Shrinkage and Selection Operator (LASSO) Cox regression with the λ.1se criterion was employed to select predictors from 21 candidate variables, and Cox proportional hazards models were subsequently constructed. Harrell's C-index and Akaike Information Criterion (AIC) were used to assess model performance, with internal validation performed using 10-fold cross-validation repeated 20 times. The incremental prognostic value of longitudinal trajectory information was assessed via Net Reclassification Improvement (NRI) and Integrated Discrimination Improvement (IDI). A 24-month landmark analysis was conducted to evaluate potential time bias.

A total of 450 patients with KOA were included, of whom 58 (12.9%) experienced rapid structural progression during follow-up. GBTM analysis identified four trajectory classes for IL-6, three for TNF-α, and three each for COMP and HA. LASSO regression identified five independent predictors: Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) pain score (hazard ratio [HR] = 1.12, 95% confidence interval [CI]: 1.05-1.20), baseline IL-6 (HR = 1.23, 95% CI: 1.06-1.43), TNF-α Class 3 trajectory (HR = 1.80, 95% CI: 1.00-3.23), COMP Class 4 trajectory (HR = 2.98, 95% CI: 1.57-5.65), and HA Class 2 trajectory (HR = 6.81, 95% CI: 3.47-13.38). The primary model demonstrated excellent discrimination, with a concordance index (C-index) of 0.9034 (cross-validated C-index = 0.8982), an AIC of 538.3, and no evidence of violation of the proportional hazards assumption (global P = 0.609). Compared with the clinical-only model, the main model incorporating biomarker trajectory features achieved an IDI of 0.5801 and an event NRI of 0.5000.

Longitudinal serum biomarker trajectory patterns provide prognostic information for rapid structural progression (SP) in KOA beyond baseline features. HA Class 2 and COMP Class 4 trajectories were identified as the strongest independent prognostic predictors. The LASSO-based main model, with only five variables, demonstrated strong discriminative ability. However, due to the single-center retrospective design and the absence of independent external validation, prospective multicenter studies are required to further validate the model.