Exploratory Spinopelvic Compensation Patterns Are Associated with a Strict Composite 12-Month MCID Endpoint After Total Knee Arthroplasty: A Retrospective Risk-Stratification Study

Background and Objectives: This study evaluated whether exploratory spinopelvic compensation patterns are associated with 12-month patient-reported recovery after total knee arthroplasty (TKA), with explicit separation between baseline-only prediction and contemporaneous association-based risk stratification. Materials and Methods: This retrospective single-center cohort included 700 complete cases selected from 723 primary TKA records after 23 exclusions for incomplete key radiographic alignment data and/or unavailable 12-month patient-reported outcome measures (PROMs). The primary endpoint was a strict composite MCID-based endpoint, defined as failure to achieve the prespecified minimal clinically important difference (MCID) for either the Oxford Knee Score (OKS) or WOMAC at 12 months. OKS- and WOMAC-specific MCID failures were reported as component sensitivity outcomes. K-means clustering was used to define exploratory compensation patterns rather than validated preoperative phenotypes. Predictor timing was separated into a baseline-only model and a contemporaneous model that additionally used 12-month alignment-change variables. Results: The strict composite MCID endpoint occurred in 537/700 patients (76.7%). Exploratory patterns showed increasing endpoint prevalence: Pattern A, 153/215 (71.2%); Pattern B, 195/258 (75.6%); and Pattern C, 189/227 (83.3%). The baseline-only ridge logistic model showed moderate discrimination (out-of-fold AUC 0.715, 95% CI 0.670–0.758). The contemporaneous model showed higher internal discrimination (out-of-fold AUC 0.864, 95% CI 0.834–0.894), but it included alignment changes measured in the same follow-up window as the outcome. Conclusions: Baseline-only prediction was modest, whereas stronger model performance was observed only in a contemporaneous association model. The exploratory patterns should not be interpreted as stable clinical phenotypes or used for individual-level decisions without external validation.