Long-term PI-LL Mismatch after OLIF-PPSF for Adult Degenerative Scoliosis: Development and Internal Validation of Exploratory Prediction Models

Background Persistent pelvic incidence–lumbar lordosis (PI-LL) mismatch may remain after oblique lumbar interbody fusion combined with percutaneous pedicle screw fixation (OLIF-PPSF) for adult degenerative scoliosis (ADS), but its determinants and clinical interpretation remain uncertain. We characterized longitudinal outcomes and developed exploratory prediction models for final PI-LL mismatch. Methods This retrospective single-center cohort included 96 adults with adult degenerative scoliosis who underwent oblique lumbar interbody fusion combined with percutaneous posterior spinal fixation and completed at least 2 years of follow-up. Final PI-LL mismatch was defined as PI-LL > 10°. Longitudinal clinical and radiographic outcomes were evaluated using mixed-effects models. Logistic regression and random forest models used the same four prespecified predictors: preoperative PI-LL, age, preoperative Oswestry Disability Index, and number of fixation segments. Both models were internally validated using a 1000-resample optimism-corrected bootstrap procedure with complete model refitting, and paired bootstrap resampling compared the four-predictor logistic model with a PI-LL-only model. Full-sample restricted cubic spline analyses assessed nonlinear associations and estimated the PI-LL values at which the fitted probability of final mismatch crossed 0.50. Results At final follow-up, 56 patients (58.3%) had PI-LL mismatch. Pain, disability, and several radiographic parameters improved after surgery. Follow-up low-back pain, leg pain, ODI, and rates of clinically meaningful improvement did not differ significantly by PI-LL status, although patients with mismatch had less favorable residual sagittal and spinopelvic alignment. Greater preoperative PI-LL (odds ratio [OR], 1.161 per degree; 95% confidence interval [CI], 1.084–1.242), older age (OR, 1.099 per year; 95% CI 1.011–1.195), and higher preoperative ODI (OR, 1.048 per point; 95% CI 1.006–1.092) were associated with final mismatch. The optimism-corrected area under the curve (AUC) was 0.909 (95% CI 0.859–0.971) for logistic regression and 0.929 (95% CI 0.897–0.958) for random forest. The four-predictor logistic model did not significantly outperform PI-LL alone (ΔAUC, 0.015; 95% CI − 0.040 to 0.047; P = 0.467). Fitted risk crossed 0.50 at a preoperative PI-LL of 13.9° (bootstrap 95% CI for the crossing location, 9.0°–22.6°) and an immediate postoperative PI-LL of 12.2° (8.9°–15.6°). Conclusion Follow-up PI-LL mismatch after OLIF-PPSF was associated with less favorable radiographic alignment but was not synonymous with clinical failure. The models and model-derived reference points are exploratory and cohort-specific; external validation is required before clinical use.