Development, validation, and visualization of a CT-based predictive model for pedicle screw loosening risk after posterior lumbar fusion

This study aimed to develop and validate a visualization-based predictive model for evaluating the risk of pedicle screw loosening (PSL) following posterior lumbar interbody fusion (PLIF) in patients with lumbar degenerative conditions. A total of 466 consecutive patients undergoing primary PLIF with pedicle screw fixation for lumbar degenerative disease were retrospectively enrolled. Patients with prior spinal surgery, deformity, fracture, tumor, ankylosing spondylitis, metabolic bone disease, fixation involving more than four levels, screw redirection or malposition, or non-PSL-related reoperation within 12 months were excluded. Patients treated in 2016–2021 formed the derivation cohort, and those treated in 2022–2023 formed the temporal internal validation cohort. Multivariable logistic regression was used to identify independent predictors and develop a PSL risk-scoring model. Model robustness was assessed using 3,000-iteration bootstrap resampling, and a 5-fold cross-validation framework was used to generate receiver operating characteristic curves and calculate the area under the curve (AUC). Calibration and decision curve analyses were performed to evaluate model calibration and clinical utility. In the derivation cohort, the PSL incidence within 1 year was 22.74% (73/321). L 3 Hounsfield unit (HU), pedicle HU, and the lowest instrumented vertebra at S 1 were identified as independent PSL predictors. According to the risk-scoring categories, the probability of PSL was 2.8% for scores 0–2, 18.8% for 3–4, and 60.2% for 5–6. In the validation cohort, the PSL incidence within 1 year was 21.38% (31/145). The predicted and observed probabilities demonstrated good agreement, with a satisfactory model fit (Hosmer–Lemeshow test, P = 0.231), and the model provided meaningful net benefit in clinical decision-making. The model showed strong discriminative ability, with AUCs of 0.868 (95% CI: 0.825–0.914) and 0.846 (95% CI: 0.795–0.893) in the derivation and validation cohorts, respectively. The risk-scoring model derived from L 3 and pedicle HU values effectively predicted PSL following PLIF, highlighting the complementary roles of global and local bone mineral density in screw stability maintenance. This scoring system serves as a reliable and easy to use tool for evaluating PSL risk and holds potential for guiding preoperative evaluation and individualized treatment planning.