Patient Specific Finite Element Modeling Outputs Outperform Clinical Metrics in Predicting Fusion Cage Subsidence

Finite element (FE) analysis of retrospective clinical cohort. To determine whether preoperative CT-derived FE model outputs can improve subsidence prediction compared with conventional clinical measurements alone in patients undergoing transforaminal lumbar interbody fusion (TLIF). Cage subsidence occurs in &#x223c;20% of spinal fusion patients and can lead to complications requiring reoperation. While individual risk factors are known, no validated tool integrates patient anatomy, bone quality, and implant characteristics to predict subsidence. Finite element models have been hypothesized to predict subsidence but lack clinical validation. Patient-specific FE models were created from preoperative CT scans of 42 TLIF patients: N=22 severe subsidence (&#x2265;4&#xa0;mm); N=20 nonsevere subsidence (<4&#xa0;mm). Vertebral geometries were segmented, and bone material properties were assigned based on Hounsfield units (HU). Cage positions from postoperative scans were registered to preoperative anatomy. Endplate and trabecular stresses and strains from FE models were compared with clinical measures using receiver operating characteristic analysis. Fifteen principal stresses and strains of the FE simulations showed significantly higher values in severely subsided patients compared with the nonsevere group. Average trabecular intermediate strain achieved the highest area under the curve score (AUC=0.809), outperforming all clinical metrics. Peak endplate minimum principal stress (AUC=0.775) was the second-best FE classifier. Traditional clinical measures showed lower discriminative ability: cage length (AUC=0.797), cage width (AUC=0.750), and cage height (AUC=0.698). Patient-specific FE model outputs significantly correlate with clinical subsidence outcomes and outperform several traditional metrics in classifying severe subsidence. Both endplate and trabecular stresses and strains are important predictors, with average values showing comparable or superior performance to peak values. Integration of FE models into the clinical workflow could provide a comprehensive preoperative subsidence prediction tool.