Background <p>Early postoperative neurological deterioration remains one of the most serious complications after thoracic spine surgery. Conventional preoperative risk assessment relies mainly on clinical and imaging variables, whereas the incremental value of electrophysiological indicators has not been fully defined. Dermatomal somatosensory evoked potentials (DSEP) may provide segment-specific functional information beyond structural imaging. This study aimed to develop and internally validate a preoperative prediction model incorporating electrophysiological parameters for early postoperative neurological deterioration after thoracic decompression surgery.</p> Methods <p>A total of 508 patients who underwent thoracic decompression surgery were retrospectively included. Candidate predictors comprised age, preoperative Japanese Orthopaedic Association (JOA) score, number of compressed levels, T2-weighted intramedullary signal change, and DSEP-derived variables including maximal N1 latency, minimal amplitude, and number of abnormal DSEP sites. Four multivariable logistic regression models were constructed: a clinical model, a clinical + imaging model, a clinical + electrophysiological model, and a combined model. Internal validation was performed using stratified five-fold cross-validation. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), calibration metrics, Brier score, decision curve analysis, and bootstrap-based optimism correction.</p> Results <p>Among the 508 patients, 107 (21.1%) met criteria for early postoperative neurological deterioration. In the combined multivariable model, T2-weighted intramedullary signal change (OR 1.734, 95% CI 1.071–2.809, <i>P</i> = 0.025) and the number of abnormal DSEP sites (OR 1.687, 95% CI 1.338–2.128, <i>P</i> &lt; 0.001) were independently associated with early postoperative neurological deterioration, whereas maximal N1 latency showed a borderline association (OR 1.092, 95% CI 0.997–1.196, <i>P</i> = 0.058). Cross-validated discrimination was limited for the clinical model (AUC 0.543, 95% CI 0.478–0.610) and improved modestly for the clinical + imaging model (AUC 0.596, 95% CI 0.536–0.659). Incorporation of electrophysiological variables substantially improved performance in the clinical + electrophysiological model (AUC 0.713, 95% CI 0.655–0.766), while the combined model showed the highest overall performance (AUC 0.715, 95% CI 0.658–0.770). Models containing electrophysiological variables also demonstrated better calibration and lower Brier scores. Bootstrap optimism correction showed only modest optimism across models.</p> Conclusions <p>Preoperative DSEP abnormalities, particularly a greater burden of abnormal DSEP sites, may improve risk stratification for early postoperative neurological deterioration after thoracic spine surgery. Models incorporating electrophysiological variables showed better discrimination, calibration, and clinical net benefit than models based on clinical variables alone. These findings support the potential role of preoperative DSEP as a functional adjunct to conventional assessment, but external validation is required before broader clinical implementation.</p>