Which Prognostic Model Best Predicts Poor Prognosis in Patients with Spinal Metastases? A Comparative Analysis of 8 Scoring Systems.

Abstract

Background: Existing scoring system's comparative effectiveness in identifying patients with poor prognosis (i.e., <6 months survival) has not been thoroughly explored.

Methods: We compared the predictive performance of 8 prognostic scoring systems (Tomita, modified Tokuhashi, modified Bauer, Rades, Oncological Spinal Prognostic Index, Lei, New England Spinal Metastasis Score, and the skeletal oncology research group [SORG] nomogram) with the area under the curve (AUC) from receiver operating characteristic curves and evaluated the predictive accuracy for 6-month survival across different primary tumor origins, and 1-month survival. Logistic regression was used to identify factors associated with 6-month survival.

Results: Six hundred forty one patients with spinal metastasis treated between 1994 and 2022 were included. The SORG nomogram showed best performance with low discriminative power in predicting 6-month survival (AUC [95% confidence interval {CI}]: 0.664 [0.584-0.744]). Logistic regression analysis identified significant factors influencing 6-month survival, including primary cancer type in Lei's classification, preoperative Frankel grades C and D, or grades A and B compared with grade E, preoperative white blood cell, preoperative albumin, and preoperative chemotherapy. For 1-month survival predictions, both the SORG nomogram (AUC [95% CI]: 0.750 [0.648-0.851]) and modified Tokuhashi score (AUC [95% CI]: 0.667 [0.552-0.781]) showed significance, albeit with moderate to low discriminative power.

Conclusions: This study shows that most scoring systems have low discriminative power, with only the SORG nomogram having moderate power for predicting poor prognosis. Recent and future advances in treatment, laboratory markers, and our understanding of tumor biology should be incorporated into prognostic models to improve their accuracy.