Primary Spinal Tumor Mortality Score (PSTMS): a novel scoring system for predicting poor survival

Background context: Although the surgical and oncological therapies of primary spinal tumors (PSTs) have changed significantly over the last few decades, the prognosis of this rare disease is still poor. The decision-making process in the multidisciplinary management is handicapped by the lack of large-scale population-based prognostic studies.

Purpose: The objective of the present study was to investigate preoperative factors associated with PST mortality and to develop a predictive scoring system of poor survival.

Study design: This is a large-scale ambispective cohort study.

Patient sample: The study included 323 consecutive patients with PSTs, treated surgically over an 18-year period at a tertiary care spine referral center for a population of 10 million.

Outcome measure: Survival was the outcome measure.

Methods: Patients were randomly divided into a training cohort (n=273) and a validation cohort (n=50). In the training cohort, 12 preoperative factors were investigated using Cox proportional hazard models. Based on the mortality-related variables, a simple scoring system of mortality was created, and three groups of patients were identified. Kaplan-Meier and log-rank analyses were used to compare the survival in the three groups. The model performance was assessed by measuring the discriminative ability (c-index) of the model and by applying a pseudo-R(2) goodness-of-fit test (Nagelkerke R(2), RN(2)). Internal validation was performed using bootstrapping in the training cohort and assessing the discrimination and explained variation of the model in the validation cohort.

Results: Patient age, spinal region, tumor grade, spinal pain, motor deficit, and myelopathy/cauda equina syndrome were significantly associated with poor survival in the multivariate analysis (p<.001, RN(2)=0.799). Based on these variables, we developed the Primary Spinal Tumor Mortality Score (PSTMS), where an eight-point scale was divided into three categories (low, medium, and high mortality). The three PSTMS categories were significantly associated with the overall survival (p<.001, RN(2)=0.811, c=0.82). The model performance remained similarly high in the validation cohort (RN(2)=0.831, c=0.81).

Conclusions: The present study identifies six predictive variables for mortality in PSTs. Using these six variables, an easy-to-use scoring system was developed that can be applied to the estimation of postoperative survival in all types of PST patients.