Introduction:
Survival prognostication plays a key role in the decision-making process for the surgical treatment of patients with spinal metastases. In the past traditional scoring systems such as the modified Tokuhashi and Tomita scoring systems have been used extensively, however in recent years their accuracy has been called into question. This has led to the development of machine learning algorithms to predict survival. In this study we aim to compare the accuracy of prognostic scoring systems in a surgically treated cohort of patients
Materials and Methods
This is a retrospective review of 318 surgically treated spinal metastases patients between 2009 to 2021. The primary outcome measured was survival from the time of diagnosis. Predicted survival at 3 months, 6 months and 1 year based on the prognostic scoring system was compared to actual survival. Predictive values of each scoring system was measured via area under receiver operating characteristic curves(AUROC). The following scoring systems were compared, Modified Tokuhashi(MT), TomitaI(T), Modified Bauer (MB), Van Den Linden(VDL), Oswestry (O), New England Spinal Metastases score (NESMS), GSTSG and SORG scoring systems.
Results:
For predicting 3 months survival, the GSTSG 0.980 (0.949-1.0) and NESM 0.980 (0.949-1.0) had outstanding predictive value, while the SORG 0.837 (0.751-0.923) and O 0.837 (0.775-0.900) had excellent predictive value. While for 6 months survival, only the O 0.819(0.758-0.880) had excellent predictive value and the GSTSG 0.791(0.725-0.857) had acceptable predictive value. For 1 year survival, the NESM 0.871(0.822-0.919) had excellent predictive value and the O 0.722 (0.657-0.786) had acceptable predictive value. The MT, T and MB scores had an AUC of <0.5 for 3 month, 6 month and 1 year survival
Discussion
Increasingly, traditional scoring systems such as the MT, T and MB scoring systems have become less predictive. While newer scoring systems such as the GSTSG, NESM and SORG have outstanding to excellent predictive value, there is no one survival scoring system that is able to accurately prognosticate survival at all 3 time points. A multidisciplinary, personalised approach to survival prognostication is needed.