Page 124 - Predicting survival in patients with spinal bone metastasesL

CHAPTER IX
decision making. In Chapter VII we aimed to evaluate the SINS in a clinical setting, based on an analysis of 110 patients who were treated for SBM. Even though previous studies by the original authors have shown that the SINS has a substantial to excellent inter-observer and intra-observer reliability, the kappa-values found in this paper and several others18,19 differed considerably when compared to the levels found in the studies performed by Fourney et al. and Fisher et al.20,21 Also, the multivariate competing risk analysis showed that only the component location was significantly associated with the cumulative incidence of an adverse event, however a hazard ratio of 0.54 (95%CI 0.30-0.96, p=0.04) for this item indicates a protective effect for developing an adverse event. Even though the components constituting the SINS classification are important factors to consider when assessing spinal instability, in its current form, clinical applicability in predicting for the occurrence of a vertebral compression fracture seems limited.22,23 As a tool for streamlining communication between physicians of different medical specialties and facilitating the decision making concerning surgical consultation, the SINS could be useful.24
Future Perspectives
In their current form and with the current set of known risk factors, it would seem that the predictive models have reached a plateau with a c-statistic in the range of 0.6-0.7.16,23 By further fine-tuning the most important variable – the primary tumor – improvements can still be made,15,25 but it is doubtful that a value of 0.8 or higher, indicating very good predictive capability, will be reached. This by no means implies that we should abandon our attempts at improvements, as the routine use of these models does provide clinicians with useful feedback concerning survival estimations, which are generally speaking too positive.6
As treatments for different cancer types tend to differ considerably between countries, let alone continents, it is unlikely that a model created based on data from the Netherlands could be applied seamlessly in Japan or the United States. We experienced this during our attempts to integrate data from Austria into our own database. Even in the larger cancer-specific groups such as prostate cancer, survival times differed considerably between the two countries. This underlines the need for a more adaptive approach when evaluating the models, where the framework (i.e. the risk factors) would remain the same, but the way in which points are assigned or patients are stratified is adjusted according to regional data.
122