Page 125 - Predicting survival in patients with spinal bone metastasesL

For instance, in our paper described in Chapter VI, we noticed early on that the Rades model would do poorly, based on its weighing of the primary tumor types. The points assigned differed greatly from the survival times in our cohort, leading to a very low c-statistic. Had we optimized this category, we would have been able to better assess whether the combination of risk factors is of value. Now we only learned that a rigid translation of the model is not applicable to our dataset. Continuous updating and fine-tuning of the predictive models is needed to ensure their clinical applicability. Considering the above, it is not unlikely that our model will require some adaptations, since it is based on data collected from patients who were treated between 2000 and 2010. This update could then be implemented in the recently released OPTIModel app for Android and iOS (available for free in the Google Play Store and the App Store). The OPTIModel is a tool to estimate survival in patients with bone metastases to the spinal column and long bones, based on the work described in this thesis and papers by Willeumier et al.26
Where stability is concerned, it is unlikely that the SINS will fill the role of a predictive tool in its current form. Even as a referral tool for non-surgical clinicians, as which it is originally meant, it lacks discriminative capability, as the majority of patients end up in the intermediate category. Also, the manner in which the points are assigned across the different items seems somewhat arbitrary. For instance, patients who experience subluxation or translation of the spinal column are awarded four points. One could argue that this is de facto spinal instability and a patient should be referred for surgical consultation immediately, irrespective of the other items. Considering the vast clinical experience of the authors involved in creating the SINS, there is no doubt that the individual items are important in assessing the metastatic spinal column. Perhaps a re-imagining and simplification of the SINS into major criteria (misalignment, deformity, collapse) and minor criteria (location, aspect of the lesion, pain, and posterior involvement) can increase its clinical usefulness.
GENERAL DISCUSSION
While the aforementioned adjustments might result in predictive models attaining a c-statistic upwards of 0.7 and an increase in the discriminative power of the SINS, it is by no means is a great leap forward. All our efforts at optimizing care for a patient population as diverse as the one of spinal metastases could improve considerably with the advent of machine learning in medicine. In recent years, several abstracts have been presented at the annual meeting of the American
IX 123