Prognostic Evaluation of Prediction Model Survival Spinal Epidural Metastases

The prediction model is web-based and used five parameters: Karnofsky performance score (KPS) (in case of sudden (<24 hours) deterioration the score was used just before the deterioration), the curative intention to treat the primary, the spinal level of the symptomatic tumor, histopathology of the metastasis, and sex. These parameters will be recorded in addition to baseline characteristics. The date of presentation will be recorded. Four months after inclusion of the last patient survival status of all patients will scored. If applicable the date of death will be noted.

Patients are eligible if they suffer from a symptomatic epidural spinal metastatic lesion warranting therapy, either radiation therapy or a combination of surgery and radiation therapy. The nature of the lesion should be verified by pathological examination. In case of an active primary that has recently been diagnosed (less than one year previous to presentation of spinal lesion), the metastatic lesion is considered to be related to this primary. However, if the patient has more than one primary in the history, pathological verification of the tissue is obligatory.

After the patients provided informed consent of using their data for scientific use including the publication of the results, they will be included. The data will be collected, and electronically and anonymized stored. Patients will be included during three successive years. It is estimated that approximately 450 patients can be included. Two centers will participate, Radboud University Medical Center and Haaglanden Medical Center.

Statistical analysis The Cox model will be validated as described earlier 4, first graphically the predictive ability will be plotted, whereas next it will be quantitatively expressed as Harrel's c-index and the Royston -Sauerbrei D-statistic, Rd2. The calibration slope is also calculated. A sample of one third of all patients will be used for validation purpose. If needed the model is adapted. Then the validation is performed using the data of all patients.