Page 103 - Predicting survival in patients with spinal bone metastasesL

independently by four expert observers (SD; orthopedic surgeon, YL; radiation oncologist, CR; radiologist, and WP; neurosurgeon). The observers were blinded to the patients’ event status. The SINS components pain and location were assessed by one observer (LB) based on the patients’ clinical files and admission forms. Based on the values for the individual SINS components, the total score was calculated and the spine was classified as being stable, potentially unstable or unstable5.
Patient characteristics were assessed using de Karnofsky Performance Status (KPS)13 and the Frankel classification14. Expected survival of each individual patient was determined using the Bollen model15. Clinical and/or radiological follow-up data for instability was collected until 12 months after initial radiotherapy. The end points of the study were considered adverse events after radiotherapy and consisted of (1) the development of a new pathologic fracture, (2) progression of an existing pathologic fracture, and (3) deterioration of alignment requiring surgical stabilization of the irradiated spinal segments. The occurrence of an adverse event was determined using patients’ medical charts and/or follow-up imaging.
Statistical analysis
Interobserver agreement between all four observers was calculated using Fleiss’ kappa16. The level of agreement for the obtained kappa was determined according to Landis et al.17 Survival curves were estimated by using Kaplan-Meier method and follow-up was assessed by using the reverse Kaplan-Meier method18. Survival times were calculated as the difference between start of treatment and date of death or last follow-up. Time to event was calculated as the difference between the date of treatment and date of occurrence of an adverse event or last follow- up, with death being considered a competing event. The cumulative incidence for the occurrence of an adverse event at six and twelve months was assessed by using a competing risk model19. Sensitivity, specificity, negative predictive value (NPV) and positive predictive value (PPV) were calculated for all observers. The results of the observer with the highest sensitivity and specificity were used to fit univariate and multivariate Fine and Gray models20, in order to estimate the effect of the risk factors on the cumulative incidence of the event. In order to perform the sensitivity analysis and the Fine and Gray’s regression analysis, the final SINS categories were reduced from three to two, aggregating the categories unstable and potentially unstable versus the category stable6,9. A P-value <0.05 was considered
VII
SINS
101