We present a survival prediction model that has identified patient subgroups with poor (Group 4) to excellent (Group 1) prognoses. In addition, pretreatment symptoms were predictive of survival and correlated with the prediction of the model. If validated, we believe that this model, possibly in conjunction with patient symptoms, may aid in determining optimal treatment strategies.
Chad Tang, Kenneth Hess, Andrew Bishop, Hubert Pan, Eva Christiansen, Nizar Tannir, Behrang Amini, Claudio Tatsui, Lawrence Rhines, Paul Brown, Amol Ghia; UT MD Anderson Cancer Center
BACKGROUND/PURPOSE: There is uncertainty in the prognosis of patients following treatment of spinal metastases. To stratify patients between poor and excellent predicted survival after stereotactic spine radiosurgery (SSRS), we created a Cox proportional hazards regression model.
PATIENTS AND METHODS: Patients who were enrolled (between 2002 and 2011) in two prospective trials investigating SSRS for spinal metastasis were analyzed. To ensure mature survival data, living patients with < 3 years of follow-up were excluded. A multivariate Cox regression model was utilized to create a survival model via backward selection at P < .05. Pretreatment variables included race, sex, age, performance status, tumor histology, extent of vertebrae involvement, prior therapy at SSRS site, disease burden, and timing of diagnosis and treatment. Four prognostic groups were generated based on the model-derived prognostic index (PI).
To assess whether pretreatment symptoms were associated with survival and survival model predictions, patients were prospectively queried for their pretreatment symptoms via the MD Anderson Symptom Inventory (MDASI) and Brief Pain Index (BPI).
RESULTS: A total of 206 patients were included in this analysis. Median follow-up time was 70 months (range: 37–133 mo) among all living patients (n = 40). Seven variables were selected in the prediction model: female sex (hazard ratio [HR] = 0.7; P = .04), Karnofsky performance status (KPS) (HR = 0.7 per 10% increase; P = .005), prior surgery at the SSRS site (HR = 0.6; P = .005), prior radiation at the SSRS site (HR = 1.7; P = .003), SSRS site as the only site of disease (HR = 0.5; P < .001), number of organ systems involved (HR = 1.4 per involved system; P < .001), and time between initial diagnosis and spine metastasis after variable normalization (HR = 0.6 per log10[time in months+5]; HR = 0.01). The c-index of the stratified and unstratified model was 0.68 and 0.70, respectively.
The median survival time among all patients included in the analysis was 25.5 months and was significantly different among prognostic groups (Group 1 [excellent prognosis]: not reached, Group 2: 32.6 mo, Group 3: 19.7 mo, Group 4 [poor prognosis]: 8.2 mo; P < .001). Patients within the excellent prognosis group exhibited a Kaplan-Meier estimated 10-year survival rate of 71%. Furthermore, all pretreatment symptom metrics were predictive of overall survival and correlated with the model-derived PI (all P ≤ .01).
CONCLUSIONS: We present a survival prediction model that has identified patient subgroups with poor (Group 4) to excellent (Group 1) prognoses. In addition, pretreatment symptoms were predictive of survival and correlated with the prediction of the model. If validated, we believe that this model, possibly in conjunction with patient symptoms, may aid in determining optimal treatment strategies.
Proceedings of the 97th Annual Meeting of the American Radium Society - americanradiumsociety.org