Predictive model for survival in patients having repeat radiation treatment for painful bone metastases

PURPOSE: To establish a survival prediction model in the setting of a randomized trial of re-irradiation for painful bone metastases. METHODS: Data were randomly divided into training and testing sets with an approximately 3:2 ratio. Baseline factors of gender, primary cancer site, KPS, worst-pain score and age were included with backward variable selection to derive a model using the training set. A partial score was assigned by dividing the value of each statistically significant regression coefficient by the smallest statistically significant regression coefficient. The survival prediction score (SPS) was obtained by adding together partial scores for the variables that were statistically significant. Three risk groups were modelled. RESULTS: The training set included 460 patients and the testing set 351 patients. Only KPS and primary cancer site reached the 5%-significance level. Summing up the partial scores assigned to KPS (90-100, 0; 70-80, 1; 50-60, 2) and primary cancer site (breast, 0; prostate, 1.3; other, 2.6; lung, 3) totalled the SPS. The 1/3 and 2/3 percentiles of the SPS were 2 and 3.6. For the testing set, the median survival of the 3 groups was not reached, 11.3 (95% C.I. 8.5 - not reached) and 5.2 months (95% C.I. 3.7-6.5). The 3, 6 and 12 month survival rates for the worst group were 64.4% (95% C.I. 55.3-72.1%), 43.0% (95% C.I. 34.0-51.8%) and 19.7% (95% C.I. 12.4-28.1%) respectively, similar to that in the training set. CONCLUSION: This survival prediction model will assist in choosing dose fractionation. We recommend a single 8 Gy in the worst group identified.