External validation of a model to predict the survival of patients presenting with a spinal epidural metastasis

The surgical treatment of spinal metastases is evolving. The major problem is the selection of patients who may benefit from surgical treatment. One of the criteria is an expected survival of at least 3 months. A prediction model has been previously developed. The present study has been performed in order to validate externally the model and to demonstrate that this model can be generalized to other institutions and other countries than the Netherlands. Data of 356 patients from five centers in Germany, Spain, Sweden, and the Netherlands who were treated for metastatic epidural spinal cord compression were collected. Hazard ratios in the test population corresponded with those of the developmental population. However, the observed and the expected survival were different. Analysis revealed that the baseline hazard function was significantly different. This tempted us to combine the data and develop a new prediction model. Estimating iteratively, a baseline hazard was composed. An adapted prediction model is presented. External validation of a prediction model revealed a difference in expected survival, although the relative contribution of the specific hazard ratios was the same as in the developmental population. This study emphasized the need to check the baseline hazard function in external validation. A new model has been developed using an estimated baseline hazard.