Validation of new prognostic and predictive scores by sequential testing approach.

Abstract

For practitioners, the question arises how their own patient population differs from that used in large-scale analyses resulting in new scores and nomograms and whether such tools actually are valid at a local level and thus can be implemented. A recent article proposed an easy-to-use method for the in-clinic validation of new prediction tools with a limited number of patients, a so-called sequential testing approach. The present study evaluates this approach in scores related to radiation oncology.
Three different scores were used, each predicting short overall survival after palliative radiotherapy (bone metastases, brain metastases, metastatic spinal cord compression). For each scenario, a limited number of consecutive patients entered the sequential testing approach (Table 1). The positive predictive value (PPV) was used for validation of the respective score and it was required that the PPV exceeded 80%.
For two scores, validity in the own local patient population could be confirmed after entering 13 and 17 patients, respectively (Figures 1 and 3). For the third score, no decision could be reached even after increasing the sample size to 30 (Figure 2).
In-clinic validation of new predictive tools with sequential testing approach should be preferred over uncritical adoption of tools which provide no significant benefit to local patient populations. Often the necessary number of patients can be reached within reasonable time frames even in small oncology practices. In addition, validation is performed continuously as the data are collected.