A drop in the basket.

Here’s a simple and logical yet extremely illuminative argument about the easy misuses of data generated from well-intended basket trials. If a patient with non-small cell lung cancer (NSCLC) is treated successfully with a targeted therapy X for a novel mutation Y, all of a sudden patients with refractory NSCLC everywhere will be getting tested for mutation Y. Great, isn’t that the point? The problem lies in that we still have no clue about the prevalence of mutation Y among NSCLC at large. Consider for a moment the prevalence of Y is 1%, and the test for Y has a 99% sensitivity and 95% specificity—honestly, all are generous assumptions. This would mean testing 1000 patients with NSCLC for mutation Y would come with achieving 10 true positives. Again, great. What else would it come with? Another 50 false positives, meaning 50 patients receiving the toxicity and price tag of drug X without any reasonable hope for benefit. Taken together with the cost of testing all 1000 patients in the first place, that’s a whole lotta bulljive for 10 hail Mary’s. TBL: Jumping the gun on wide clinical applications of isolated basket trial wins may mean accepting totally unacceptable positive predictive values. | Shankaren, J Clin Oncol 2019