The optimal assessment of the risk-to-benefit ratio of a new treatment, particularly in oncology, has always been based on the results of prospective randomized trials, ideally carried out in a double-blind manner with a large sample size and overall survival as the primary endpoint. This is still the best level of evidence available and has been an indisputable metric for discussing approval and reimbursement with health authorities for decades. This remains true in clinical oncology. However, it is no longer an appropriate standard in personalized medicine, particularly when focusing on rare subtypes. Indeed, molecular screening can categorize common diseases into several ultra-rare pathologies, each grouping a limited number of patients. For example, patients with NTRK-rearranged tumors are textbook cases. They could be identified and receive personalized treatment; but, in some countries, access to treatment is suspended pending the results of randomized trials. However, the demand is insurmountable.