Abstract
Clinical trials of a treatment in traditional mass medicine are based on the concept of proof of efficacy. It must be proven for a group of subjects that meet certain selection criteria. Subject variability must be demonstrated to exist and yet not to invalidate the proof of efficacy. If so, it is assumed that new patients meeting the same selection criteria would have a uniform response to treatment, irrespective of their individual traits. However, the variability that can be ignored for a group should not be ignored for an individual. Standard statistical methods are designed to estimate an average effect size for large enough groups, but they cannot predict an expected effect size for a single patient. Such predictions based on the patient's individual characteristics, rather than on their classification as a member of a target population or study group, are possible in personalized medicine. The latter employs multivariable predictive models via advanced mathematical methods implemented in Artificial Intelligence (AI), and it incorporates the subject variability in the predictive models to improve their accuracy and selectivity. There is a common misconception that personalized medicine belongs in a narrow area of rare diseases or genotype-guided care. In this paper, we argue that AI has potential to improve the treatment success estimates in traditional mass medicine as well at no extra cost to researchers. The clinical trial data on subject variability that are already routinely collected only need to be analyzed and interpreted using the methods of personalized medicine. To implement such improvements in medical practice, they need to be acknowledged and regulated by FDA and its counterparts in other countries.