This article, presented by Ostinelli et al., describes the development and assessment of a novel online tool called PETRUSHKA which aims to give personalized recommendations for antidepressants to patients with depression and their providers. The tool’s predictive algorithms use individual participant data from double-blind randomized controlled trials of 16 antidepressants as well as electronic health records from the Qresearch dataset to predict the efficacy, probability of discontinuation, and probability of adverse events of multiple antidepressants. Machine learning and statistical methods were combined to first predict these metrics for the antidepressant with the most data (the reference) and then to produce values for other antidepressants compared to the reference. These two results are then combined to predict outcomes for every antidepressant, and personalized recommendations that consider the probabilities of adverse effects that patients indicated they want to avoid are given. A web-based platform was developed to allow patients and clinicians to enter relevant information and to let patients indicate their preferences towards certain adverse effects. A dedicated group of patient representatives, as well as several clinicians, were involved in the production of this tool. An international randomized trial involving 504 participants with depressive disorder was designed to compare the PETRUSHKA tool to usual care.