The invention relates to a computer-implemented method for providing a trained function (36) to determine a treatment. To allow a provision of improved treatment recommendations the method comprises the following steps: receiving (S11) a plurality of training data sets (30) each relating to a former patient and each comprising treatment data (31), which represents a former treatment chosen by a physician for the respective former patient, and patient data (32), which relates to a respective health condition of the respective former patient; determining (S12) a propensity score (33), which represents a relative frequency for a former treatment of the plurality to be chosen for each health conditions; determining (S13) an outcome score (34), which is a measure of how the health data (32) and/or the health condition of the respective patient changed; and determining (S14) a treatment recommendation score (35) based on the propensity score (33) and the outcome score (34); and providing (S15) the trained function (36).