A method includes, during a first time, for each user in a population: accessing a clinical assessment for depression in the user; accessing a set of biosignal data of the user; deriving correlations between the set of biosignal data and the clinical assessment for depression; and compiling the correlations into a depression model configured to predict risk of future depression diagnosis. The method further includes, during a second time: accessing a series of biosignal data collected by a device worn by a first user; accessing a target duration; calculating a risk of presentation of depression symptoms by the first user within the target duration based on the series of biosignal data; and in response to the risk exceeding a threshold, serving a notification to a care provider associated with the first user, the notification indicating the risk and including a prompt to investigate the first user for treatment.