Abstract: Systems and methods are disclosed herein for predicting whether a patient has been exposed to an agent. Physiological data is recorded from the patient during a first time interval, and one or more features are extracted from the physiological data. A plurality of classifiers is identified, wherein each classifier is trained using training data for a respective specific post-exposure time interval. For each classifier and based on a respective subset of the one or more features, a patient state classification that indicates an initial prediction of whether the patient has been exposed to the agent is determined. An indication of a prediction that the patient has been exposed to the agent is provided when a number of patient state classifications indicating a positive initial prediction that the patient has been exposed to the agent exceeds a first threshold.