These techniques provide early warning of infection by monitoring physiological states before symptoms become apparent. Using machine learning models, these methods train classifiers with non-invasive patient data of physical waveforms, such as those from electrocardiography, hemodynamics, and temperature. They then apply those classifiers over a number of time intervals when analyzing an unknown patient’s data. Either the number of predictive patient state classifications or an aggregated patient state classification can then be used to indicate whether a patient has been exposed to an infectious or chemical agent. These are techniques with high sensitivity and low specificity.