Predicting core temperature in humans using autoregressive model with exogenous inputs

Abstract

Although heat illnesses are preventable, they are difficult to predict because humans lack early warning mechanisms of impending heat illnesses. In our previous work, we showed that, due to the large thermal inertia of the human body, the core temperature in humans can be predicted with an autoregressive (AR) model for up to 20 minutes ahead of time. However, when used for real-time predictions, the AR model introduces a lag into the predicted core temperature, thus effectively reducing the prediction horizon. In order to reduce the prediction lag, an exogenous input which precedes the rise in the core temperature can be used. Currently available wearable physiologic monitors allow measuring several vital signs simultaneously. We used the Equivital physiologic monitoring system (Hidalgo Limited, UK) to collect the core temperature and the heart rate (HR) data in order to develop and test our algorithm. The core temperature data were collected using a telemetry pill (VitalSense, Respironics, USA). The algorithm uses the antecedent samples of core temperature along with the current readings of the heart rate to predict the core temperature variations 20 minutes ahead in real time. We compared the performance of the new algorithm with the previously developed AR algorithm and found that the prediction time lag is reduced by 50 % from 10 to 5 minutes.