Human Core Temperature Prediction for Heat-Injury Prevention

Area of Science:

• Physiology
• Biomedical Engineering
• Environmental Health

Background:

• Autoregressive (AR) models were previously developed to predict human core temperature and prevent hyperthermia.
• Existing AR models often provided delayed predictions, limiting their utility as real-time warning systems.

Purpose of the Study:

• To develop and evaluate novel alert algorithms for predicting hyperthermia by combining AR model point estimates with statistically derived prediction intervals (PIs).
• To assess the performance of three new algorithms: AR model plus PI, median filter of AR model plus PI decisions, and an adaptation of the sequential probability ratio test (SPRT).

Main Methods:

• Field-study data from 22 soldiers, including five who experienced hyperthermia, were used.
• The performance of alert algorithms was assessed using AR-model prediction windows ranging from 15 to 30 minutes.
• Cross-validation simulations were employed to evaluate prediction horizons and accuracy.

Main Results:

• A 20-minute prediction window offered a reasonable balance between prediction horizon and accuracy.
• The AR model plus PI and SPRT algorithms achieved the largest effective prediction horizons (≥18 minutes).
• The "model plus PI" algorithm demonstrated high sensitivity and a long effective prediction horizon, while SPRT offered fewer decision switches.

Conclusions:

• The "model plus PI" algorithm is recommended when high sensitivity and a long prediction horizon are critical for early warning of heat illness, provided decision switches are acceptable.
• SPRT presents a viable alternative for an early warning system when minimizing decision switches is a priority.
• These enhanced algorithms represent a significant improvement for real-time monitoring and prevention of heat-related illnesses in operational environments.