The systematic bias of ingestible core temperature sensors requires a correction by linear regression

An accurate measure of core body temperature is critical for monitoring individuals, groups and teams undertaking physical activity in situations of high heat stress or prolonged cold exposure. This study examined the range in systematic bias of ingestible temperature sensors compared to a certified and traceable reference thermometer. A total of 119 ingestible temperature sensors were immersed in a circulated water bath at five water temperatures (TEMP A: 35.12±0.60 °C, TEMP B: 37.33±0.56 °C, TEMP C: 39.48±0.73 °C, TEMP D: 41.58±0.97 °C, and TEMP E: 43.47±1.07 °C) along with a certified traceable reference thermometer. Thirteen sensors (10.9%) demonstrated a systematic bias greater than ±0.1 °C, of which 4 (3.3%) were greater than 0.06±0.24 °C. Limits of agreement (95%) indicated that systematic bias would likely fall in the range of -0.14 to 0.26 °C, highlighting that it is possible for temperatures measured between sensors to differ by more than 0.4 °C. The proportion of sensors with systematic bias greater than ±0.1 °C (10.9%) confirms that ingestible temperature sensors require correction to ensure their accuracy. An individualised linear correction achieved a mean systematic bias of 0.00 °C, and limits of agreement (95%) to 0.00 to 0.00 °C, with 100% of sensors achieving ±0.1 °C accuracy. Alternatively, a generalised linear function (Corrected Temperature (°C) = 1.00375 x Sensor Temperature (°C) - 0.205549), produced as the average slope and intercept of a sub-set of 52 sensors and excluding sensors with accuracy outside ±0.5 °C, reduced the systematic bias to less than ±0.1 °C in 98.4% of the remaining sensors (n = 64). In conclusion, these data show that using an uncalibrated ingestible temperature sensor may provide inaccurate data that still appears to be statistically, physiologically, and clinically meaningful. Correction of sensor temperature to a reference thermometer by linear function eliminates this systematic bias (individualised functions) or ensures systematic bias is within ±0.1 °C in 98% of the sensors (generalised function).