Performance characterization of non-contact infrared thermometers (NCITs) for forehead temperature measurement.

The ability to assess the performance of a non-contact infrared thermometer (NCIT) may be limited due to the algorithms necessary to predict a reference site temperature (e.g., oral) from a measurement of the forehead skin temperature. The algorithm not only adjusts for the difference between the reference site temperature and forehead temperature, but may also account for hardware corrections, bias adjustments and emissivity settings. These algorithms are proprietary to the manufacturer and can be unique for each device. ASTM E1965-98 (2016) is a standard test method for the evaluation of NCITs. It includes forehead thermometers; however, the algorithm must be known or an unadjusted calibration mode must be accessible. This study evaluates 6 NCIT models (10 units of each) against the ASTM standard error criterion using a blackbody source. Units were tested within the manufacturer's operating and temperature measurement range specification. A method to evaluate measurement outliers and characterize each model's performance when the adjustment algorithm is unknown is proposed. Using this method, 5 of the 6 models had a predicted error > 0.3°C.

Clinical evaluation of non-contact infrared thermometers.

Non-contact infrared thermometers (NCITs) are being widely used during the COVID-19 pandemic as a temperature-measurement tool for screening and isolating patients in healthcare settings, travelers at ports of entry, and the general public. To understand the accuracy of NCITs, a clinical study was conducted with 1113 adult subjects using six different commercially available NCIT models. A total of 60 NCITs were tested with 10 units for each model. The NCIT-measured temperature was compared with the oral temperature obtained using a reference oral thermometer. The mean difference between the reference thermometer and NCIT measurement (clinical bias) was different for each NCIT model. The clinical bias ranged from just under - 0.9 °C (under-reporting) to just over 0.2 °C (over-reporting). The individual differences ranged from - 3 to + 2 °C in extreme cases, with the majority of the differences between - 2 and + 1 °C. Depending upon the NCIT model, 48% to 88% of the individual temperature measurements were outside the labeled accuracy stated by the manufacturers. The sensitivity of the NCIT models for detecting subject's temperature above 38 °C ranged from 0 to 0.69. Overall, our results indicate that some NCIT devices may not be consistently accurate enough to determine if subject's temperature exceeds a specific threshold of 38 °C. Model-to-model variability and individual model accuracy in the displayed temperature were found to be outside of acceptable limits. Accuracy and credibility of the NCITs should be thoroughly evaluated before using them as an effective screening tool.