Objective The objective was to compare agreement between three non-invasive measures of temperature and rectal temperatures and to estimate the sensitivity and specificity of these measures to detect a rectal temperature of 38°C or higher.

Methods We conducted a study of the diagnostic accuracy of oral, tympanic membrane (TM) and temporal artery (TA) thermometry to measure fever in an urban emergency department (ED). Data were collected from adult patients who received rectal temperature measurement. Bland-Altman analysis was performed; sensitivity, specificity and 95% CIs were calculated.

Results 987 patients were enrolled. 36% of the TM and TA readings differed by 0.5°C or more from rectal temperatures, 50% of oral temperatures. TM measures were most precise—the SD of the difference from rectal was 0.4°C TM, and 0.6°C for oral and TA (p<0.001). The sensitivities of a 38°C cutpoint on oral, TM and TA measures to detect a rectal temperature of 38°C or higher were: 37.0%, 68.3% and 71.1%, respectively (oral vs TM and TA p<0.001). The corresponding specificities were 99.4%, 98.2% and 92.3% (oral, TM and TA) with oral specificity significantly higher than the other two methods (p<0.01). TM and TA cutpoints of 37.5°C provided greater than 90% sensitivity to detect fever with specificity of 90% and 72%, respectively.

Conclusions None of the non-invasive methods met benchmarks for diagnostic accuracy using the criterion of 38°C to detect rectal temperature of 38°C. A TM cutpoint of 37.5°C provides maximum diagnostic accuracy of the three non-invasive measures.