Not much is known about how accurate and reproducible different thermometers are at diagnosing patients with suspected fever. The study aims at evaluating which peripheral thermometers are more accurate and reproducible. We searched Medline, Embase, Scopus, WOS, CENTRAL, and Cinahl to perform: (1) diagnostic accuracy meta-analysis (MA) using rectal mercury-in-glass or digital thermometry as reference, and bivariate models for pooling; (2) network MA to estimate differences in mean temperature between devices; (3) Bland-Altman method to estimate 95% coefficient of reproducibility. PROSPERO registration: CRD42020174996. We included 46 studies enrolling more than 12,000 patients. Using 38 °C (100.4 ℉) as cut-off temperature, temporal infrared thermometry had a sensitivity of 0.76 (95% confidence interval, 0.65, 0.84; low certainty) and specificity of 0.96 (0.92, 0.98; moderate certainty); tympanic infrared thermometry had a sensitivity of 0.77 (0.60, 0.88; low certainty) and specificity of 0.98 (0.95, 0.99; moderate certainty). For all the other index devices, it was not possible to pool the estimates. Compared to the rectal mercury-in-glass thermometer, mean temperature differences were not statistically different from zero for temporal or tympanic infrared thermometry; the median coefficient of reproducibility ranged between 0.53 °C [0.95 ℉] for infrared temporal and 1.2 °C [2.16 ℉] for axillary digital thermometry. Several peripheral thermometers proved specific, but not sensitive for diagnosing fever with rectal thermometry as a reference standard, meaning that finding a temperature below 38 °C does not rule out fever. Fixed differences between temperatures together with random error means facing differences between measurements in the order of 2 °C [4.5 ℉]. This study informs practitioners of the limitations associated with different thermometers; peripheral ones are specific but not sensitive.
Keywords: Body temperature; Diagnostic tests; Fever; Systematic review; Thermometers.