Quantity of Research Available
A total of 523 citations were identified in the literature search. Following screening of titles and abstracts, 498 citations were excluded and 25 potentially relevant reports from the electronic search were retrieved for full-text review. Five potentially relevant publications, of a total of 14, were retrieved from the grey literature search. Of these 30 potentially relevant articles, 10 publications were excluded for various reasons, while 20 publications met the inclusion criteria and were included in this report. Of the studies included, four are systematic reviews and sixteen are non-randomized studies. Appendix 1 describes the PRISMA flowchart of the study selection.
The summary of study characteristics table is provided in Appendix 2, the results of the critical appraisal are in Appendix 3, and the main study findings and author conclusions are provided in Appendix 4.
Summary of Study Characteristics
The accuracy of tympanic thermometers for detecting febrile individuals
A total of fifteen studies were identified that evaluated the accuracy of tympanic thermometers. Of these publications, four were systematic reviews6–9 and eleven were non-randomized studies.3,10–19 Among these publications, tympanic temperature was used as the reference in four studies6,10,11,13 where results did not focus on the tympanic measurements.
Country of origin
Systematic reviews originated from China,8,9 New-Zealand,7 and France.6 The non-randomized studies were from Korea,10 China,11 Thailand,12 New-Zealand,13 USA,14 Spain,3 Gabon,15 United-Kingdom,17 Pakistan,18 Malaysia,16 and Belgium.19
Population
The mean age of patients in the included studies ranged from neonates9 to 80.919 years and a majority of studies reported a ratio of male/female near 1:1. Some studies only included pediatric patients,8,9,15 whereas one study only included geriatric patients.19 Studies included inpatients or patients presenting at hospital,3,6–9,11,14–19 or travelers presenting at borders.10,12,13 The sample size of the non-randomized studies ranged from 2116 to 2000.18 The systematic reviews included from 37 to 318 studies with samples from 97 to 72,3276 participants.
Interventions and comparators
Devices used to measure tympanic temperature varied across studies. The BraunThermoScan and the FirstTemp Genius were those used most often, whereas one study did not report the model of the device.11 The number of measurements and the mode of the device (i.e. the algorithm transforming the actual reading into the predicted body temperature) may have been different between studies, but were not always specified.
The accuracy of tympanic thermometers was compared with rectal temperature in six studies,8,9,14–16,19 with oral temperature in two studies,12,18 with pulmonary artery catheter temperature in two studies,3,7 with axillary temperature in one study,15 and with nasopharyngeal temperature in one study.17
Years of publication
The years of publication ranged from 2009 to 2014.
The accuracy of handheld infrared skin thermometers for detecting febrile individuals
Seven studies evaluated the accuracy of handheld infrared skin thermometers. Of these publications, one was a systematic review6 and six were non-randomized studies.3,17,20–23
Country of origin
The systematic review originated from France.6 The non-randomized studies were from Bolivia,23 Italy,20 Spain,3 USA,21,22 and United-Kingdom.17
Population
The age of the patients included in the studies ranged from 1 month21 to over 8017 years and most of the studies reported a similar proportion of males and females. Some studies only included pediatric patients.20–23 All studies included inpatients or patients presenting at hospital. The sample size of the non-randomized studies ranged from 6117 to 855.22 The six studies included in the systematic reviews have samples ranging from 176 to 72,327 participants.6
Interventions and comparators
Devices used to measure skin temperature varied across studies. The Thermofocus and the Exergen thermometers were those used in the non-randomized studies, whereas one study did not report the model of the device3 and the SR included studies with other kinds of devices.6 The number of measurements and the mode of the device (i.e. the algorithm transforming the actual reading into the predicted body temperature) were different or not reported between studies.
The accuracy of handheld infrared skin thermometers was compared with rectal temperature in two studies,21,23 with pulmonary artery catheter temperature in one study,3 with tympanic thermometers in one study,6 with axillary temperature in one study,20 with nasopharyngeal temperature in one study,17 and with a reference that could be either oral, rectal or axillary temperatures in one study.22
Years of publication
The years of publication ranged from 2009 to 2013.
The accuracy of thermal scanners (infrared cameras) for detecting febrile individuals
Six studies evaluated the accuracy of thermal scanners. Of these publications, one was a systematic review6 and five were non-randomized studies.10,11,13,22,24
Country of origin
The systematic review originated from France.6 The non-randomized studies were from Korea,10 China,11 New-Zealand,13 and USA.22,24
Population
The age of the patients included in the studies ranged from 6 months22 to 9224 years and most of the studies reported a similar proportion of males and females. One study only included pediatric patients.22 Studies included inpatients or patients presenting at hospital,6,11,22,24 or travelers presenting at borders.10,13 The sample size of the non-randomized studies ranged from 60810 to 2873.24 The studies included in the systematic reviews have samples ranging from 176 to 72,327 participants.6
Interventions and comparators
Devices used to measure skin temperature varied across studies. The FLIR and the OptoTherm ThermoScreen were those used in the majority of studies.
The accuracy of thermal scanners was compared with tympanic thermometers in four studies,6,10,11,13 with oral temperature in one study24 and with a reference that could be either oral, rectal or axillary temperatures in one study.22
Years of publication
The years of publication ranged from 2009 to 2014.
Summary of Critical Appraisal
The SRs7–9 had a pre-specified protocol, except Bitar et al.6 which did not describe an a priori study design. Two8,9 SRs had study selection performed in duplicate by independent reviewers. The literature search strategy, including grey literature search, was described in two SRs,8,9 whereas Jefferies et al. did not mention a grey literature search7 and Bitar et al. did not perform a comprehensive literature search (only one database search, no mention of grey literature).6 Excluded studies were not disclosed in any of the SRs. A list of the included studies, with their characteristics and an assessment of their individual quality were reported in two SRs,7,8 whereas Zhen et al.9 and Bitar et al.6 did not report individual quality. The conclusions of one SR were in line with the quality of its results,7 whereas Zhen et al. expressed different conclusions that were in contradiction with each other.8 However, it was the only SR8 where publication bias has been assessed. The heterogeneity and the comparability of the data was assessed in the three SRs7–9 where a meta-analysis was planned, even though it could not be performed in on case.7 Conflicts of interest of the included studies were not documented in any of the SRs. Other issues were: subjective inclusion criteria,9 lack of information on the included studies,7 poor statistical methods in the included studies,9 and the non-blinded design of the included studies.6–9
The common strengths of the non-randomized studies were the objectivity of the measurements (i.e. body temperature) and the comparison in a single group. The common limitation of the non-randomized studies is that they all used, at least in part, non-blinded investigators for the assessment of temperatures. Some studies also lacked a sample size calculation,10,11,14,18,19,21,23 a description of the percentage of eligible participants who were enrolled,3,11,14–18,20,21,23 a description of the thermometers used,3,11 statistical tests or P-values,13,14,16 or a description of statistical analyses.11 Moreover, a few studies based their conclusions on very small numbers of feverish subjects10,13,18 or failed to clearly state the outcomes in the introduction.12,23 Cho et al. conducted a retrospective study and the variability of measurements have not been correctly reported.10 Oyakhirome et al. did not use the gold standard comparator (rectal temperature) in all patients and did not state explicit inclusion/exclusion criteria.15 Since Priest et al. only analyzed a small proportion of eligible participants the representativity of their sample is questionable.13 Participant characteristics were not well desribed in both Oyakhirome and Priest studies.13,15 Other issues encountered were conclusions that could not be extrapolated to febrile patients17 and a study where some patients of interest have been excluded.12
The time elapsed between the different measurements was short or nonexistent in twelve studies,3,12,14–21,23,24 but has not been specifically reported in four studies.10,11,13,22. The choices for the reference temperature were reasonable in most of the studies as the utilization of the actual core temperature (pulmonary artery catheter) or the common gold standard (rectal temperature) is not always feasible, depending of the context of the study. Nevertheless, the studies which compared to these two standards are more likely to have an accurate reference for the core body temperature.
Summary of Findings
The accuracy of tympanic thermometers for detecting febrile individuals
The systematic review of pediatric studies from Zhen et al.8 reported a sensitivity of 0.70 (95% confidence interval [CI] 0.68 to 0.72), a specificity of 0.86 (95% CI 0.85 to 0.88), a positive likelihood ratio of 9.14 (95% CI 6.37 to 13.11), a negative likelihood ratio of 0.24 (95% CI 0.17 to 0.34), a diagnostic odds ratio of 47.3 (95% CI 29.79 to 75.18) and an area under the receiver operating characteristic curve (AUROC) of 0.94 when comparing tympanic thermometry with rectal thermometry. The overall pooled mean difference between tympanic and rectal temperature was 0.22°C (95% limits of agreements [LOA] −0.44 to 1.30°C).9 This difference was reduced to 0.15°C (95% LOA −0.32 to 1.10°C) when considering a subgroup of febrile children.9 Compared to pulmonary artery catheter temperature, the SR from Jefferies et al. reported a mean difference from tympamic temperature within the ±0.2 °C range.7 Two SRs expressed conclusions in favor of the utilization of tympanic thermometry, 7,8 whereas one stated that its accuracy (with an LOA spanning over 1.74 °C) is poor.9
Compared with oral temperature, Chue et al reported a mean difference ranging from 0.05°C (95% CI 0.01 to 0.08) to 0.12°C (95% CI 0.07 to 0.17) depending of the investigator.12 Rabanni et al. reported a mean difference of 0.1°C, a correlation of 0.723, a sensitivity of 66%, a specificity of 99.6%, a positive predicitive value (PPV) of 91%, and a negative predictive value (NPV) 98%.18
When comparing with rectal temperature, Barnett et al. reported a mean difference of 0.22°F (95% CI −1.61 to 2.05),14 Smitz et al. found 95% LOA of −0.83 to 0.42°C for ThermoScan and −1.32 to 0.20°C for Genius19 and Oyakhirome reported mean difference of 0.3°C (95% CI 0.2 to 0.3, LOA −1 to 2).15 Across studies, reported sensitivities were of 63.6%,16 74.12%,14 75%15 and 94%;19 reported specificities were of 86.22%,14 95%,15 97.4%16 and 96–98%;19 reported PPV were of 55.26%,14 87.5%,19 85–89%16 and 94%;15 reported NPV were 76%,15 90.5%,16 93.55%14 and 99%;19 reported correlation coefficients were 0.80616 and 0.84–0.9119. Oyakhirome reported an optimal fever cutoff point of 100.2°F with an AUROC of 0.878.14
Compared to pulmonary artery catheter temperature at a threshold of 38.5°C,3 tympanic thermometry had a specificity of 98%, a PPV of 89%, a NPV of 98% and an AUROC of 0.987±0.007.3 Sensitivity was not reported. Compared to nasopharyngeal probe, tympanic thermometry had a mean difference of 0.19°C (95% LOA −0.32 to 0.71) or 0.98°C (95% LOA 0.42 to 1.54) depending on the device.17
Six studies expressed conclusions in favor of the utilization of tympanic thermometry,3,12,16–19 whereas one study stated that the variability of measurements with tympanic termometry was too high.14 One study did not express conclusions in favor or against use of the device.15
The accuracy of handheld infrared skin thermometers for detecting febrile individuals
The systematic review from Bitar et al.6 reported sensitivities ranging from 4.0 to 89.6%, specificities ranging from 75.4 to 99.6%, positive likelihood ratios ranging from 0.9 to 76.0%, negative likelihood ratios ranging from 86.1 to 99.7%, correlation coefficients ranging from 0.25 to 0.71, and AUROC ranging from 0.86 to 0.96 when comparing infrared non-contact thermometers (including both skin thermometers and cameras) with tympanic thermometry. The authors of this SR highlighted the poor scientific evidence available for the utilization of infrared skin thermometers and thermal scanners for mass screening.6
Across studies, comparators were rectal temperature,21,23 axillary temperature,20 pulmonary artery catheter temperature,3 nasopharyngeal probe temperature,17 or either oral, rectal or axillary temperature.22 Teran et al. reported a mean difference of 0.029 ± 0.01°C and of −0.02 ± 0.277°C depending of the model used.23 Chiappini et al. reported a mean difference of 0.11°C.20 Fortuna et al. reported a mean difference of −0.1°F.21 Mangat et al. reported a mean difference of 0.66°C (95% LOA −0.15 to 1.48).17 Reported sensitivities were 76.8%,22 89%20 and 91–97%;23 reported specificities were of 79.4%,22 83%,3 90%20 and 97–99.6%;23 reported PPV were of 47%,3 70%20 and 95.2–99.3%;23 reported NPV were of 96%,3 97%20 and 94.6–98.1%;23 reported correlation coefficients were of 0.48,21 0.66,22 0.83720 and 0.950–0.952;23 reported AUROC were of 0.85222 and 0.853 ± 0.051.3
Three studies expressed conclusions in favor of the utilization of infrared skin thermometry,20,22,23 whereas three studies stated that this type of device is lacking accuracy.3,17,21
The accuracy of thermal scanners (infrared cameras) for detecting febrile individuals
The findings of a systematic review6 that included studies both with infrared skin thermometers and thermal scanners have been described in the previous sub-section.
Nguyen et al. compared thermal scanners with oral thermometers.24 Selent et al. compared with oral, rectal or axillary temperature.22 All other studies compared with tympanic temperature9,13 or tympanic + oral temperatures.11 Cho et al. reported a mean difference of −1.31°C but this was not statistically different from tympanic temperature (P = 0.316)10 whereas Chan et al. reported a mean difference of −3.10°C.11 Reported sensitivities were 87%,11 83.0–83.7%,22 86%13 and 80.0–91.0%;24 reported specificities were 34–43%,11 85.7–86.3%,22 71%13 and 65.0–86.0;24 reported PPV were 10–11%,11 1.5%13 and 5.7–18.4%;24 reported NPV were 97–98%11 and 99.1–99.6%;24 a positive likelihood of 1.33–1.5311 was reported; a negative likelihood of 0.29–0.3711 was reported; reported correlation coefficients were < 0.511 and 0.75–0.78;22 reported AUROC were 0.780–0.815,11 0.922–0.923%22 0.8613 and 0.78–0.96.24
Four studies expressed conclusions in favor of the utilization of thermal scanners for fever detection,10,13,22,24 whereas one study stated that this type of device is unsuitable for this purpose due to a high proportion of false positives.11
Limitations
The most common limitation across studies is that they all used, at least in part, non-blinded investigators for the assessment of temperature, but given the objective nature of temperature measurement, this should not be considered a major biasing limitation. Moreover, for many studies it is not clear if they were powered to find a difference between their devices.10,11,14,18,19,21,23 Also, many studies failed to reveal the percentage of eligible participants who were actually enrolled.3,11,14–18,20,21,23 This is of importance since it is not clear if the samples were representative of the population. The profile of people who refused to participate to the studies has not been described. Therefore, it is plausible that feverish or very ill people might be underestimated in those studies.
Across studies, many potential confounders of body temperature have been mentioned such as sweat, gender, age, the range of temperature, the rater, physical activity, the use of antipyretic drugs and the emotional state, but the list is not exhaustive. It has to be kept in mind that those factors can bias the results of the study reviewed, especially when using non-contact infrared (including tympanic, skin or scanners) thermometers. Many studies were specifically conducted on a pediatric population and one was conducted on an geriatric population. Since age is a potential counfounder,11,18,22 the generalizability of those studies to the adult population is questionable.
As mentioned by Zhen et al.,8,9 there is a lot of heterogeneity in the data between studies. Some specific factors affect the comparability of the studies. Reviewed studies have been using different thermometric devices that, depending of the brand, model and mode used, convert the actual reading to a different output measure following their own algorithm. Also, threshold temperatures for fever varied across studies. Some studies aimed to find the optimal threshold for their device even if it was different by many degrees from the reference.11
