Since the first case of coronavirus disease 2019 (COVID-19) was diagnosed in December 2019, the COVID-19 pandemic has evolved with the appearance of several variants (Alpha, Beta, Gamma, Delta and Omicron) (1–6) which affected billions of people around the world and became a significant global public health crisis.

Considering that, unlike previous strikes of COVID-19, the current epidemic is small-scale and cannot be eliminated in a short time, it is not feasible to adopt severe measures such as nationwide nucleic acid measurement and city-wide blockades (7–10). To identify and exclude close contacts of COVID-19 infectors is a difficult but necessary challenge that must be undertaken. Too strict COVID-19 management policy will lead to stagnation or even retrogression of social and economic development (11). Screening out the possible COVID-19 infectors without affecting the social order is conducive to the simultaneous development of economy while ensuring people's health (12, 13). Fever screening is an effective and low-cost method to detect infectors associated with different variants of COVID-19 based on the fact that most infectors with COVID-19 have fever symptoms (14–17). Among 55,924 laboratory confirmed cases of COVID-19 in China, 87.9% had a “fever” (no cut-off point reported). Therefore, the WHO announced that temperature screening was sufficient to detect the majority of COVID-19 cases, and recommended that fever screening checkpoints be positioned at the entrance of public places for screening out potential COVID-19 infectors (18).

Rapid testing, accuracy, and convenience are the key features required for fever screening (19, 20). Compared with mercury axillary thermometers (MATs), non-contact infrared thermometers (NCITs) are more suitable for fever screening because of the rapidity and convenience (21, 22). Nevertheless, authoritative data is lacking in defining “fever” at different body surface sites when using NCITs. Previous studies indicated that the temperature measurements of NCITs were influenced by body surface measurement sites, ambient temperatures and instruments (23–26). Meanwhile, regarding differences in religious beliefs and modes of dressing habits, as well as certain personal factors (local trauma or disability, for example), it is essential to determine the optimal diagnostic fever thresholds at different body surface measurement sites.

In our previous preliminary study (27), we have calculated the theoretical optimal diagnostic threshold (the maximal sum of sensitivity and specificity) at each body surface site. For instance, when 37.4 °C at neck was used as fever threshold, the sensitivity of fever screening was 0.866 while the specificity was 0.846. However, a sensitivity of 0.866 does not meet the criteria for fever screening. In the actual fever screening, we tend to reduce the fever threshold to improve the sensitivity. Meanwhile, for advanced age and frailty can blunt febrile responses to illness, the temperature used to define fever can influence the clinical recognition of COVID-19 symptoms. A lower fever threshold would increase the number of residents recognized as having symptomatic infection, potentially leading to earlier detection and more rapid implementation of therapeutic interventions and infection prevention and control measures (28, 29).

Hence, in this study, we attempted to ascertain the optimal diagnostic threshold for fever screening using NICTs at different body surface sites, unify fever reference standards for improving accuracy and standardization of fever screening under COVID-19.

The study initially recruited 2,013 participants. However, 153 participants who withdrew or whose data was completely lost were excluded. The final analysis included 1,860 participants. The participants' characteristics are summarized in Table 1. The mean age of participants was 3.45 ± 2.85 years for children and 28.56 ± 7.25 years for adults. In addition 1,304 (70.1%) participants were children (≤12), and 683 (36.7%) were male because we are a specialized hospital for women and children. Among the participants, 823 (44.2%) were tentatively diagnosed with fever based on axillary temperature measured by MATs ≥ 37.3°C. Of these, 691 (84.0%) were children and 132 (16.0%) were teens and adults. The humidity ranged in 30–90 %RH (mean: 58.30 ± 12.90) and ambient temperature ranged in 9°–32°C (mean: 22.5°C ± 4.09 °C).

Temperatures at the wrist, forehead, temple and neck measured by NCITs and axillary temperatures measured by MATs were compared. The results revealed that the neck temperature was closest to the axillary temperature, whereas the wrist, forehead and temple temperatures lower than that (Table 1). Among several major participant populations, the average body temperature in children was higher than that of teens and adults (p < 0.001), and that of pregnant women was slightly higher than that of non-pregnant women (p = 0.027) (Figure 2A); the temperature of females was slightly higher than that of males, but with no statistical difference.

Axillary temperature of ≥ 37.3 °C was used as the diagnostic criterion of fever, ROC curves were used to reflect the accuracy of NCITs at different body surface sites in different groups (Figure 1). Surprisingly, when all participants were involved in the ROC curve, the maximum AUC was 0.922 among the four body surface sites. The maximum specificity was only 0.623 when the sensitivity fulfilled the requirements of fever screening (single sensitivity > 0.95) (Table 3, Figure 1A). Due to the difference of average body temperature among different groups, ROC analysis was conducted respectively. Remarkably, the accuracy of NCITs varied greatly among different groups of individuals. In children, the highest AUC of body surface sites with NCITs was the neck (0.902) whereas AUC of neck, temple and forehead in teens and adults (female) were 0.981, 0.973 and 0.961 respectively (Figures 1B,C). Among teens and adults (female), the accuracy of wrist temperature was the lowest; the AUC was only 0.887 (Table 2, Supplementary Figures 1A–D).

Considering the vast majority of patients with fever are children, and the uncertainty of the accuracy of axillary temperature in children, linear correlation analysis was performed to analyze the correlation between non-febrile participant ages and neck temperature (Figure 2B). Our results revealed that non-febrile participant ages and their neck temperature were negatively correlated (p < 0.001). The effect of non-febrile participant covariates on neck temperature was further analyzed using multiple linear regression. Similar to the results of univariate analysis, neck temperature was significantly associated with participants' age and pregnancy (beta-value: 0.372, 95% CI: 0.145, 0.599. P = 0.002), but was not found to have a correlation with ambient temperature, sex, investigators, mode of transportation. The neck temperature was negatively correlated with participant's age; neck temperature decreased by 0.03 °C when the participant's age increased by 1 year (beta-value: −0.030, 95% CI: −0.036, −0.024; P < 0.001) (Table 3).

The fever diagnostic thresholds at the four body surface measurement sites were listed according to Youden index (Table 4). The optimal fever diagnostic threshold needs to fulfill the requirements of fever screening (single sensitivity > 0.95) and display as high specificity as possible. According to this principle, the optimal diagnostic thresholds of fever at these four body surface sites are 36.75 °C at the neck (Youden index: 0.851, sensitivity: 0.993, specificity: 0.858), 36.55 °C at the temple (Youden index: 0.848, sensitivity: 0.974, specificity: 0.874), 36.45 °C at the forehead (Youden index: 0.744, sensitivity: 0.961, specificity: 0.813), and 36.15 °C at the wrist (Youden index: 0.385, sensitivity: 0.951, specificity: 0.434). Meanwhile, the diagnostic threshold of fever with NCITs at the neck was 36.7–37.8 °C, which all exhibited quite high accuracy (Supplementary Table 2).

The body temperature detected by NCITs needs to be 5 cm away from the body surface, therefore, the effect of ambient temperature on NCITs was further evaluated (Supplementary Figure 2). When the ambient temperature is below 18 °C, the accuracy of NCITs decreases as ambient temperature decreases. Nevertheless, neck temperature is less affected by ambient temperature., the AUC of neck temperature was still high at 9°C (0.923) and commonly remains above 0.950 when ambient temperature fluctuated between 9–29 °C.

In addition, the participants' preference data on body temperature measurement sites from the standardized questionnaire were counted (Supplementary Table 1). Participants preferred to measure their temperature at their wrists or foreheads, mostly for “convenience” and “speed.” More than half of the participants (54.2%) were indifferent to body temperature measurement sites whereas 457 (25.2%) disliked axillary temperature due to “inconvenience” and “time-consuming.” It is noteworthy that our investigators also disliked axillary temperature and were uncertain about the accuracy of axillary temperature measurement with MATs for children because the children did not cooperate well when having the temperatures taken. Failure of body temperature measurement with MATs on children is common.

This further study collected more fever screening data and attempted to ascertain the optimum diagnostic threshold of fever with NICTs at different body surface measurement sites for improving the standardization and accuracy in fever screening under COVID-19.

Consistent with previous studies, in our study, the average body temperature was inconsistent at different body surface measurement sites and different population groups (33, 34). The average body temperature of children was higher than that of teenagers and adults, and that of pregnant women was slightly higher than that of non-pregnant women. This is obvious because children experience more vigorous physiological metabolism and pregnant women undertake the metabolism of the fetus (35, 36). Meanwhile, other factors affecting body temperature [ovulation cycle and circadian rhythm (37, 38), for example] were not involved in this study. However, the method of fever screening should have universal applicability to meet the screening requirements in various situations. Research based on the real-world practice may also meet the universality of fever screening standards.

In our study, the temperatures at the forehead, temples and wrists measured by NCITs were lower than the axillary temperatures measured by MATs whereas the neck temperature was slightly higher; these results are similar to those of previous studies (30, 39). This is obvious because the forehead and temple are bare and are more likely to be affected by ambient temperature. Meanwhile, the wrist is away from the core of the body, resulting in less heat supply. Hence, the higher accuracy of neck temperature can be explained. NCITs are indeed more affected by ambient temperature than MATs because they are not in direct contact with the skin. Moreover, different body surface sites are differently affected by ambient temperatures. As mentioned above, the forehead and temple sites are more affected due to exposure with the neck wrapped in a collar. Therefore, the accuracy of neck temperature is high under different ambient temperatures.

It is worth mentioning that there are also some factors affecting the measurement results in the use of NCITs, although they have been controlled in this study, such as the distance from the skin, the measurement angle and the ambient light intensity (40, 41). According to the standardized body temperature measurement process, we strictly measured body temperature at a distance of 5 cm perpendicular to the skin in a same semi-enclosed artificial light source measurement environment. The body surface temperature measured at same body surface site with NCITs in different studies could be different, which may fluctuate between 34 and 37 °C (42–45). As we said, the accuracy of NCITs may vary according to ambient temperature, measurement distance, the measurement angle, the ambient light intensity, model and manufacturer, etc., the measurement result even has an error of more than 1°C (41). That's why we strictly measured body temperature at a distance of 5cm perpendicular to the skin in a same semi-enclosed artificial light source measurement environment according to the standardized body temperature measurement process. When NCITs are used for fever screening, they shall be measured perpendicular to the surface in strict accordance with their recommended measurement distance in a suitable artificial light environment to improve the accuracy of the results. Meanwhile, a previous study has shown that the additional monitoring of ambient temperature can be strongly helpful to improve the accuracy in the detection of human body temperature, the measurement of which can be affected by the environmental humidity (46). This means that monitoring environment humidity while measuring body temperature or using instruments that can simultaneously measure body temperature and humidity may also increase the accuracy of body temperature measurement.

From our findings, there seems to be a significant difference in the accuracy of NCITs between children and adults. However, the criterion for the accuracy of NCITs is the axillary temperature measured by MATs. Previous studies revealed that the accuracy of MATs in measuring axillary temperature of children has been questioned (47, 48). Because MATs require subjects to clamp their arms for a long time, it is difficult for active children. Moreover, due to the rapidity of NCITs and the researchers' measurement in strict accordance with the standardized process, the obtained temperature data of NCITs were highly accurate in adults. There is reason to believe that NCITs has the same high accuracy in children, and it may be a more accurate method of child thermometry.

In our previous preliminary studies, we have calculated the theoretical optimal diagnostic threshold (maximal sum of sensitivity and specificity) of each body surface site (27). For example, when 37.4 °C at the neck is the diagnostic threshold of fever, the sensitivity is 0.866 while specificity is 0.846. However, the sensitivity of 0.866 does not meet the criteria for fever screening. In actual fever screening, we tend to reduce the fever threshold to improve the sensitivity. Moreover, in the preliminary study, the inaccuracy of children's axillary temperature data was not taken into account, which may slightly affect the determination of the optimal threshold. In this study, we determined the optimal fever diagnostic thresholds at the four body surface measurement sites when the sensitivity fulfilled the requirements for fever screening (single sensitivity > 0.95). It is noteworthy that wrist temperature does not appear to be suitable for fever screening due to its low accuracy (49–51).

Although this is the third year of the COVID-19 outbreak, COVID-19 is still quite threatening due to the potential for several dangerous variants to evolve. It is therefore crucial to explore further measures for pandemic prevention and control, which may prepare for potential variant of COVID-19 or future other virus pandemics (13). Fever screening is an effective and low-cost method to detect infectors associated with different variants of COVID-19 or future other virus based on the fact that most infectors with virus infection have fever symptoms (11). Determining the optimal diagnostic threshold for fever screening using NICTs at different body surface sites is instrumental in improving the accuracy of fever screening and providing theoretical reference for healthcare policy. Vaccinate before outbreak of the epidemic (52, 56). And in case of sporadic outbreak, the possible infectors can be found as soon as possible through fever screening and nucleic acid detection, so as to avoid the occurrence of pandemic.

This study is large-scale and with highly accurate data attributed to the standardized body temperature measurement process by the same investigators using the same instruments. Compared with the previous NCITs studies whose main participants had normal body temperatures, nearly half of the participants had “fever,” which can provide a more optimal diagnostic threshold of fever.

This study adds to our understanding of the optimal fever diagnostic thresholds of NCITs at the wrist, forehead, temple and neck under COVID-19. However, there are some limitations that should be considered. First, ambient temperature fluctuated between 9 and 32°C during the study period. Therefore, the accuracy of NCITs beyond these ambient temperatures cannot be assumed. Second, although fever screening is an effective and low-cost method to detect infectors associated with different variants of COVID-19 since most infectors with COVID-19 have fever symptoms, fever screening cannot screen out all infected persons. Third, the study only involved body surface sites commonly used in fever screening in China, and other body surface sites deemed to have high accuracy, like the inner ear and canthus (53–55), were not included. Meanwhile, the circadian rhythm and female menstrual cycle may also be influencing factors of body temperature, but this study is conducted in the outpatient and emergency department, so the patient's temperature cannot be measured continuously, the effect of the circadian rhythm and female menstrual cycle on body temperature cannot be analyzed.

Although our study confirmed the high accuracy of NCITs in fever screening when used correctly, it should be noted that NCITs need to be measured perpendicular to the surface in strict accordance with their recommended measurement distance in a suitable artificial light environment to ensure the accuracy of the results. Given the findings of our study, we recommend 36.15, 36.45, 36.55, and 36.75 °C as the diagnostic thresholds of fever at the wrist, forehead, temple and neck, respectively. Among the four surface sites, neck temperature exhibited the highest accuracy and stability. These results are instrumental in improving the accuracy of fever screening and providing theoretical reference for healthcare policy.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by Ethics Committee of the Chengdu Women's and Children's Central Hospital (No. 202040). Written informed consent to participate in this study was provided by the participants' legal guardian/next of kin.

XY, YLin, and YLi contributed to the policy-making and administrative work in this study. FL conceived and designed this study and also joined in the data collection. XL and TL analyzed, interpreted, and visualized the data. XW, QW, SC, SW, YX, QH, XZ, and YY participated in the data collection and analysis. All co-authors contributed to the drafting and revising of this research and showed their approval for the publication of the article. All authors agree to be accountable for the accuracy and integrity of this work.

Financial support of this work was provided by Chengdu Science and Technology Bureau (22ZDYF1597, 2021-YF09-00048-SN, 2022-YF05-01326-SN, and 2020-YF05-00145-SN), Chengdu Municipal Health Commission (Nos. 2020216 and 2022441), and Chengdu Women's and Children's Central Hospital (No. 2020JC01).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.