Feasibility and effectiveness of daily temperature screening to detect COVID-19 in a prospective cohort at a large public university

SARS-CoV-2 is a novel coronavirus that causes coronavirus disease 2019, otherwise known as COVID-19 [1]. One of the most challenging features of the COVID-19 epidemic to date is considerable pre-symptomatic [2] and asymptomatic transmission [3, 4], currently estimated to comprise anywhere from 6 to 41% of infectious individuals [5]; many other individuals may experience only mild symptoms. This presents barriers to epidemic control by impeding rapid isolation of cases, and makes it necessary to develop nuanced screening approaches that will both limit infectious exposure and be a useful trigger for SARS-CoV-2 testing. Accordingly, a strong desire to have students and employees return in person to school or work has driven workplaces, businesses, and colleges to look for methods to rapidly assess risk of infection and prevent entry for those who are possibly infectious to others.

One common strategy to prevent transmission has been temperature checks, which have increased in popularity as a non-invasive measure to rapidly screen individuals for elevated body temperature (i.e., fever). Temperature screening has been used during other global outbreaks, including the severe acute respiratory syndrome (SARS) outbreak in 2003, the H1N1 influenza epidemic in 2009 [6], and recent major Ebola virus disease outbreaks in Sub-Saharan Africa in 2014 and 2018 [7]. However, multiple studies have found low sensitivity or specificity of temperature monitoring during these prior epidemics [8, 9], even in cases where fever was a very common symptom among people with the disease in question [6, 10, 11]. Large infrared fever screening systems or no-contact temperature screening at building entrances [12] and hospital entryways [13] and wearable devices to continuously monitor individual temperature [14] have all been widely employed in an attempt to prevent the spread of SARS-CoV-2, yet limited evidence exists concerning the sensitivity and specificity of these approaches for the current pandemic [15], especially on a university campus. Of note, the United States Food and Drug Administration (FDA) released a statement in June 2020 noting that non-contact temperature assessment devices “are not effective if used as the only means of detecting a COVID-19 infection” [16].

Nonetheless, a number of college campuses are now implementing systems that require students, faculty, and staff to attest that they are free of symptoms, and/or are afebrile before coming to campus [17‐21]. These policies almost universally rely on dichotomous temperature cutoffs for fever, e.g. temperature ≥ 100.4 °F, in alignment with the CDC guidelines [22]. Such strategies have numerous pitfalls, including reliance on an outdated sense of “normal” body temperature [23], disregard of the effect of time of day [24] and ambient temperature on body temperature [25], disregard of numerous studies that have found meaningful variation in normal body temperature between individuals [26, 27], and incentives to not voluntarily disclose symptoms in order to preserve access to work spaces and therefore safeguard career and financial stability [28].

Given the rapid increase in reliance on temperature-based strategies to restrict campus access for prevention of COVID-19 spread, we aimed to assess the feasibility, acceptability, and effectiveness of temperature monitoring and other fever-based strategies to prevent spread of SARS-CoV-2 in a longitudinal cohort of 2916 university-affiliated students and employees, known as the Berkeley COVID-19 Safe Campus Initiative.

We found that daily temperature monitoring to screen for SARS-CoV-2 infection was acceptable to a wide variety of people affiliated with a large public university. While only 30.4% of participants recorded their temperature every day they completed at least some of the daily survey, on average participants completed 40 daily surveys (median: 40, IQR: 30–53) during the study period, with 77 participants taking 70 or more daily surveys during a study period that was 78 days from day 1 of rolling enrollment through study end. Particularly among students (who were enrolled for substantially longer than non-students on average), the number of surveys completed per day started to decrease in early August (see Supplemental Figures 1 and 2). It is expected that many people would eventually look for efficiencies in their study participation, including frequently completing a very brief symptoms survey without taking the extra step to take their temperature and record the quantitative reading.

We also found daily temperature monitoring to be feasible for participants; however, almost 3 out of 4 people, particularly students, did not already own thermometers, and purchasing and disseminating thermometers of sufficient quality during a global pandemic proved nearly impossible: taken together, these thermometers cost over $18,200 just to cover the needs of our study participants, calling into question the feasibility of this strategy on a campus-wide level. This highlights the challenges of accessing this type of equipment during a true public health emergency (particularly a global one), and underscores one of the challenges to requiring self-monitoring of body temperature as a condition of entry to a college campus, i.e. it is likely unrealistic to expect accurate responses to required self-attestations of being fever and symptom-free, especially for students.

Overall, the sensitivity of temperature or symptoms screening in our study never rose above 40.5%, and the best-performing quantitative temperature threshold had a sensitivity of only 30.6%, but considered fever to be anything at or above 98.7 °F, which would be considered normal body temperature in most settings and resulted in a positive predictive value of only 3.5% (95% CI 2.2–5.8%). These findings align well with recently published results from an Australian hospital, where fever ≥100.4 °F was only detected in 16 of 86 patients testing positive for SARS-CoV-2 over a 2-month period (sensitivity of 19, 95% CI 11–28%), when using a variety of in-hospital temperature measurement methods, mostly with temporal thermometers [38]. While their sensitivity using this fever threshold was higher than we observed in this study, this is unsurprising given that these were among hospital patients, not people who were mostly feeling well at time of testing. While the specificity of the strategies tested in our study was reasonably good (> 93% in all cases), this study was conducted during the summer months, well before the typical influenza season in the San Francisco Bay Area. Specificity of temperature monitoring for SARS-CoV-2 can be expected to worsen in the fall and winter months, when the prevalence of influenza-like illness for reasons other than infection with SARS-CoV-2 is substantially increased.

There were significant associations between measured body temperature and PCR test results in longitudinal mixed models when controlling for individual variation in body temperatures; however, these regression models were designed to account for within-person variation in baseline temperature that is not scalable to real-life monitoring strategies, and does not apply to the most common temperature monitoring policies, which use dichotomous fever thresholds for permitting building entry or triggering symptoms-based SARS-CoV-2 testing, rather than a more sophisticated and individualized algorithm. Further, the performance of self-reported qualitative fever symptoms (“feeling feverish”) in this study suggests that daily quantitative temperature monitoring may not offer additional protection as a screening measure to prevent SARS-CoV-2 transmission on a college campus. Research by others that has found substantial measurement error with regard to body temperature [26] bolsters the idea that qualitative self-report may be a preferable strategy.

Our study had a number of limitations. First, temperature data included here could be inaccurate for multiple reasons, including error using the thermometer or reporting error, such as participants not recording their temperature despite taking it or recording the temperature incorrectly. However, such limitations also apply to any scaled, non-research campus-level system for temperature monitoring. Second, while participants were asked to take their temperature in the morning, we did not control for time of day in this analysis, because there was substantial information bias in the daily survey timestamp (i.e. some participants anecdotally reported taking their temperature in the morning as requested but completing the survey at a later time). This may have made us less likely to detect fevers among our participants [24]; however, since many school or work-based temperature screenings likely take place in the morning, this limitation would also apply to any scaled campus-level temperature screening system. Similarly, we did not control for differences in thermometer type in our analysis, given that the large majority of participants had an oral thermometer, despite the fact that type of thermometer used is related to mean body temperature measurements [39]. However, this too is a limitation that would apply in real-life application of any campus-level temperature monitoring strategy. Third, individual basal temperature is known to vary by fertility cycle for women not taking hormones [40]; however, we did not collect information about birth control or other hormone use during this study, so could not account for this in our analysis. A sensitivity analysis (not shown) stratifying our adjusted regression model by sex showed nearly identical results for women compared to the overall study population, suggesting that this hormonal temperature variation was not an important factor. Fourth, our results related to feasibility and acceptability of temperature monitoring were likely biased by the incentives provided to participants to encourage daily monitoring, as well as the enthusiasm for such activities among the type of person who would enroll in an intensive study of this nature, compared to a general campus population. Finally, it is possible that some participants had SARS-CoV-2 infection that was not detected by PCR testing, and therefore our effectiveness results may be biased in an unpredictable direction.

Given the popularity of temperature monitoring as a non-invasive measure for rapid screening for SARS-CoV-2 infection before allowing students or employees access to college campuses, workplaces, or congregate venues, our findings offer important insight into the insufficiency of such methods on college campuses, despite widespread adherence to and acceptability of daily temperature monitoring strategies.

In concordance with evidence from prior global pandemics [6, 8‐11], our findings suggest temperature screening using dichotomous fever thresholds during the COVID-19 pandemic may be little more than “security theater” [41]. This term, largely attributed to computer security expert Bruce Schneier, describes measures that provide a sense of security, despite having no actual positive impact on security. Temperature monitoring strategies may be a rational response to community calls for action, even if those demands for action are based on inaccurate or outdated information about the risks and effective strategies for COVID-19 mitigation [42, 43]. Other strategies to disinfect or otherwise prevent fomite transmission have also been implemented during the COVID-19 pandemic, often at great cost and with questionable preventative benefit [44‐46]. Millions of dollars are invested in similar “security theater” measures each year in the United States, from radio-frequency identification (RFID) ankle bracelets to prevent hospital newborn abductions to removing shoes and prohibiting liquids above 3.4 oz at airport security [47, 48], in order to reassure the public that sufficient action is being taken in the face of a serious threat. Measures like this can have some benefit; namely, some individuals with SARS-CoV-2 would undoubtedly be detected by temperature screening measures and prevented from transmitting the virus to others. Further, routine temperature monitoring can reassure members of a campus community or workplace that those in charge care about their health and are taking the pandemic seriously. Yet an ineffective action can be harmful, if people believe that because temperature screening measures are in effect they are safe, and therefore other measures to prevent spread of SARS-CoV-2 – such as wearing a face covering and practicing social distancing – are not taken seriously [49].

Further attention is needed to the benefits and drawbacks of various strategies to detect and prevent transmission of SARS-CoV-2 on college campuses and in similar close-knit communities, to ensure the community is fully aware of the practical and behavioral implications of any strategies employed. Colleges may want to consider using qualitative measures for self-report of feverishness as a symptom rather than attempting quantitative temperature monitoring, and focus resources on additional strategies known to prevent SARS-CoV-2 transmission, such as low-barrier access to SARS-CoV-2 testing without requiring disclosure of specific risks or symptoms [50], rather than relying on temperature screening or self-attestations of good health as a condition of campus access.