Impact of Fever Thresholds in Detection of COVID-19 in Department of Veterans Affairs Community Living Center residents

Taissa Bej MS; Sonya Kothadia MD, MS; Brigid M. Wilson PhD; Sunah Song MS; Janet M. Briggs RN, NP; Richard E. Banks BS; Curtis J. Donskey MD; Federico Perez MD, MS; Robin L. P. Jump MD, PhD


J Am Geriatr Soc. 2021;69(11):3044-3050. 

In This Article


Study Design, Setting, and Data Sources

We conducted a retrospective cohort study of residents living in any of 134 VA CLCs between March 1 and November 30, 2020. On March 17, 2020 the VA issued a memorandum requiring daily screening of residents for temperature >100.4°F, cough, shortness of breath, or sore throat. Screening residents for signs and symptoms of COVID-19 infection continued throughout the study period; providers were notified of residents with a positive screen and could choose to order a SARS-CoV-2 diagnostic test as clinically appropriate. A memorandum issued on April 14, 2020 called for widespread baseline testing of all CLC residents for SARS-CoV-2. On June 11, in alignment with guidance issued by the Centers for Medicare and Medicaid Services (CMS), the VA issued a memorandum for CLCs to test residents for COVID-19 within 48 h of admission, following detection of a new confirmed case in the CLC, and weekly thereafter until no new cases were identified and at least 14 days passed since the most recent positive test.[9]

We used the Department of Veterans Affairs Informatics and Computing Infrastructure to access clinical databases from the U.S. Veterans Healthcare Administration (VHA). Data were extracted from the VHA's Corporate Data Warehouse, the VHA's Vital Status File, and the VA COVID-19 Shared Data Resource. The Institutional Review Board at the VA Northeast Ohio Healthcare System approved the study protocol.

Case Ascertainment and Clinical Characteristics

The cohort studies included all VA CLC residents tested for SARS-CoV-2 using an RT-PCR-based assay. If residents had a temperature of >100.4°F on the day of or within the 14 days before a positive SARS-CoV-2 test, they were considered symptomatic. If their temperature was >100.4°F in the 14 days following a positive SARS-CoV-2 tests, residents were considered pre-symptomatic at the time of testing. Asymptomatic residents had a temperature of ≥100.4°F in the 2 weeks before and after a positive SARS-CoV-2 test. In accordance with the recognition that nursing home residents may have a blunted febrile response, we also conducted analyses using >99.0°F as the temperature threshold for fever.[10,11] To asses for sensitivity and specificity, we considered all tests performed in CLC residents, excluding tests following first positive result in residents who tested positive in between March 1, 2020 and November 30, 2020. Additionally, we assessed age, gender, self-reported race and ethnicity, underlying comorbid conditions, and the Charlson comorbidity index (CCI) based on International Classification of Diseases (ICD) codes.[12] For residents with a positive SARS-CoV-2 test, all-cause mortality was evaluated at 30 days following their first positive test. For those with only negative test results, the date of the SARS-CoV-2 test closest to the start of the study period was used.

Statistical Methods

Differences in continuous values across patient groups were assessed using F-tests, and post-hoc pairwise tests of differences are presented with Tukey-adjusted p-values. Kaplan–Meier survival curves were estimated for CLC residents based on their categorization as being asymptomatic, pre-symptomatic, or asymptomatic at the time of their first positive SARS-CoV-2 test as well as for those with negative SARS-CoV-2 tests. Time at risk for residents who were negative for SARS-CoV-2 during the study period (March–November) and who later tested positive was censored at the date of the positive test. The survival curves were compared using an omnibus log-rank test and post-hoc pairwise comparisons with p-values adjusted using the Holm method. Statistical analyses were performed using R (version 3.5.1; Vienna, Austria)[13] including functions from additional packages.[14–18] Diagnostic test evaluations were performed using MedCalc (Ostend, Belgium).[19]