Factors influencing sensitivity of a rapid influenza diagnostic test in a community‐based population of kindergarten through 12th‐grade students: Wisconsin 2015–2020

Abstract Rapid influenza diagnostic tests (RIDTs) have variable sensitivity. In a community‐based population of kindergarten through 12th‐grade (K‐12) students, we assessed factors that may influence RIDT performance using 2368 paired results from Sofia® influenza A + B fluorescent immunoassay and reverse transcription polymerase chain reaction (RT‐PCR). RIDT sensitivity and specificity were 76.1% (95% CI: 72.8–79.1) and 97.2% (96.2–97.9), respectively. Factors associated with sensitivity included runny nose (OR = 3.0, p < 0.001), nasal congestion (1.59, p = 0.045), days from symptom onset (per day; 0.75; p < 0.001), myalgia (0.61; p = 0.014), age (per 5 years; 0.55; p = 0.001), and detection of another virus (0.50; p = 0.043). Understanding these factors can aid in interpreting negative results.


| BACKGROUND
Rapid influenza diagnostic tests (RIDTs) have long been used in clinical settings to identify patients infected with influenza. Studies have shown that point-of-care RIDTs reduce the use of antibiotics and improve clinical outcomes, but in many cases, test performance is unreliable. 1 Although specificity is consistently over 90%, some studies report sensitivity rates as low as 17.8%. 2 New advances in technology have improved overall sensitivity, but performance is still inconsistent and false negatives are fairly common. [3][4][5][6] The COVID-19 pandemic resulted in a federal policy that allowed rapid antigen testing in homes, schools, and other community settings for the first time. This policy shift occurred, however, without formal evaluations of performance characteristics of specific overthe-counter tests in community settings. Many academic institutions conducted routine SARS-CoV-2 rapid antigen testing and experienced high specificity and varying sensitivity, similar to observations in clinical settings when testing for influenza. 2 In the aftermath of the pandemic, we are likely to see the emergence of additional rapid testing technologies intended for school or community use. In addition to bolstering existing disease surveillance systems that are often reliant on clinical data, the near real-time results in densely populated settings like kindergarten-12thgrade (K-12) schools may help contain outbreaks of influenza and other respiratory viruses. Accordingly, it is necessary to evaluate the performance characteristics of rapid tests in community settings.
Such information can be essential to educating the public on how best to interpret results. Thus, we conducted a detailed analysis of the performance characteristics of Sofia ® influenza A + B fluorescent immunoassay (FIA) to identify factors that could influence the sensitivity and specificity of RIDTs for K-12 students in a community setting.

| Data collection
The ORegon CHild Absenteeism due to Respiratory Disease Study (ORCHARDS) is a prospective study of K-12 student absenteeism and acute respiratory illness (ARI) in a community setting. ORCHARDS is ongoing and has been described in depth elsewhere. 7,8 The study is based in the Oregon School District, located in Southcentral Wisconsin. The presence of influenza within the school district is verified through home visits where research staff record demographic and symptom information and collect a nasal specimen and an oropharyngeal or a nasopharyngeal (NP/OP) specimen from eligible students. To participate, children must be experiencing at least two ARI symptoms that began within 7 days of a parent or guardian calling the study phone line, as well as a Jackson score of at least 2. 9 Research staff perform a RIDT on the student's nasal specimen

| Analysis
Our analysis was performed with R version 4.02 and used the same approach as our previously published analysis examining the performance of RIDT in a clinical setting. 10 Binomial logistic regression models were used for adjusted associations. Adjusted models were fit for predicting improvements in sensitivity and specificity based on age, gender, number of days from symptom onset, presence and absence of individual signs and symptoms (chills, cough, sore throat, malaise, myalgia, runny nose, nasal congestion, and headache), presence of an influenza-like illness (ILI), illness severity (mild, moderate, or severe), influenza vaccination status, co-detection of other pathogens, and season (early, peak, or late). Specimens meeting inclusion criteria that had missing information were included in the overall sensitivity and specificity estimates, but they were removed from the adjusted model. We defined ILI as having a fever with cough and/or sore throat [23]. A quadratic term for age was not included because it did not significantly improve the model (LRT p = 0.46). Interaction terms between age and days from onset were also considered but ultimately excluded because of lack of significance (LRT p = 0.11).
Some students had multiple specimens present in the dataset. Specimens, even from the same student, were assumed to be independent for testing and modeling.
T A B L E 1 Demographics and distribution of sample characteristics for K-12 students

| RESULTS
Of the 2378 total student ARI episodes between January 5, 2015 and March 12, 2020, 2368 (99.6%) were included in this analysis. Nine specimens (0.38%) were excluded from the overall analysis because the child's symptoms began more than 7 days before the samples were collected, and one (0.04%) was excluded because an RIDT was not performed. For the adjusted model, 17 records were excluded because they were missing the severity of illness (n = 12; 0.50%) or were marked as missing the number of days from symptom onset because of an invalid test date (n = 5; 0.21%). Three (0.13%) specimens had inconclusive Sofia ® test results. These were imputed as

| Participant characteristics
The average age of participants was 10.2 years with slightly more males than females represented (Table 1). More than half of the participants reported symptoms that met the definition of an ILI. Cough, malaise, and nasal congestion were reported in over 80% of participants, whereas runny nose and sore throat were reported in more than 70%. Myalgia was the least common symptom and was reported in 28% of participants. Slightly more than half of the children (52.4%) were vaccinated against influenza. Influenza was confirmed by RT-PCR in 710 (30.0%) specimens (influenza A: n = 443; influenza B: n = 263; dual detection: n = 4).

| Rapid influenza diagnostic test performance
Overall sensitivity and specificity of Sofia ® were 76.1% (95% CI: 72.8-79.1) and 97.2% (96.2-97.9), respectively, with slightly higher sensitivity for influenza A than for influenza B (Table 2) Sensitivity was highest (84.2%) 1 day after symptoms began and dropped considerably by the fourth day (52.4%; Figure 2). Fever was excluded from the adjusted analysis because the variance inflation factor was 18.09, meaning it was highly collinear with other variables. None of the factors explored in the adjusted model were associated with specificity.

| DISCUSSION
Among a population of K-12 students evaluated in a community-based setting, Sofia ® demonstrated high specificity and moderately high sensitivity. In an adjusted analysis, we found that older age, increased days from illness onset, the presence of additional viruses, and the presence of myalgia could result in lower sensitivity. Alternatively, the presence of runny nose and nasal congestion could improve sensitivity.
The effect of age on RIDT sensitivity is well documented, and our results demonstrate that this holds true even among children in a community setting. 3 Previous studies have also noted the importance of the duration between when symptoms started and when the test was performed. 11 The effect of individual symptoms on test sensitivity is not well understood, but our previous clinic-based study also showed that the presence of a runny nose improves sensitivity. 10 One explanation for this could be that having a runny nose or nasal congestion is correlated with a greater quantity of viral antigen within the sample collected. Absence of these symptoms may help guide the interpretation of negative results.
Our study benefited from a large sample recruited from a community setting over six influenza seasons, which is underrepresented in   The US Office of Management and Budget approved all forms used in this study.

PARTICIPANT CONSENT
Written informed consent was obtained from parents/guardians of minor students and from students aged ≥18 years.

PEER REVIEW
The peer review history for this article is available at https://publons.