Principal Component Patterns of Pediatric Respiratory Viral Testing Across Health Care Settings Open Access

We conducted a secondary analysis using data collected from an ongoing ARI surveillance as part of the Centers for Disease Control and Prevention–funded New Vaccine Surveillance Network^9,10  in Nashville, Tennessee, between November 2, 2017, and February 28, 2020. Year-round recruitment at a single southern children’s hospital occurred 4 days/week in the ED and 7 days/week in the inpatient setting. Children who were enrolled in the ED and subsequently admitted to the hospital were classified as an inpatient. Institutional review boards at our institution and the Centers for Disease Control and Prevention reviewed and approved this study (see 45 C.F.R. part 46; 21 C.F.R. part 56).¹¹ 

Eligible children were younger than age 18 years, resided within the catchment/surveillance area (ie, Davidson, Williamson, Cheatham, Dickson, Rutherford, Montgomery, Roberson, and Sumner counties), and presented to an ED or were hospitalized within 48 hours of enrollment with 1 or more of the following symptoms within 14 days of onset: fever, cough, earache, nasal congestion, runny nose, sore throat, posttussive vomiting, wheezing, shortness of breath/rapid or shallow breathing, myalgia, apnea, apparent-life threatening event, or brief resolved unexplained event (eg, unresponsive; Supplemental Fig 3).^9,10  A child was ineligible if any of the following criteria were met: chemotherapy-associated fever and neutropenia (absolute neutrophil count <500/µL), newborns never discharged since birth, transferred from another hospital >48 hours after initial admission, recently hospitalized (≤5 days), or previously enrolled in the study within the past 14 days.^9,10 

In addition, we excluded infants age <30 days because of the increased risk of serious bacterial infections compared with other groups^6,12  (eg, group B Streptococcus) and unique clinical management/practices. Children with multiple ARI visits were considered susceptible to a new ARI infection (either viral or bacterial) if each visit was ≥90 days from the previous visit.

After informed consent, trained research personnel interviewed the child’s parent/guardian to collect demographic information (age, sex, race, and ethnicity), clinical symptoms, and medical history using a standardized case report form. Race and ethnicity were included as social construct variables to understand potential health care disparities driven by interpersonal and systemic racism and bias. Physical examination findings, vital signs at the time of presentation, and provider-ordered viral testing (ie, rapid antigen and NAAT) were abstracted from medical charts. Final clinical disposition was determined by the highest level of care provided to the child during the visit (ie, children enrolled in the ED and later hospitalized were considered an inpatient). During the study period, providers at our institution did not have any diagnostic stewardship programs or testing cost information guiding their decisions and were able to order the following viral tests: rapid influenza antigen, rapid RSV antigen, and NAAT testing. Demographic, interview, medical record, and laboratory results were maintained in a secure REDCap (Research Electronic Data Capture, Vanderbilt University, Nashville, TN) database.¹³ 

Our main outcome was a provider-ordered viral test, which included at least 1 of the following: rapid influenza antigen, rapid RSV antigen, and/or NAAT (ie, BioFire FilmArray Respiratory Panel); dichotomized into tested versus not tested.

A principal components analysis (PCA; a dimensionality-reduction method) was used to derive different presentation patterns (ie, principal components—the set of intercorrelated variables that embody the greatest amount of variance [information]) using the following variables: age (years), illness duration (days), maximum temperature (none, moderate [≥100.4 °F to <102.0 °F], or severe [≥102.0 °F]), pulmonary symptoms (cough, nasal congestion, rhinorrhea, self-reported wheezing, and shortness of breath; yes or no), number of nonpulmonary symptoms (fatigue, earache, myalgia, diarrhea, vomiting, sore throat), respiratory rate (breaths/minute), heart rate (beats/minute), oxygen saturation on room air (%), wheezing (yes or no), retractions (yes or no), time of year (respiratory season [October–April] or nonrespiratory season), any underlying medical condition (yes or no), underlying respiratory condition (yes or no), underlying asthma/reactive airway disease (yes or no), underlying cardiovascular condition (yes or no), underlying neurologic condition (yes or no), underlying oncologic/immunosuppressive condition (yes or no), prematurity (yes or no), public insurance (yes or no), private insurance (yes or no), none/self-pay insurance (yes or no), and previous antibiotic use in past 14 days (yes or no). Continuous variables were standardized to have a mean of 0 and SD equal to 1.

Presentation patterns were derived for children in each health care setting (ie, ED and hospital). The principal components derived from the PCA analysis were rotated to maximize the variance explained by each pattern. Scree plots were generated to summarize the total number of principal components (PCs) with the greatest amount of variance explained. The strongest PCs whose eigenvalues (ie, the coefficients that provide us with summary statistics that depict the amount of variance carried from each PC) were greater than 1 were chosen to represent the independent presentation patterns. Final PCs were dichotomized at the median using cut-points among children who did not receive a provider-ordered test (n = 1322). Our pediatricians (N.H. and S.K.) assigned respiratory virus/coviral names to each PC based on their clinical suspicion from the presentation patterns represented. Coviral refers to a presentation pattern of a child with more than 1 respiratory virus.

PCs were validated using test results from our research specimens. Research specimens (eg, nasal, oropharyngeal, tracheal aspirates) were collected from all enrolled children and placed in viral transport medium by study personnel. All research specimens were transported to our research laboratory and stored at 4 °C until processed, within 48 hours. Testing was performed using quantitative reverse transcription-polymerase chain reaction assays for the following viruses: adenovirus, human metapneumovirus, influenza (A, B, and C), parainfluenza (types 1–4), RSV, rhinovirus, and endemic coronaviruses (OC43, 229E, NL63, and HKU1). Research specimen collection and molecular testing were performed in accordance with the New Vaccine Surveillance Network study protocol.^9,10 

Sociodemographic and clinical characteristics of children by setting were summarized using frequency for categorical variables and mean (SD) for continuous variables. We performed multiple imputation for 266 children with ≥1 missing covariates (missingness: 1 covariate [n = 258] children; 2 covariates [n = 6]; 3 covariates [n = 2]), using predictive mean matching with 10 imputation samples.¹⁴  Our final analytic sample size consisted of 4107 children.

PC-specific models were built using unconditional logistic regression with robust sandwich estimators (to account for repeat visits) to estimate odds ratios (ORs) and 95% confidence intervals (CIs) between different clinical presentation patterns and provider-ordered viral testing. All analyses were performed in R (version 3.6.1).

From November 2017 to February 2020, 4271 (57%) of 7482 eligible children were enrolled, among whom 4212 (99%) had a research specimen tested. A total of 151 enrollments were excluded based on a previous enrollment occurring within <90 days. Our final cohort included 4107 children, of which 2616 (64%) were from the ED and 1491 (36%) were hospitalized (Fig 1). A total of 455 children from our final cohort had multiple ARI visits ≥90 days apart, with some children incurring multiple visits in both health care settings (n = 12).

Enrolled children were a median age of 3 years, predominately Black, non-Hispanic, and 31% had at least 1 underlying medical condition (Table 1). Cough, nasal congestion/runny nose, and fever were the most frequently reported symptoms. A provider-ordered viral test occurred in 49% (1294/2616) of children with an ARI-related ED visit, 97% of whom received 1 test (1255/1294; 87% rapid influenza; 1% rapid RSV; 13% NAAT), 3% received 2 tests (38/1294; 100% rapid influenza; 42% rapid RSV; 58% NAAT), and <1% had 3 tests (1/1294). In total, 64% of children had at least 1 virus detected from their research specimen, with influenza being most common (Table 1).

Among 1491 hospitalized children, the median age was 1.6 years, 46% were white, non-Hispanic; 47% of hospitalized children had at least 1 underlying medical condition and the mean length of stay was 3 days. Twenty percent were admitted to the ICU. Cough, nasal congestion/runny nose, and shortness of breath were the 3 most common symptoms reported; on physical examination, 51% and 36% had retractions and wheezing, respectively (Table 1). A provider-ordered test occurred in 64% (955/1491) of the children with an ARI-related hospitalization, with 82% receiving 1 test (784/955; 22% rapid influenza, 3% rapid RSV, 75% NAAT), 15% had 2 tests (143/955; 95% rapid influenza; 46% rapid RSV; 59% NAAT), and 3% had 3 tests (28/955). Overall, 61% of children had at least 1 virus detected from their research specimen, with RSV (31%) and rhinovirus (20%) most commonly detected.

A total of 4 PCs accounting for 63% of cumulative variance were identified in the ED (Table 2, Supplemental Fig 4). PC1_ED described 26% of variance and had a pattern mimicking RSV bronchiolitis: younger children with a higher temperature, more likely to have respiratory symptoms, short illness duration, lower oxygen saturation, no underlying medical conditions, and not regularly attending daycare. PC2_ED had a pattern resembling rhinovirus-like presentation: children older than those in PC1_ED, less likely to have fever present, more likely to present with respiratory symptoms, have wheezing and retractions present on physical examination, and no underlying medical conditions. PCs 3_ED-4_ED accounted for 12% and 11% of variance, respectively, and both patterns comprised older children. PC3_ED represented coviral-like (eg, influenza, rhinovirus) infections with a pattern of a short illness duration, high temperature, more likely to present with respiratory symptoms, low respiratory rate, and low oxygen saturation. PC4_ED closely followed an influenza-like presentation with a pattern of longer illness duration, respiratory symptoms present, high respiratory rate and low oxygen saturation, and underlying medical conditions dominated by respiratory disorders. Univariable summary statistics for PCs can be found in Supplemental Table 4.

Children who were more likely to represent the coviral-like presentation pattern (≥median, PC3_ED) had 44% (OR, 1.44; 95% CI, 1.24–1.68) increased odds of receiving a provider-ordered viral test during their visit than those children showing clinical symptoms less representative of coviral-like infection (<median, PC3_ED; Table 3). In contrast, children with symptoms representing the rhinovirus-like pattern (≥median, PC2_ED) had 71% (OR, 0.29; 95% CI, 0.24–0.34) decreased odds of receiving a provider-ordered viral test compared with those with symptoms less representative of the rhinovirus-like pattern (<median, PC2_ED). No associations were observed for the patterns of PC1_ED (RSV-like) and PC4_ED (influenza-like) with provider-ordered testing.

A total of 3 PCs accounting for 49% cumulative proportion of variance were identified among hospitalized children (Table 2, Supplemental Fig 4, and Supplemental Table 4). PC1_IP explained 22% of observed ordering variation and resembled influenza-like illness. These children were older (school-aged) and characterized by short illness duration, high temperature, increased frequency of nonrespiratory symptoms (eg, myalgia, vomiting, diarrhea), low respiratory and heart rates, and underlying conditions. PC2_IP described 15% variance and consisted of children with rhinovirus-like illness. These patients were younger than those in the PC1_IP group (influenza-like) and presented with a short illness duration, no fever, greater likelihood of respiratory symptoms, low heart rate, low oxygen saturation, wheezing on physical examination, and underlying conditions dominated by respiratory disorders (ie, asthma/reactive airway disease). PC3_IP mimicked RSV-like illness and captured young children presenting with a high respiratory rate, low oxygen saturation, and wheezing and retractions on physical examination.

Among hospitalized children, children with symptoms representing the rhinovirus-like pattern (≥median, PC2_IP) had 39% (OR, 0.61; 95% CI, 0.49–0.76) decreased odds of receiving a provider-ordered viral test during their stay than those with dissimilar symptoms (≤median, PC2_IP; Table 3). An association was not observed for the patterns of PCs 1 (influenza-like) and 2 (rhinovirus-like) with provider-ordered testing.

Among children who were tested, the frequency of viral detection from a research specimen was highest in PCs_ED 1 and 4 (68% and 67%, respectively; Fig 2). RSV (20%) and ≥2 respiratory viruses (20% [RSV/coronavirus codetection = 9%;17/182]) were the most often detected in PC1_ED (RSV-like), whereas influenza was most common in PC3_ED (30%; coviral-like) and PC4_ED (30%; influenza-like). In PC2_ED (rhinovirus-like), rhinovirus (25%) and RSV (19%) were the 2 most frequent viruses detected.

Among children who had a provider-ordered viral test, positivity was highest in PC3_IP (RSV-like), with RSV accounting for 36% of all virus-positive detections (Fig 2). In PC2_IP (rhinovirus-like), rhinovirus (30%) and RSV (29%) were the 2 most commonly detected viruses, whereas in PC1_IP (influenza-like), rhinovirus and influenza were identified in 22% and 21% of virus-positive specimens, respectively.

In our prospective study of 4107 ARI-related ED visits and hospitalizations in Nashville, TN, we demonstrate provider-ordered viral testing practices vary across health care settings. In the ED, nearly one-half of children with an ARI-related illness were administered a provider-ordered viral test, with rapid influenza accounting for 87% of all tests performed. In comparison, two-thirds of hospitalized children underwent viral testing with more than 60% receiving a molecular test. Coviral-like presentation patterns were more likely to receive provider-ordered testing in the ED, whereas rhinovirus-like patterns were less likely to receive testing in either setting. Thus, presentation patterns among children with ARI vary by health care setting and influence decisions to initiate viral testing.

In the ED, we found that the coviral-like presentation pattern was associated with provider-ordered viral test. Rapid influenza tests accounted for more than three-quarters of all tests administered, whereas 66% of virus-positive research specimens in the coviral-like pattern were non-influenza viruses. The abundance of rapid influenza tests may be attributed to the decision of initiating prompt antiviral therapy, which was only available for influenza before the SARS-CoV-2 pandemic. Furthermore, these findings suggest that non–influenza-like presentation patterns, such as RSV (common cause bronchiolitis), were not targeted with appropriate diagnostic testing. The limited use of RSV testing adheres to the American Academy of Pediatrics clinical practice guidelines for bronchiolitis, which recommends clinicians diagnose children based on history and physical examination and advises against routine viral testing.⁶  In the pediatric guidelines for community-acquired pneumonia (common viral ARI condition), rapid influenza testing is strongly recommended as a first step to reduce additional diagnostic testing, initiate antiviral therapy, and minimize unnecessary antibiotic therapy.¹⁵  From a diagnostic standpoint, the clinical patterns and associated viral testing practices we identified in the ED appear to be well aligned with the common ARI condition(s) guidelines. However, the confluence of clinical factors, laboratory findings, and institutional practices that shape management of children discharged from the ED with a non-influenza ARI and a negative influenza rapid test deserves systematic analysis.

The proportion of provider-ordered viral tests for hospitalized children were nearly 20% more compared with the ED, with the majority performed using NAAT. This difference might be attributed to time allowance for testing results in the inpatient setting, potential antiviral use, the need to rule out serious bacterial infections, insurance reimbursement, or decision-making about further laboratory evaluation, such as blood tests, cerebrospinal fluid assessment, or radiography.^4,16 

Despite the availability of NAAT to our ED clinicians, this resource was sparingly used, perhaps a reflection of the influence of a positive result on the clinical management plan. A meta-analysis assessing the use of rapid viral tests in the ED revealed null effects of viral testing on provider decisions about clinical management.³  In a retrospective study using the National Hospital Ambulatory Medical Care Survey, rapid influenza testing in US EDs decreased the use of antibiotics and ancillary testing.¹⁷  Most studies assessing rapid viral testing and its impact on clinical management in the ED are methodologically aimed at influenza,³  and findings have not been uniformly replicated in hospitalized patients.⁵  Discordant results across settings could be due to variability in patient acuity, provider practices, or the prevalence of viral testing performed. Studies evaluating pathogen-appropriate testing and influence of results on patient management and outcomes are needed to define the true clinical impact of testing on pediatric ARI.

In our study, rhinovirus was detected in nearly a quarter of children with a research specimen (23%, including codetections). Despite being common in both the ED and inpatient settings, children with rhinovirus-like patterns were less likely to receive a provider-ordered test. Rhinovirus is often associated with asthma exacerbations, bronchiolitis, and community-acquired pneumonia in children, particularly among those with underlying respiratory conditions (ie, asthma/reactive airway disease).^18–20  Limited use of viral testing in both the ED and inpatient settings for rhinovirus-like presentation pattern might be ascribable to clinical management guidelines, as previously mentioned, or specific clinical presentations observed in children with rhinovirus ARIs.¹⁹  Of note, in a prospective multicenter study assessing differences among children younger than age 2 years hospitalized with bronchiolitis caused by RSV and rhinovirus, patients with rhinovirus ARI presented similarly to older children with asthma.²¹  Differences in clinical presentation may underlie variations in patient management, including viral testing, highlighting opportunities for improvement initiatives and standardization of care across settings.

A major strength of our study is systematic, active, prospective enrollment and molecular viral testing conducted over consecutive years among children with ARI in both the ED and inpatient settings. However, certain limitations apply. First, our study only included children enrolled before March 2020 and does not reflect practices during the COVID-19 pandemic. Second, we used principal components, a data-driven approach, to understand the presentation patterns influencing provider-ordered viral testing. Third, our findings represent a single center and may not be generalizable to other states or regions.²  Fourth, we were unable to assess clinician years of experience in either setting, which may affect viral testing practices. Finally, our study focused on testing practices for viral ARIs and did not assess clinical testing for bacterial ARIs.

Variability in provider-ordered viral testing was observed across presentation patterns and testing options by health care setting. Rapid influenza testing was the most common test performed in the ED, whereas utilization of NAATs was more common in the inpatient setting. The diversity of testing practices and clinical presentations we identified across the ED and inpatient setting may assist in the development of setting specific diagnostic stewardship programs and highlights the need for broadening viral testing in the ED. Future research should target identification of individual drivers of pediatric viral testing practices and policies and impact of those decisions on ARI management and outcomes.

HALASA/CHAPPELL Research Investigators: Rebekkah Varjabedian, Jessica Villarreal, Marcia Blair, Lauren J. Ezell, Samarian J. Hargrave, Jennifer L. Luther, Bryan P. M. Peterson, Laura L. Short, Margaret E. Whitsett.

We thank our research personnel and the families who participated in this study.