key: cord-0843744-398vvkrn
authors: Weng, C.-H.; Butt, W. W. W.; Brooks, M. B.; Clarke, C.; Jenkins, H.; Holland, S. D.; Chiang, S. S.
title: Diagnostic Value of Symptoms for Pediatric SARS-CoV-2 Infection in a Primary Care Setting
date: 2021-03-31
journal: nan
DOI: 10.1101/2021.03.29.21254600
sha: 7c8e2f2ef1594e4209a6f96926940759b2d3a269
doc_id: 843744
cord_uid: 398vvkrn

Purpose To evaluate the diagnostic value of symptoms used in the screening approaches by daycares and schools for identifying children and adolescents with possible SARS-CoV-2 infection, we designed a large observational study utilizing the data from primary care settings. Methods This cohort study included children and adolescents evaluated in a network of clinics in Rhode Island. Participants were age-stratified: 0-4, 5-11, and 12-17 years. We estimated the sensitivity, specificity, and area under the receiver operating curve (AUC) of individual symptoms and three symptom combinations: a probable case definition published by the Rhode Island Department of Health (RIDOH), and two novel combinations generated by different statistical approaches to maximize sensitivity and AUC. We evaluated the test characteristics of symptom combinations both with and without consideration of COVID-19 exposure. Results Two-hundred seventeen (39.1%) of 555 participants were SARS-CoV-2-infected. Fever was more common among 0-4 years-olds (p=0.002); older children more frequently reported fatigue (p=0.02) and anosmia or ageusia (p=0.047). In children >5 years old, anosmia or ageusia had 94-98% specificity. In all age groups, exposure history most accurately predicted infection. In combination with COVID-19 exposure history, various symptom combinations had sensitivity >95% but specificity <30%. No individual symptom or symptom combination had AUC [≥]0.70. Conclusions Anosmia or ageusia in children [≥]5 years old and dyspnea in children 5-11 years old should raise providers' index of suspicion for COVID-19. However, our overall findings underscore the limited diagnostic value of symptoms and the critical need for widely available, efficient testing.

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 31, 2021. ; https://doi.org/10.1101 /2021 doi: medRxiv preprint model. We checked for interactions on the multiplicative scale between known COVID-19 149 exposure and all other covariates. 150 151 For each age group, we calculated the sensitivity, specificity, and AUC of COVID-19 exposure 152 history and each symptom for identifying SARS-CoV-2 infection. Myalgia, headache, sore throat, 153 abdominal pain, nausea, and anosmia or ageusia were not assessed in children 0-4 years old due 154 to their lower ability to report these symptoms. We then evaluated the diagnostic value of three 155 symptom combinations: (1) the probable case definition published by RIDOH(9); (2) a 156 combination generated by a backward elimination approach; and (3) a combination generated by 157 classification and regression tree (CART) analysis. We evaluated the test characteristics of the 158 three symptom combinations both with and without consideration of COVID-19 exposure. 159 160 According to the RIDOH criteria, a probable COVID-19 case has one of the following: new 161 cough, shortness of breath, or anosmia or ageusia. A case also qualifies as probable COVID-19 162 by having at least two of the following: fever, chills, myalgia, headache, sore throat, nausea or 163 vomiting, diarrhea, fatigue, or new congestion or rhinorrhea.(9) As we did not collect 164 information on chills, we excluded this symptom in our analysis of the RIDOH criteria. 165 166 For each age group, we used a backward elimination approach to generate a symptom 167 combination that maximized specificity without sacrificing sensitivity. First, we calculated the 168 sensitivity, specificity, and AUC if any of the symptoms were present (the baseline combination). 169

Then, we manually removed symptoms one at a time, in order of ascending AUC. Symptoms 170 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted March 31, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 with the same AUC were removed in order of ascending sensitivity. We selected the 171 combination with the highest specificity but the same sensitivity as the baseline combination. 172

We used CART analysis to identify the symptoms that best predicted SARS-CoV-2 infection in 174 each age group. In CART analysis, measures of predictive importance were assigned to each 175 symptom, entailing both marginal and interaction effects involving this variable. The data set 176 was then split into increasingly homogenous sub-groups, using improvement in the Gini gain 177 score, to identify the explanatory variable that gave the best discrimination between the two 178 outcome classes . Maximal trees were created and then pruned 179 based on relative misclassification costs, complexity, and parsimony. Ten-fold cross-validation 180 was performed, in which the whole data set was randomly split into learning and test data sets. 181 CART analysis was then applied to determine model performance and predictive accuracy in 182 these test sets, removing the need for a validation data set. We calculated discriminatory 183

properties of having at least one of the most important symptoms identified as nodes on the final 184 derived trees for each age group. 185

To determine the impact of recall bias from applying the standardized template after PCR testing, 187

we performed a sensitivity analysis restricted to participants who were evaluated for exposure 188 and symptoms before testing. We decided a priori to assess the diagnostic value of symptom 189 combinations only if there were meaningful differences in AUCs of exposure history and 190 individual symptoms. 191 192 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 31, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 Twenty-two (4.0%) participants with unknown race/ethnicity were grouped into the "NH other" 193 group and included in all analyses. BMI-for-age percentile-which is measured in children at 194 least two years old-was missing for 128 (23.1%) participants, 96 of whom were younger than 195 two years. These participants were excluded from comparisons of BMI between infected and uninfected children only. Four (0.7%) participants had no data for one symptom; 197 they were excluded from regression models that examined the association between the number of 198 presenting symptoms and SARS-CoV-2 infection, as well as from calculations of sensitivity, 199 specificity, and AUC for the missing symptom only. There were no other missing data. Before June 22, 2020, SARS-CoV-2 PCR was performed in 803 individuals <18 years of age 214 who were registered as PCHC patients. We included 555 (69.1%) who were evaluated using the 215 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 31, 2021. ; standardized template. PCHC clinicians assessed 506 (91.2%) participants in primary care and 216 five (0.9%) in urgent care prior to PCR testing. Forty-four (7.9%) participants were assessed by 217 PCHC clinicians after pre-procedure screening (n=10), hospital admission (n=21), and 218 emergency room visit (n=13). The 248 excluded patients were tested at a PCHC specialty clinic 219 or a non-PCHC facility without subsequent evaluation using the standardized template 220 (Supplementary Material). 221

Of the 555 participants, 283 (51.0%) had known COVID-19 exposure and at least one symptom; 223 183 (33.0%) had at least one symptom but no known exposure; 56 (10.1%) had known exposure 224 but no symptoms; and 33 (5.9%) participants had neither symptoms nor known exposure. Two-225 hundred eighty-nine (52.1%) participants were tested after reopening on May 9. Median age was 226 9 (IQR: 3.5-15) years; 293 (52.8%) were female, 459 (82.7%) were Hispanic, and 37 (6.7%) 227

were uninsured (Table 1) CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 31, 2021. ; https://doi.org/10.1101/2021.03.29.21254600 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) (3 271 vs. 2 symptoms). Test positivity did not differ between participants evaluated before and after 272 reopening. The multivariable regression analyses showed consistent results (Table 1) (Tables 3-5) . 281

When exposure history was considered, all symptom combinations had 97-100% sensitivity. 282

When exposure history was not considered, the RIDOH criteria and the combination generated 283 by backward elimination had the highest sensitivity: 95% in 0-4 year-olds, 87% in 5-11 year-olds, 284 and 92% in 12-17 year-olds. All combinations had <30% specificity. 285 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 31, 2021. ; https://doi.org/10.1101/2021.03.29.21254600 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 31, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 31, 2021. ; is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 31, 2021. ;  or anosmia or ageusia cough, or headache 97 (77.6) 108 ( is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 31, 2021. ;  In the sensitivity analysis of participants evaluated with the standardized template before PCR 340 testing, the AUCs of exposure history and individual symptoms were similar to those calculated 341 for the entire study population, with differences of In this study, we assessed the diagnostic properties of symptoms for SARS-CoV-2 infection in a 346 large pediatric cohort, >90% of whom presented to primary care and were evaluated with a 347 standardized symptom template before PCR testing. We identified symptom combinations with 348 high sensitivity, particularly in conjunction with COVID-19 exposure; however, all symptom 349 combinations had poor specificities. We failed to identify any individual symptom or symptom 350 combination with AUC >0.70, underscoring the importance of widely available SARS-CoV-2 351 testing with rapid turnaround. 352

The AUCs observed in our study likely are higher than in the general population for several 354 reasons. First, few study participants were asymptomatic, thus maximizing sensitivity of 355 symptoms. Second, our study took place in the spring and summer; specificities of symptoms are 356 expected to decrease further in the winter as more respiratory viruses circulate. Third, the 357 reliability of COVID-19 exposure history probably was higher in our cohort since many 358 participants were tested during a stay-at-home order. Therefore, they likely had few contacts and 359

were better able to stay informed of the infection status of their contacts. 360 361 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 31, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 Reports of pediatric COVID-19 symptoms mostly include hospitalized participants.(3, 14-17) 362

One exception is a study conducted in Alberta, Canada, which used provincial databases to 363 assess the association of symptoms with SARS-CoV-2 PCR positivity.(8) This study found a 364 high positive predictive value for anosmia or ageusia; similarly, we observed a high specificity 365 of these symptoms. Our study differs in a few ways. First, in Alberta, the symptom questionnaire 366 was applied after test results were known, whereas symptoms were assessed before testing in 367 >90% of our cohort, reducing recall bias. Second, we age-stratified participants and detected 368 differences in COVID-19 presentation between age groups. These differences were similar to 369 those reported by the BRAVE study, which evaluated children with a close SARS-CoV-2-370 infected contact (18): elementary school-aged children were most likely to have asymptomatic 371 COVID-19 (though the difference did not reach statistical significance in our cohort), the 372 youngest children were most likely to be febrile, and adolescents were more likely to report 373 anosmia or ageusia compared to elementary school-aged children. 374 375 Our findings have clinical and public health implications. The low AUCs we observed strongly 376 argue against the use of symptoms to diagnose pediatric COVID-19. However, anosmia or 377 ageusia in children ≥ 5 years old and dyspnea in children 5-11 years old are highly specific, and 378 their presence should alert providers to quickly isolate and test the patient. Exposure history most 379 accurately predicted SARS-CoV-2 infection and should remain a cornerstone of quarantine 380 recommendations. Because of differences between our cohort and daycare and school attendees, 381 the diagnostic characteristics we observed may not be generalizable to that group. However, our 382 findings suggest that different age groups need distinct symptom screening criteria. Additionally, 383 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 31, 2021. ;  because of the low specificities of most symptoms, easily accessible tests with rapid turnaround 384 times are critical to minimize unnecessary absences. 385 386 With respect to the secondary aim of our study, we identified Hispanic ethnicity as an 387 independent risk factor for COVID-19 compared to the reference group of NH Black. Both the 388 BRAVE study and another study conducted in Washington, DC similarly found significantly 389 higher SARS-CoV-2 positivity in Hispanic children. (18, 19) Further investigation is needed to 390 clarify the contribution of various factors-including multigenerational or multi-family housing, 391 the inability to work from home, and language barriers-to these higher positivity rates.(20-24) 392

The DC study reported that NH Blacks also had higher positivity rates than NH Whites, but we 393 did not detect a difference between these groups, potentially due to insufficient statistical power. 394 395 This study had limitations. Data were collected early in the pandemic; however, symptoms are 396 not expected to change over the course of the pandemic, and the clinical and public health 397 implications of this study remain relevant and practical. As previously discussed, the AUCs of 398 exposure and symptoms that we observed may represent a "best case scenario," but this 399 possibility only strengthens the overarching message that symptoms are poorly predictive of 400 402 Despite these limitations, our assessment of the diagnostic value of symptoms fills an important 403 gap in the pediatric COVID-19 literature. The poor AUCs we observed mean that symptoms 404

should not be used alone to identify pediatric SARS-CoV-2 infection, and underscore the 405 importance of widely available and efficient testing. 406 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted March 31, 2021. ;

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted March 31, 2021. ; https://doi.org/10.1101/2021.03.29.21254600 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted March 31, 2021. ;

American Academy of Pediatrics and the Children's Hospital Association. Children and 456 COVID-19: State Data Report

SARS-CoV-2) Infection in Children and Adolescents: A 461 Systematic Review

Epidemiology of COVID-19 Among 463 Children in China

Infectious Disease Perspective of SARS-CoV-2 and COVID-19 in Children

Signs and 468 symptoms to determine if a patient presenting in primary care or hospital outpatient settings has 469 COVID-19 disease

471 Presenting symptoms of COVID-19 in children: a meta-analysis of published studies

Symptoms associated with a positive 474 result for a swab for SARS-CoV-2 infection among children in Alberta

Guidance 476 for COVID-19 outbreak response in Pre K-12

COVID-19 Information

An Easy Way to Report ROC Analysis. R package version 3.5 486 ed2020. 487 14. Team CC-R. Coronavirus Disease 2019 in Children -United States

Clinical 490 characteristics of children and young people admitted to hospital with covid-19 in United 491 Kingdom: prospective multicentre observational cohort study

COVID-19 in 7780 pediatric patients: A systematic review

Systematic 495 review of reviews of symptoms and signs of COVID-19 in children and adolescents

SARS-CoV-2 Infections Among Children in the Biospecimens from Respiratory Virus

Kids (BRAVE Kids) Study. Clin Infect Dis. 2020. 500 19

Racial/Ethnic and Socioeconomic Disparities of SARS-CoV-2 Infection Among Children. 502 Pediatrics

Uninsured Hispanic/Latino Population: a Retrospective Cohort Study

COVID-19 and African Americans

Associated With Racial And Ethnic Disparities In COVID-19 Rates In Massachusetts. Health 510 Aff (Millwood). 2020:101377hlthaff202001040

Persistently high SARS-CoV-2 positivity rate 512 and incidence for Hispanic/Latinos during state reopening in an urban setting: a retrospective 513 cohort study

The authors thank Jennifer Friedman and Philip Chan. 408 409Funding: None