Assessment of a scoring system for long-term COVID-19 symptoms in children

We conducted a secondary analysis of data collected from children enrolled in a multicenter, prospective, longitudinal cohort study conducted between 4 August 2020 and 22 February 2022 (7, 10). Participating institutions (Supplementary Table S1) included 14 Pediatric Emergency Research Canada (PERC) (11) tertiary-care pediatric emergency departments (EDs). Participating sites obtained research ethics board approval, and caregivers provided informed consent; assent was obtained per institutional policy. We followed the Statement for Reporting Studies of Diagnostic Accuracy guidelines (12).

Children less than 18 years old who underwent testing for SARS-CoV-2 due to symptoms or epidemiologic risk factors for infection (e.g., close contact with positive individuals) were eligible. Specimens were collected at the treating physician's discretion and analyzed per local laboratory standards. To identify potentially eligible children, team members received a list each day of children who had SARS-CoV-2 testing performed. Research assistants attempted to contact by telephone the caregivers of all children who tested positive, followed by contacting those who tested negative. To minimize selection bias, recruitment was standardized across sites by attempting to contact potentially eligible participants consecutively based on the chronological time of specimen collection.

For this sub-study, to align with the age categories included in the PASC derivation study (9), participant eligibility was restricted to those aged 6 to <18 years. Eligible participants completed follow-up surveys, which were added to the protocol on 1 November 2021, at 6 and/or 12 months after their index ED visit; participants enrolled prior to 1 November 2020 were ineligible. These time points were employed to standardize follow-up timing; in the derivation study, follow-up was performed a median of >500 days after infection (9).

Research into the pathophysiology of PASC and the conduct of therapeutic trials has been challenging, particularly in children (13). As an extensive evaluation failed to identify any laboratory tests that could serve as useful biomarkers of PASC (14), the Researching COVID-19 to Enhance Recovery (RECOVER) initiative prospectively enrolled and collected symptom data from children infected and uninfected with SARS-CoV-2 to improve our understanding of pediatric PASC (9). Reported symptoms were analyzed, and those most strongly associated with prior infection were included in an index that has been proposed as a research tool to identify children with PASC.

In this study, we sought to assess the psychometric properties of the PASC index through an evaluation of construct validity, which was evaluated via exploratory factor analysis (EFA) and concurrent validity by examining the association between the total PASC index and the PedsQL™-4.0 inventory and caregiver-reported overall health status. Reliability was assessed via Cronbach's α and a generalizability (G) study. Finally, to assess the relationship between the PASC score and SARS-CoV-2 infection, we assessed if scores above the RECOVER study's age-specific cut points (6 to <12 years, score ≥5.5; 12 to <18 years, score ≥5) (9) are associated with SARS-CoV-2 test status.

Clinical information from the index ED visit was collected from caregivers and supplemented by medical record review performed 14 days later to identify the index ED SARS-CoV-2 test result and results of additional tests performed. Follow-up data were collected from caregivers 6- and 12-month post-index ED visit using a modified version of the International Severe Acute Respiratory and Emerging Infection Consortium (ISARIC) long-COVID pediatric survey tool (Supplementary Table S2) (15). Self-reported race and ethnicity were collected as disparities exist in PASC prevalence based on these social constructs (16). To standardize data collection, caregivers completed follow-up questionnaires as most surveys were completed during the day when older children and teenagers were unavailable—they were at school or the caregiver was at work.

SARS-CoV-2 status was classified as positive if a nucleic acid test performed on a swab obtained from the nares, nasopharynx, or oral cavity at the index ED visit or during the subsequent 14 days was positive. Non-nucleic acid tests performed at home were infrequently available during the enrollment period, and as such we relied on results of nucleic acid testing. However, all participating sites are in Canada, with provincial health authorities, enabling easy access to results performed in any laboratory or hospital across their province. Participants with negative tests constituted the comparison group. Acute SARS-CoV-2 hospitalization status incorporated events until 14 days after the index ED visit (17). Testing and reporting of variants of concern (VoC) varied by institution and over time. When a VoC or a variant linked to a VoC was identified, that report was used for classification purposes. If VoC testing was not performed or results were inconclusive, the SARS-CoV-2 variant was classified based on the predominant variant circulating at the time of testing (10).

The RECOVER study's PASC index includes 10 symptoms (e.g., trouble with memory or focusing; back, neck, or stomach pain; headache; and fear about specific things) for those 6 to <12 years and 8 symptoms (e.g., change or loss of smell or taste; body, muscle, or joint pain; daytime tiredness; sleepiness or low energy; and tired after walking) for those 12 to <18 years. The maximum achievable scores are 35 and 25 for these age groups, respectively. Follow-up surveys collected data related to all symptoms included in the PASC index; however, the precise language or collection approach varied slightly from the derivation study (). Supplementary Table S3

Quality of life (QoL) was quantified using age-specific versions of the PedsQL™-4.0 (18), which includes 23 distinct items divided into four subscales that address physical, emotional, social, and academic functioning (19). Caregivers selected survey responses using a five-point Likert scale, ranging from "never" to "almost always" and described the child's well-being in the preceding 7 days. PedsQL™ domain-specific scores were derived by summation of item scores divided by the number of items answered (20). The average of the subdomain scores provided the cumulative PedsQL™ score (19, 20). Participant overall health status was reported by caregivers on a 0- to 100-point visual analog scale.

We estimated the sample size in each group (i.e., infected and uninfected) for participants aged 6 to <12 years and 12 to <18 years, to identify a difference between groups in the categorization of participants according to the RECOVER PASC index with 80% power and a two-sided type 1 error of 0.05. Calculations employed age-specific estimates of PASC scores of 20% and 4% for children (6 to <12 years) and 14% and 3% for adolescents (12 to <18 years), among children infected and uninfected with SARS-CoV-2, respectively (9). For the latter group of children, these estimates do not imply they have PASC (as they were uninfected) but rather symptoms yielding scores that meet the threshold proposed for the diagnosis of PASC in SARS-CoV-2 infected individuals. To account for potential site-level clustering effects, as participants were recruited from 14 sites, the sample size was adjusted using a design effect approach. Given that the exact magnitude of the intra-class correlation coefficient (ICC) is unknown, we conducted sample size estimates assuming ICC values of 0.01, 0.05, and 0.10 (21). Using these estimates, the required sample size per group is 67, 76, and 87 participants aged 6 to <12 years, respectively, and 107, 131, and 162 participants aged 12 to <18 years, respectively.

Data were summarized using descriptive statistics, and baseline categorical variables were compared using the χ² test, Fisher's exact test, or Mann–Whitney U test, as appropriate. Confidence intervals for the difference between proportions were calculated using the Wald method; the Agresti–Caffo method was used when the event rate was <20% (22). Participants with missing or incomplete outcome data were excluded. All analyses were two-tailed, with significance set at P < 0.05. Analyses were performed using SPSS 29.0 (IBM Corp., Armonk, NY, USA) and R version 4.3.3.

Construct validity was assessed using EFA. A two-factor solution was explored through principal axis factor analysis with direct oblimin rotation because we anticipated some correlation between the items included in the index (23). Factor loadings in the resulting pattern matrix reflect the correlation between an item in the index and a specific factor; higher loadings indicate a stronger association with that factor. We used a cutoff of 0.4 to determine whether an item was significantly related to the factor (24). Due to the small sample size (n < 50), EFA was not performed for the PASC in the 12- to <18-year-old group at 6 months.

Internal reliability was evaluated using Cronbach's α with values ≥0.7 indicative of good internal reliability (25). For generalizability, we conducted a G study on total PASC scores to assess the variance components associated with study sites, assessment time points, and individual participants. This approach provides a framework to evaluate the reliability of a scoring system (26 –28), by quantifying the proportion of total variance attributable to different sources. We fitted a random-effects ANOVA model (variance components model) to estimate the variance attributable to study sites, assessment time points, and individual participants. Based on these variance components, we calculated generalizability coefficients to evaluate the relative and absolute reliability of the PASC index. Additionally, we also explored how the generalizability coefficient would improve with additional repeated assessments per participant.

To assess concurrent validity, we calculated Pearson correlation coefficients to examine the relationship between the total PASC index and the PedsQL™-4.0 inventory and caregiver-reported overall health reported on a 0- to 100-point scale. To assess the relationship between exposure to COVID infection, PASC index scores were calculated and classified. To estimate the odds of having a score exceeding the threshold in relation to SARS-CoV-2 test status, we applied logistic regression using generalized estimating equation (GEE) models, adjusting for clustering by site and the presence of chronic conditions at baseline. Separate GEE models were used for each age and survey time point subgroup. As several of the measures included in the PASC index scores include measures from the PedsQL™-4.0 inventory which includes a response option of "sometimes" (), a sensitivity analysis was conducted with this option categorized as the symptom being "absent." Supplementary Table S3

A total of 785 eligible participants completed either or both follow-up time points (Figure 1). There were no clinically significant differences between those who were and were not lost to follow-up based on SARS-CoV-2 classification (Supplementary Table S4). The median age of the participants was 9 years (IQR: 7, 13), 51.5% (N = 401) were male, 46.4% (N = 364) self-identified as White, 86.5% (N = 679) were discharged from the ED at the index visit, and 1.7% (N = 13) were asymptomatic at the time of testing (Table 1).

EFA of the PASC index in both age groups and time points identified two factors. Among the children 6 to <12 years old, at 6 months, factor 1 was primarily associated with physical symptoms, while factor 2 was associated with psychological/anxiety or cognitive symptoms (e.g., trouble with memory or focusing). At the 12-month follow-up, factor 1 remained associated with physical symptoms, while factor 2 was linked to fearfulness and memory problems (Table 2). The two factors explained 38.5% and 31.5% of the total variance at 6 and 12 months, respectively. For children aged 12 to <18 years, at 12 months, factor 1 included physical symptoms, whereas factor 2 reflected fatigue-related symptoms. These factors explained 39.7% of the total variance.

An inverse correlation, i.e., higher PASC scores and lower PedsQL^™ scores, occurred at all time points within all age groups (Supplementary Table S5 and Figure S1). For children aged 6 to <12 years, the correlations did not differ between SARS-CoV-2 positive and negative groups, ranging from −0.71 (positives at 12 months) to −0.74 (negatives at 12 months). For participants aged 12 to <18 years, the inverse correlation was greater among SARS-CoV-2 negative participants at 6 months (difference: 0.36; 95% CI: 0.28, 0.85), but the opposite (i.e., stronger among SARS-CoV-2 positives) was detected at 12 months (difference: −0.19; 95% CI: −0.58, 0.03).

Correlation analysis between PASC index scores and overall health status revealed inverse correlations being strongest among SARS-CoV-2 positive participants aged 12 to <18 years at the 6-month follow-up (−0.59; 95% CI: −0.79, −0.28) and weakest among SARS-CoV-2 negative participants aged 6 to <12 years at the 6-month follow-up (−0.34; 95% CI: −0.47, −0.18) (). Overall, the strength of correlations did not differ between SARS-CoV-2 positive and negative groups at either time points or age groups. Supplementary Table S6 and Figure S2

Internal reliability was greatest among SARS-CoV-2 positive children aged 6 to <12 years at 6 months (α = 0.67; 95% CI: 0.64, 0.69) suggesting moderate internal consistency. It was lowest among participants aged 12 to <18 years at 6 months (α = 0.49; 95% CI: 0.45, 0.53) reflecting low internal consistency (Supplementary Table S7). In the generalizability analysis, the largest variance was attributable to changes in PASC scores across the two time points, accounting for 70.8% and 62.8% of the total variance, among the younger and older age cohorts, respectively. The variance attributed to individual subjects was 15.1% and 36.8%, respectively. For children aged 6 to <12 years, adding five repeated assessments would increase the G coefficient to 0.6, while adding one additional assessment yields a similar improvement for the 12- to <18-year-old group (Supplementary Table S8).

The individual symptoms reported by participants, stratified by age group, survey time point, and SARS-CoV-2 status, are reported in Supplementary Table S9 and depicted in Supplementary Figure S3. PASC scores were right skewed with means greater than medians in all age groups and time points (Table 3, Figure 2). At 6 months, 25.2% (29/115) and 21.7% (31/143) of 6- to <12-year-old SARS-CoV-2 positive and negative participants, respectively, had PASC scores exceeding the 5.5-point threshold (difference = 3.5%; 95% CI of the difference: −6.7%, 14.1%). These proportions increased to 27.0% (41/152) and 31.5% (103/327), respectively, at 12 months (difference = −4.5%; 95% CI of the difference: −12.8%, 4.4%). Among 12- to <18-year-olds, 17.9% (5/28) and 9.8% (5/51) of SARS-CoV-2 positive and negative participants had PASC scores exceeding the threshold at 6 months (difference = 8.1%; 95% CI of the difference: −6.9%, 26.6%). Among these children, at 12 months, the difference was 3.5% (95% CI of the difference: −6.4%, 16.7%). GEE analysis revealed no difference in PASC index categorization based on SARS-CoV-2 test status in either age groups or time points. The presence of a chronic pre-existing condition was however associated with a PASC score exceeding the threshold in children aged 6 to <12 years of age at the 12-month follow-up (aOR: 2.64; 95% CI: 1.82, 3.84). In our sensitivity analysis, we did find that at 6 months, 17.9% (5/28) and 2.0% (1/51) of 6- to <12-year-old SARS-CoV-2 positive and negative participants, respectively, had PASC scores exceeding the 5.5-point threshold (difference = 15.9%; 95% CI of the difference: 1.0%, 31.4%).