This is a prospective, multicenter (12 U.S. centers) survey of parents of children from a subset of Tier 1 (3,568 patients from 46 centers in 10 countries) conducted between January 1, 2020, and April 30, 2021 (Supplementary Table S1). Centers screened using locally approved methods including chart review and hospital registries. Local regulatory approval was obtained at each study site. Adult participant and child assent informed consent forms were used per local regulatory guidelines. The University of Pittsburgh Institutional Review Board (STUDY20060012) approved the Data Coordinating Center (DCC) at the University of Pittsburgh to receive and analyze de-identified data. Families were contacted by sites using a variety of IRB-approved methods, thus representing a convenience sample.

A study manual of operations was created and disseminated to sites. Study procedures were reviewed on DCC-led webinars. A custom, secure REDcap Case Report Form was created for parent input of data (Supplementary Appendix). Data collected included a symptom list (new since discharge or present during hospitalization and continued post-discharge), health care utilization (emergency department or hospital admission), new medications, new referrals, proxy-report child HRQOL [PedsQL (24)], global function [PROMIS Parent Proxy Global Health and Functional Status Scale (FSS) (25, 26)], family [PedsQL Family Impact Module (27)], school and vaccination status, and socioeconomic outcomes of the pandemic [Coronavirus Impact Scale (CIS) (28)] (Supplementary Table S2).

Each parent was assigned a unique study identification number by the DCC and provided by the local site upon informed consent. This gave the family access to complete the study surveys within REDcap and allowed the DCC to merge patient-specific data with post-discharge outcomes. If families were unable to complete the surveys directly in REDcap, local study teams could offer assistance.

The Clinical Research, Investigation, and Systems Modeling of Acute Illness (CRISMA) Center at the University of Pittsburgh managed central data collection, quality, security, and analysis. Centers with a data use agreement in place with the University of Pittsburgh previously submitted Tier 1 patient data during the index hospitalization to the DCC using encrypted email or via upload to a secure cloud (https://www.globus.org/). Race and ethnicity were collected in the parent study to evaluate as risk factors for outcomes given prior literature. Data from Tier 1 were linked with data collected prospectively for this study. Data were stored on a password-protected network, with additional periodic secure offsite backups to the database. Data were screened for missing or implausible information and queries were issued for clarification and adjusted.

The primary outcome was a parent report of recovery status at the time of follow-up as either recovered or not recovered. Caregivers were asked, “Do you feel your child has fully recovered from his or her COVID-19 or MIS-C related illness?” The possible answers were Yes, No, or Unsure, with No and Unsure combined into “Not Recovered” status for a binary outcome. Secondary outcomes are listed in Supplementary Table S2. Poor HRQOL was defined as a total PedsQL score less than 1 standard deviation below the mean of the sample.

Non-parametric tests were used, and data were presented as median (interquartile range). Comparisons were made between children post-discharge recovered vs. not recovered. Backwards stepwise multivariable logistic regression modeling was performed to identify patient, family, disease, and treatment characteristics associated with recovered status and HRQOL. Covariates included for recovered status that had less than 25% missing and p < 0.1 in univariate analysis: sex, any comorbidity, steroid and intravenous immune globulin (IVIG) treatment, acute COVID vs. MIS-C status, lymphocytes and C-reactive protein on admission, and CIS score. Covariates included for HRQOL that had less than 25% missing and p < 0.1 in univariate analysis: family impact total score, neurologic comorbidity, remdesivir and IVIG treatment, new impairment at hospital discharge, CIS score, ethnicity, and acute COVID vs. MIS-C status. Missing data were not imputed (thus sample sizes for variables varied). All p-values were two-sided, and p < 0.05 was considered statistically significant. Data were analyzed using Stata (Version 17, Statacorp LLC, College Station, TX). This article was written according to the STROBE (Strengthening the Reporting of Observational studies in Epidemiology) initiative (29).

Among Tier 1 sites that were invited to participate in Tier 2, 12 sites obtained IRB approval, 10 had at least one family complete informed consent, and 9 had at least one family submit survey data (Figure 1). Among 1,238 patients eligible from Tier 1 at participating sites, 885 were contacted by sites, 186 consented, and 79 completed surveys. Follow-up surveys were completed at 530 (382, 652) days from the hospital admission. Children whose caregivers participated in Tier 2 were numerically younger, of less female sex and minority race, and more frequently had a preexisting comorbidity than children whose caregivers who did not participate in Tier 2 (Supplementary Table S11).

The median age was 6.5 [interquartile range (IQR) 2.0, 13.0] years and 38.0% were female (Table 1). Of the 51 (64.6%) children with a pre-existing condition, 24 (30.4%) had neurologic conditions. Most (75.5%) children had acute SARS-CoV-2 and 36.7% were admitted to the intensive care unit (Supplementary Table S1). Fewer Hispanic patients were represented in Tier 2 (12.1%) than in Tier 1 (31.1%) and slightly fewer patients in Tier 2 were admitted to the ICU (36.7%) compared with Tier 1 (45.8%). Fewer children had impairment at hospital discharge in Tier 2 (9.1%) vs. Tier 1 (17.0%).

Among families responding to the question, “Do you feel your child has recovered from his or her COVID-19 or MIS-C related illness?”, 70.9% defined their child as recovered and 29.1% as not recovered. Recovered and not recovered groups had similar time to survey, personal characteristics, severity of illness, and frequency of impairment (Table 1). Children who were recovered had decreased median (IQR) initial C-reactive protein (CRP) levels compared to children who were not recovered [1.3 (0.4, 6.3) vs. 5.7 (1.3, 25.1) mg/dl, p = 0.02]. Children with MIS-C had higher median (IQR) initial CRP compared with those with acute SARS-CoV-2 [15.0 (5.7, 25.1) vs. 1.0 (0.4, 4.4) mg/dl, p < 0.001].

Overall, 34.2% of patients received treatment directed at a SARS-CoV-2 related condition, with no difference by recovered or not recovered status (28.6% vs. 47.8%, p = 0.12) (Table 3). Steroids (29.1%) and IVIG (20.3%) were the most common therapies provided in the hospital.

Overall, two-thirds of patients had at least one symptom that was new (27.4%) or continued (56.3%) after hospital discharge (Table 2) (Supplementary Table S3).

There were no group differences in the frequency of new/continued symptoms (66.1% vs. 69.6%, p = 1.000). By category, the frequency of new/continued symptoms was more common in the not recovered vs. recovered groups, except for the respiratory category. Headaches were the most frequent new symptom reported for children recovered (5.4%) with very few others reported. Among children not recovered, new symptoms most frequently reported were anxiousness or sleep problems (both 21.7%). The most common continued symptoms for children who were recovered were fatigue (28.6%), weakness (16.1%), cough (16.1%), fever (14.3%), and loss of appetite (14.3%), while for children not recovered, they were fatigue (47.8%), headache (43.5%), muscle aches (39.1%), and weakness (39.1%).

Thirty-one (39.2%) children had at least one emergency department visit since hospital discharge, with a median of 2 (IQR 1, 3) visits (Table 3). There were no group differences in the occurrence of emergency department or hospital readmission.

Post-discharge care was most frequently provided by the child's primary care provider (73.4%), followed by cardiology (35.4%) and neurology (22.8%).

New medications were prescribed post-hospital discharge for heart and lung problems (both 7.6%) and seizures (6.3%). Of the children prescribed steroids (6.3%), more children were not recovered vs. recovered (17.4% vs. 1.8%, p = 0.02).

Total HRQOL [87 (77, 95) vs. 77 (51, 83), p = 0.005] was better in children who were recovered vs. not recovered (Table 4). Subdomain scores were similarly better for the recovered group except for school scores (n = 23 ag ≥2 years), which were not different between groups.

Global functioning as assessed by FSS was better in recovered vs. not recovered groups [6 (6, 6) vs. 6 (6, 8), p = 0.01], but PROMIS measures were not different (Supplementary Tables S5, S6).

Family functioning total score was not different between recovered groups [78 (65, 93) vs. 73 (60, 82), p = 0.13] (Supplementary Table S7).

Many families reported changes in their family's socioeconomics and mental health statuses (Supplementary Table S8). The sub-domains in which the largest numbers of families reported any change included routines (e.g., work, education, social life, hobbies, religious activities) (90.4%) and stress related to the pandemic (83.6%). For changes in routine, most families reported a severe status change with a change in 3 or more categories (32.9%). Parents with children who were not recovered had a worse CIS score than those with recovered children [8 (5, 13) vs. 6 (4, 9), p = 0.03]. They also reported changes in food access (47.6% vs. 21.1%, p = 0.01), accessing mental health treatment (47.6% vs. 19.2%, p = 0.002), and stress due to the pandemic (100.0% vs. 76.9%, p = 0.04).

In the children >5 years of age (n = 50), 80.0% of children were back to exclusive in-person schooling and 14% were exclusively homeschooled (Supplementary Table S9). There was no difference in vaccination status between groups at the time of the survey.

In a multivariable logistic regression, lower CRP [adjusted odds ratio (aOR) 0.90, 95% confidence interval 0.82, 0.99] and higher lymphocyte count (aOR 1.001, 95% CI 2.37–5.15) on hospital admission were associated with recovered status (Table 4). These results were unchanged when acute SARS-CoV-2 vs. MIS-C was forced into the model (data not shown).

A total of 12 out of 79 (15.2%) children had poor HRQOL at follow-up. In a multivariable logistic regression, worse family function scores (0.94, 0.91, 0.98) and neurologic comorbidity (10.57, 1.71, 65.16) were associated with poor HRQOL (Supplementary Table S10).

The generalization of findings from this work is limited by the fact that participating centers are only from the U.S. and the sample size is relatively small and at risk of response bias. General pandemic effects and other primary diagnoses (e.g., trauma) upon index hospitalization may have affected outcomes (20, 70). Limitations include nonresponse bias and parental recall bias of symptoms and recovered status. This research study lacks a control group and does not include a detailed neuropsychological assessment.