Risk factors for long COVID in children and adolescents: a systematic review and meta-analysis

Daniel G. Rayner ·Elaine Wang·Cloris Su·Om D. Patel·Stephanie Aleluya·Alessandra Giglia·Evelyn Zhu·Maha Siddique

Abstract Background The long-term sequelae of COVID-19 in children and adolescents remain poorly understood and characterized.This systematic review and meta-analysis sought to summarize the risk factors for long COVID in the pediatric population.Methods We searched six databases from January 2020 to May 2023 for observational studies reporting on risk factors for long COVID or persistent symptoms those were present 12 or more weeks post-infection using multivariable regression analyses.Trial registries,reference lists of included studies,and preprint servers were hand-searched for relevant studies.Random-effects meta-analyses were conducted to pool odds ratios for each risk factor.Individual study risk of bias was rated using QUIPS,and the GRADE framework was used to assess the certainty of evidence for each unique factor.Results Sixteen observational studies (N =46,262) were included,and 19 risk factors were amenable to meta-analysis.With moderate certainty in the evidence,age (per 2-year increase),allergic rhinitis,obesity,previous respiratory diseases,hospitalization,severe acute COVID-19,and symptomatic acute COVID-19 are probably associated with an increased risk of long COVID.Female sex,asthma,comorbidity,and heart diseases may be associated with an increased risk of long COVID,and Asian and Black races may be associated with a decreased risk of long COVID.We did not observe any credible subgroup effects for any risk factor.Conclusions The current body of literature presents several compelling risk factors for the development of long COVID in the pediatric population.Further research is necessary to elucidate the pathophysiology of long COVID.

Keywords COVID-19·Long COVID·Post-acute COVID-19 syndrome·Pediatrics·Risk factors

Introduction

Post-COVID syndrome,also known as long COVID,is characterized by persistent symptoms,such as brain fog,fatigue,and dyspnea,present at least 3 months from the onset of acute COVID-19 infection [1— 3].The prevalence of long COVID remains uncertain,with an estimated prevalence of 10%—20% of COVID-19 survivors who experience persistent symptoms following acute infections [4].

Children with long COVID may require long-term clinical support and follow-up,and long COVID may have substantial economic consequences [5].Evidence from the adult population suggests that individuals with long COVID have greater mortality and healthcare utilization in the year following their acute COVID-19 infection compared to noninfected individuals [6].Thus,it is critical to identify the risk factors for the development of long COVID in children previously infected with SARS-CoV-2.

Several risk factors,including demographic characteristics and comorbidities,have been found to be associated with long COVID in the adult population [7].However,research exploring the risk factors for long COVID-19 are limited in the pediatric population.Previous systematic reviews have identif ied older age,female sex,severity during acute COVID-19 infection,and hospitalization as potential risk factors for long COVID in this population;however,their results were limited to narrative synthesis,and the robustness of these risk factors remains unclear [8,9].Thus,this systematic review aims to identify and summarize the potential risk factors associated with the development of long COVID in children and adolescents and to quantitatively pool their results.

Methods

We reported this systematic review and meta-analysis in accordance with the Preferred Reporting Items for Systematic Reviews of Interventions (PRISMA) 2020 statement and guidance from the Cochrane Handbook [10,11].The completed PRISMA checklist is presented in Table S1.We prospectively registered this review on PROSPERO(CRD42023402878).

Study identif ication

A reviewer (DR) experienced in systematic reviews of prognosis conducted a database search of MEDLINE,Embase,CENTRAL,CINAHL,Web of Science (Core Collection),and PubMed Clinical Queries from January 1st,2020,to May 22nd,2023.We hand-searched the reference lists of the included studies,preprint servers (medRxiv and Research Square),and clinical trial registries (ClinicalTrials.gov and the WHO International Clinical Trials Registry) for additional eligible records.The full search strategies are presented in Tables S2— S7.

Eligibility criteria

We included cohort,case—control,and cross-sectional studies of children and adolescents (aged ≤ 17 years) with conf irmed or suspected COVID-19,as well as secondary analyses of randomized controlled trials.Studies reporting neonates and infants were also eligible for inclusion.Eligible studies evaluated predictors of long COVID-19 or persistent symptoms using multivariable regression analyses and reported adjusted risk ratios (RRs),hazard ratios(HRs),or odds ratios (ORs).Studies assessing long COVID or persistent symptoms with an average follow-up duration of less than 12 weeks were excluded.We did not impose any language restrictions.When multiple publications analyzed the same cohort,we included all unique predictors with the largest sample size possible.Eligible risk factors were those related to the characteristics of the patients,their acute infection,or symptomatic period.Predictors related to the medical therapies received by children and adolescents (e.g.,previous vaccination,antivirals) were not eligible.

Study selection

Eight reviewers (DR,EW,CS,OP,SA,AG,EZ,and MS)performed title and abstract screening independently and in duplicate using Covidence (Veritas Health Innovation,Melbourne,Australia).Records deemed eligible for inclusion by two reviewers were subsequently retrieved and underwent an independent and in-duplicate full-text screening process.We resolved disagreements during the title and abstract and fulltext screening phases through discussion or by recruiting a third reviewer to attain consensus.

Data extraction

Data extraction was performed independently and in duplicate by two reviewers (DR,EW) using standardized and pre-piloted extraction sheets.Data related to the study participants,the center(s),recruitment period,and prevalence of long COVID were extracted.From each included study,reviewers extracted key information about each predictor,including the prevalence,def inition,effect estimate and conf idence interval (CI),and all covariates included in the f inal regression model.In case of missing or unclear data,we made attempts to contact the corresponding authors for unpublished information or to seek clarif ication.

Risk of bias assessment

We assessed the risk of bias of our included studies using the Quality in Prognostic Studies (QUIPS) tool [12].The QUIPS tool assesses risk of bias in six domains: (1) study participation,(2) study attrition,(3) prognostic factor measurement,(4) outcome measurement,(5) study confounding,and (6) statistical analysis and reporting.When three or more domains were judged to be at a moderate risk of bias or if one or more domains were judged to be at a high risk of bias,we classif ied the study as being at an overall high risk of bias;otherwise,we classif ied the study as being at an overall low risk of bias.In assessing the risk of bias related to the adjustment of important confounders (as part of the study confounding domain),we judged studies to be at a low risk of bias if they adjusted for age,sex,obesity,any measure of acute COVID-19 severity (including but not limited to hospitalization,ICU admission,or number of symptoms),and at least one chronic comorbidity.If they did not adjust for this minimum set of covariates,we rated their approach to adjustment as having a moderate or high risk of bias.Risk of bias assessments were completed independently and in duplicate by two reviewers (DR,EW).We resolved disagreements through discussion to attain consensus.

Data synthesis and data analysis

We classif ied risk factors as follows: (1) those related to the demographic characteristics of children and adolescents,(2) comorbidities,and (3) infection and post-infection complications.The included studies reported point estimates and 95% CIs in the form of HRs,ORs or RRs.We converted all estimates from each study to ORs using baseline risk estimates from each respective study.When needed,we estimated the baseline risk of long COVID in individual studies using the prevalence of the risk factor,the overall risk of long COVID,and the reported measure of association between the risk factor and long COVID[13].

When studies reported continuous predictors as ordinal variables,if we observed a linear relationship between the increasing ordinal variable and the odds of the outcome,we averaged the beta coefficients across categories to obtain an effect estimate associated with a unit change for the meta-analysis.We conducted sensitivity analyses to evaluate whether these transformed measures differed from those originally reported as continuous.When possible,we pooled effect estimates for each predictor through the inverse variance approach using random-effects meta-analysis.

We addressed statistical heterogeneity through visual inspection of forest plots and evaluated it based on the consistency of point estimates and the overlap between the CIs across the individual studies.Heterogeneity was not assessed with theI2 statistic,as this statistical measure is not useful in prognostic studies with very large sample sizes and narrow CIs [14].When 10 or more studies were included in the meta-analysis,we constructed funnel plots and assessed publication bias through visual inspection.

The STATAmetanandmetafunctions provided the platform for all statistical analyses in this systematic review.All comparisons applied a two-sidedPvalue of 0.05 or less to denote statistical signif icance.

Subgroup analyses

Our review focused on four pre-specif ied subgroup analyses:(1) study risk of bias (high vs.low),(2) duration of followup (<6 vs.≥ 6 months),(3) hospitalization status (inpatient vs.outpatient),and (4) follow-up method (electronic health record vs.in-person visits,phone calls,and online surveys).For all aforementioned subgroup analyses,if we observed a signif icant difference across groups,we restricted our f inal meta-analysis to studies with a low risk of bias,the same duration of follow-up,the same hospitalization status,and/or those assessing outcomes using the approach and applied our certainty of the evidence assessments only to these studies.

Certainty of evidence

To assess the certainty of evidence related to a given risk factor,we used the Grading of Recommendations,Assessment,Development,and Evaluation (GRADE) framework[15].For each risk factor,using the overall risk of long COVID and the prevalence of the factor,we converted the relative measure of association (OR) to an absolute risk difference.We used this absolute risk difference in our judgment of imprecision.We used a 5% absolute risk difference threshold to signify a meaningful risk factor.Our judgment ref lects our certainty in the inference that the true absolute risk difference lies around these predetermined thresholds[16].We presented the synthesized results and their associated certainty in the evidence in a summary of f indings tables.

Results

The database searches yielded 15,214 records,of which 11,421 were screened following de-duplication.One hundred and twenty-one full texts were retrieved for further screening,and 15 studies were included.A list of citations excluded during the full-text screening phase can be found in Result S1.In addition,one relevant study was retrieved from the reference lists of the included studies.In total,16 observational studies (N=46,262) were included in the current review [17— 32].The PRISMA f lowchart for the study selection process is illustrated in Fig.1,and study characteristics are tabulated in Table S8.The included studies reported a median of 664 (range: 58—20,601) participants with a median mean age of 9.82 (range of means:4.33—16.50) years.More than half of the included studies were conducted in Europe (N=9),followed by Asia (N=3)and North America (N=2).The symptoms considered as long COVID manifestations by the included studies are presented in Table S9.Risk factors not amenable to metaanalysis are reported in Fig.S1.

Fig.1 PRISMA f low diagram for study selection

Risk of bias of individual studies

Demographic characteristics

Our review identif ied 6 risk factors related to children and adolescents’ demographic characteristics that were amenable to meta-analysis (Table 1;Figs.S2— S7).With moderate certainty in the evidence,age (per 2-year increase) is probably associated with an increased risk of long COVID.Low-certainty evidence suggests that Asian and black races may be associated with a decreased risk of long COVID,and female sex may be associated with an increased risk of long COVID.Low-certainty evidence also suggests that mixed race may not be associated with the risk of long COVID.We are very uncertain about the association between the index of multiple deprivation and the risk of long COVID.None of our subgroup or sensitivity analyses proved meaningful for any of the demographic characteristic factors(Figs.S8— S15).Visual inspection of the funnel plots showed no publication bias for age (Fig.S16),but we observed compelling publication bias for the meta-analysis investigating female sex as a risk factor (Fig.S17).

Comorbidities

We identif ied 10 risk factors related to comorbidities that were amenable to meta-analysis (Table 2;Figs.S18— S27).With moderate certainty,allergic rhinitis,obesity,and previous respiratory diseases probably increase the risk of long COVID;however,previous neurological disorders are probably not associated with the risk of long COVID.Lowcertainty evidence indicates that asthma,any comorbidity,and heart diseases may be associated with an elevated risk of long COVID.We are very uncertain about the prognostic value of dermatitis and eczema,gastrointestinal problems,and urticaria on the risk of long COVID.None of our subgroup analyses proved meaningful for any of the comorbidity risk factors (Figs.S28,S29).

Infection complications

We identif ied three risk factors related to infection complications that were amenable to meta-analysis (Table 3;Figs.S30— 32).With moderate certainty in the evidence,hospitalization during acute COVID-19,severe acute COVID-19,and symptomatic acute COVID-19 are probably associated with an increased risk of long COVID.We were unable to explore whether our subgroup analyses proved meaningful for any of the infection complication factors.

Discussion

Principle f indings

Our systematic review identif ied several risk factors for the development of long COVID in the pediatric population.With moderate certainty in the evidence,we found that age (per 2-year increase),allergic rhinitis,obesity,previous respiratory disease,hospitalization,severe acute COVID-19,and symptomatic acute COVID-19 are probably associated with an increased risk of long COVID.In addition,based on low certainty evidence,female sex,asthma,comorbidity,and heart disease may be associated with an increased risk of long COVID,and Asian and Black races may be associated with a decreased risk of long COVID.The evidence remains very uncertain regarding the prognostic value of the index of multiple deprivation,dermatitis and eczema,gastrointestinal problems,and urticaria regarding the development of long COVID in the pediatric population.

Strengths and limitations

Our systematic review has several notable strengths.We systematically searched all relevant bibliographic databases and included studies regardless of the language of publication.In addition,we manually searched gray literature sources and contacted authors regarding unpublished data.Therefore,our inferences regarding the reported risk factors are comprehensively informed by all published and unpublished publications on this topic.Furthermore,we sought to limit confounding bias that may have been caused by known and measurable risk factors by limiting our included studies to those using multivariable regression analyses.To clinically contextualize the f indings of our review,when summarizing our f indings,we translated our pooled relative effects to absolute risk differences.Finally,our review also benef ited from the GRADE framework,which informed our certainty in the conclusions we drew regarding our identif ied risk factors.

Our systematic review also had several notable limitations.For certain risk factors,such as continuous age,the measure of association may not meet the linear relationship assumption of regression analyses.The current methods for systematic reviews and meta-analyses on risk factors are unable to incorporate analytic techniques for handling non-linear relationships,such as splines.Moreover,the studies included in our review used a variety of def initions and measurement techniques to capture the phenomenon of long COVID,which may represent a source of heterogeneity in our analyses.Finally,the studies included in our systematic review adjusted for a variety of covariate sets in their multivariable regression analyses.Rarely did we observe two studies adjusting for the exact same set of confounders.These different sets of covariates may have contributed to the statistical heterogeneity that we observed in some of our analyses.To account for these methodological differences,we conducted all meta-analyses using random-effects models.

Relation to previous work and implications for future work

Since the beginning of the COVID-19 pandemic,several studies have sought to investigate the potential prognostic value of various demographic characteristics for the development of long COVID in children.Consistent with previous systematic reviews,our study found that older age is associated with a greater risk of long COVID [8,9].Furthermore,our study identif ied female sex as a potential risk factor,which has shown mixed prognostic value in the published literature [8,9,27].However,through visual inspection of our funnel plots,we identif ied compelling evidence for publication bias in relation to female sex as a risk factor.The prognostic value of female sex in predicting long COVID needs to be further explored and conf irmed in future studies.

Our systematic review also found several underlying medical conditions,including allergic rhinitis,obesity,asthma,respiratory disease,and heart disease,that may be compelling risk factors for long COVID in the pediatric population.These f indings agree with previous studies demonstrating that obesity,respiratory disease,and heart disease are key prognostic factors for hospitalization and severe COVID-19—known risk factors for long COVID [8,33,34].Conversely,our f indings for allergic rhinitis and asthma disagree with previous f indings,which suggest that they have no prognostic benef it for adverse COVID-19 outcomes [35,36].Further research is warranted to explore the relationship between asthma and other allergic conditions and the risk of long COVID-19 in children.

Our review identif ied low-grade evidence that Black race may be associated with a decreased risk of long COVID in children and adolescents.This f inding differs from previous work that found that Black adults are at a greater risk of long COVID and severe acute COVID-19 [37,38].Our f indings may disagree with previous studies due to differences in sample size and study design.The cross-sectional nature of the studies evaluating Black race as a risk factor [19]does not permit the adjustment of competing risks,such as death following COVID-19.Furthermore,Atchison et al.’s study populations included a small proportion of Black children and adolescents [19].Further research is necessary to explore and validate the prognostic value of Black race in long COVID-19 risk.

Out in the king s garden, said she, under the great oak that stands on the left hand, just as one goes out from the castle, is a little bush, rather brown than green, with hairy leaves and long spikes4

The results of our systematic review may have important public health utility in identifying which children with SARS-CoV-2 infection are likely to develop COVID-19 in the future.Previous studies suggest that COVID-19 vaccination may reduce the risk of long COVID in the pediatric population [26,27].Likewise,studies from the adult population have also suggested potential benef its of vaccination in reducing the risk of long COVID-19 [39].The robust risk factors identif ied in this systematic review may be used to identify vulnerable children who may benef it from targeted public health interventions to promote COVID-19 vaccination uptake.

Conclusion

In this systematic review and meta-analysis,we identif ied several compelling risk factors for the development of long COVID in children and adolescents.We demonstrated that increasing age,allergic rhinitis,obesity,previous respiratory disease,hospitalization,severe acute COVID-19,and symptomatic acute COVID-19 are probably associated with an increased risk of long COVID.Further research is required to explore and conf irm other risk factors of interest,which will enable a greater understanding of the pathophysiology of the condition and may lead to the development of future treatment and prevention strategies for long COVID.

Supplementary InformationThe online version contains supplementary material available at https:// doi.org/ 10.1007/ s12519-023-00765-z.

Author contributionsConceptualization: DGR.Methodology: DGR.Investigation: DGR,EW,CS,ODP,SA,AG,EZ,MS.Data curation:DGR.Formal analysis: DGR.Project administration: DGR.Resources:DGR.software: DGR.Validation: DGR.Visualization: DGR.Writing—original draft: DGR.Writing—review and editing: EW,CS,ODP,SA,AG,EZ,MS.Supervision: DGR.All the authors approved the f inal version of the manuscript.

FundingNone.

Data availabilityAll available data associated with this study are presented in the supplemental material.

Declarations

Conflict of interestNo f inancial or non-f inancial benef its have been received or will be received from any party related directly or indirectly to the subject of this article.

Ethical approvalThis is a systematic review of previous published studies;thus,ethics approval was not required.