Meta-analyses of ORs (cases vs. controls)
It was only possible to perform meta-analyses of ORs comparing cases and controls for 13 symptoms (Supplementary Figure 1). Cases were defined as patients that had a confirmed COVID infection, and controls as patients without COVID. When compared to controls, children with long COVID had a higher risk of persistent dyspnea (OR: 2.69; 95%CI, 2.30-3.14), anosmia/ageusia (OR: 10.68; 95% CI, 2.48, 46.03), and/or fever (OR: 2.23; 95% CI, 1.2-4.07). There was significant heterogeneity for 5 out of the 13 meta-analyses.
The controls were chosen in a very different way among studies, which might have introduced significant heterogeneity. The following were the different definitions of controls: 1) children with other infections (e.g., common cold, pharyngotonsillitis, gastrointestinal, urinary tract infections, pneumonia of bacteria or unknown origin) 16; 2) children with no antibodies testing 18 mixed with other children with other infections 16; 3) children with a negative antibody test 19, 4) children with a negative PCR test that were symptomatic 20; and 5) children who did not have a positive test recorded in the database 14.
The adjustments among studies also varied. Several studies adjusted their OR by age, sex, ethnicity, socioeconomic status, and comorbidities 20. However age and sex 14 only adjusted for sex, only age 16 only adjusted for age, and Knoke et al did not adjust, or by OR without adjusting previous conditions 18 (Supplemental Figures 2 and 3).
The prevalence of symptoms over the course of long COVID for cases and controls is showed in Supplementary Table 1. Given the heterogeneity in the definition of controls and the low number of subjects, no formal statistical comparison was done for the crude prevalence.
Symptoms that were presented in a single study and, therefore, unable to be incorporated into the meta-analyses included: orthostatic intolerance, cold hands/feet, chapped lips, adenopathy, fainting, twitching of fingers and toes, chills, swollen toes/fingers, and hallucinations. One study reported statistically significant differences between clinical cases and controls for systolic blood pressure, left ventricular ejection fraction, relative myocardial wall thickness, and tricuspid annular plane systolic excursion 21. However, given that these variables were only evaluated in this study, we could not perform a meta-analysis for these outcomes.
Studies included in the meta-analyses evaluated whether certain variables increased the risk of long COVID-19 and found that age, sex, severe acute-COVID-19, obesity, allergic disease, and long-term health conditions were associated with high risk to develop long COVID-19 22-25. Further, two of the studies evaluated the duration of symptoms. A study from Denmark reported that symptoms resolved in a minimum of 54–75% of children (varied with age) within 1–5 months 15. Another, from England, which used the UK ZOE COVID Symptom Study app, reported that 4.4% of children still had symptoms four weeks after COVID-19 onset, which decreased to 1.8% at 8 or more weeks 24.
Quality of studies
Regarding the quality of studies, all had a score of 7 or more. Supplementary Table 1 presents a list of methodological strengths or, conversely, limitations for each study. All studies included laboratory-confirmed COVID-19 infection, PCR or antibody test. Two-thirds of the studies included over 100 children. Six meta-analyses had low heterogeneity (I2<25%) for the following symptoms: vomiting and nausea, nasal congestion, dysphonia, urinary problems, neurological abnormalities, and dysphagia. Three meta-analyses had medium heterogeneity for the following symptoms: abdominal pain, changes in menstruation, and speech disturbances. All other meta-analyses had high heterogeneity (I2>75%). It was not possible to stratify by any variable (e.g., age, sex, country, past comorbidities, or severity) to evaluate where the heterogeneity originated.
The prevalence of long COVID in children and adolescents, following a COVID-19 infection was 25.24%. The five most prevalent clinical manifestations were mood symptoms (16.50%), fatigue (9.66%), sleep disorders (8.42%), headache (7.84%), and respiratory symptoms (7.62%). It was only possible to perform meta-analyses of ORs comparing cases and controls for 13 symptoms. When compared to controls, persons with COVID-19 had a higher risk of presenting persistent dyspnea, anosmia/ageusia, and/or fever.
The most frequent symptoms reported were related to mood. COVID-19 pandemic has initiated an explosion of future mental illnesses 26, that is affecting both society as a whole as well as those who recover from COVID-19. Studies have shown that the pandemic has profoundly impacted society by affecting children’s development through isolation, poverty, food insecurity, loss of parents and caregivers, loss of time in education, and increased stress 27. The presence of these symptoms in the general population, regardless of COVID-19 status, has been coined long-Pandemic Syndrome 28.
Interestingly, many of the symptoms identified in these meta-analyses, such as mood, fatigue, sleep disorders, orthostatic intolerance, decreased concentration, confusion, memory loss, balance problems, exercise intolerance, hyperhidrosis, blurred vision, body temperature dysregulation, dysfunction on heart, rate variability and palpitations, constipation or diarrhea, and dysphagia, are commonly present in dysautonomia 29. Dysautonomia is defined as a dysfunction of the sympathetic and/or parasympathetic autonomic nervous system Postural orthostatic tachycardia syndrome, chronic fatigue syndrome (CFS), and myalgic encephalomyelitis (ME) are subclassifications of this condition30. Moreover, the constellation of symptoms because of long COVID can vary from patient to patient, fluctuating in their frequency and severity 31. Several viruses have been shown to trigger ME/CFS, including the Epstein Barr Virus, Ross River virus, and earlier coronaviruses (e.g., SARS and MERS) 32. However, it remains unclear whether dysautonomia may occur as a direct result of the SARS-CoV-2 infection, interaction with other viruses, or due to immune-mediated processes such as cytokines, which are known mediators of the inflammatory response 33-36.
Similar to adults, the following risk factors in the pediatric population were associated with long COVID: older age, female gender, severe COVID-19, overweight/obesity, comorbid allergic diseases, and other long-term co-morbidities. Protective factors leading to milder severity and duration of COVID-19, and possibly also long COVID, in children include fewer comorbidities, strong innate immune responses, reduced expression of ACE2 receptors, and active thymic function, which leads to the increased presence and decreased depletion of T cells which recognize viral proteins. Further protections include a range of environmental or non-inheritable factors such as vaccines, past infections, nutrition, and/or the gut microbiome 22-25,37.
The prevalence of symptoms is highly dependent on how much time has passed after having acute COVID-19. The follow-up time in our meta-analyses varied between 1 to 13 months. Even though most symptoms improve with time 38, there is evidence in adult studies that suggests some symptoms can persist one year after COVID-19 diagnosis 39. It is important to understand which symptoms are associated with certain periods of time, so future studies should assess the prevalence of each symptom at different time points (e.g., 6 months, 12 months, 2 years) to determine which symptoms are associated with which time period.
As with other meta-analyses, the strength of this study centers on the large sample size 40 which helps provide identify the signs and symptoms present after acute SARS-CoV-2 infection.. Further, there were some limitations to our meta-analyses. The quality of the meta-analyses results depends on the quality of the studies included. Table 3 contains a list of all the methodological aspects that future studies need to consider. We can observe that all studies had a high probability of bias, including lack of standardized definitions recall, selection, misclassification, nonresponse, and/or loss of follow-up. Additionally, the included studies have the limitations inherited in all observational studies, including bias due to residual and unmeasured confounding. Another limitation relates to the high level of heterogeneity. To account for heterogeneity, we used a random-effects model 41. However, ideally one should stratify the meta-analysis to identify what is causing the heterogeneity. This was not possible because most studies did not include data on different groups. The differences between studies were likely due to differences in study designs, settings, populations, follow-up time, symptom ascertainment methods, inconsistent terminology, little details on stratification on pre-existing comorbidities, and prior receipt of COVID-19 therapeutics and vaccines. Only four studies mentioned what percentage of the population was already vaccinated 14,15,23,28 (Table 3). It has been shown that vaccines reduce the risk of long COVID. A study in Israel compared the prevalence of symptoms of long COVID and found that fully vaccinated participants who had COVID-19 were 54% less likely to report headaches, 64% less likely to report fatigue, and 68% less likely to report muscle pain than were their unvaccinated control group 42. More studies are needed to analyze the relationship between vaccines in children and long COVID.