Participants

The CLoCk study recruited 31,012 CYP in England aged 11-to-17-years at study invitation and matched on SARS-CoV-2 test result according to month of testing (between September 2020 and March 2021), sex at birth, age, and geographic area. Study design is described in detail elsewhere^7,14. In brief, potential participants were contacted with study information via post by Public Health England (now UK Health Security Agency). Information included a web link for electronic consent and questionnaire completion. SARS-CoV-2 polymerase chain reaction test results were sourced from laboratory information management systems at the UK Health Security Agency, to which reporting by hospitals and laboratories was mandatory during study recruitment.

CLoCk received ethical approval from the Yorkshire and the Humber–South Yorkshire Research Ethics Committee (REC reference: 21/YH/0060). All research was performed in accordance with relevant guidelines/regulations, including the Declaration of Helsinki. All participants provided (electronic) informed consent.

In this analysis, we included CLoCk participants who tested positive for SARS-CoV-2 between January and March 2021 and responded to questionnaires at 3-, 6-, 12-, and 24-months post-testing. This sub-cohort were enrolled 3-months post-testing (i.e., between April and June 2021) and have previously been characterised in Nugawela et al¹⁵. We selected this sub-cohort given definitive ascertainment of SARS-CoV-2 positive status as part of routine national testing, availability of data at four time-points, and completeness of responses over follow-up. We did not include a comparison group of participants testing negative for SARS-CoV-2 at baseline given many of such participants may have been infected during follow-up.

Measures

At all time-points, participants self-completed questionnaires about their physical and mental health, containing elements of the International Severe Acute Respiratory and emerging Infection Consortium Paediatric COVID-19 follow-up questionnaire¹⁶. Participants could ask for help completing questionnaires from parents/carers or by contacting the research team. The 3-month post-testing (i.e., at study enrolment in April-June 2021, 3-months after testing in January-March 2021) questionnaire also collected demographics, and retrospective reports of whether participants often felt very tired prior to the pandemic (in early March 2020) and their main symptoms at their SARS-CoV-2 test (between January-March 2021). At all time-points, fatigue was assessed using the Chalder Fatigue Scale (CFQ)^12,13and single-item assessment¹⁴. In addition, SARS-CoV-2 testing data were linked to the national Personal Demographic Service by the UK Health Security Agency to provide further data on age at infection, sex at birth, and the 2019 English Index of Multiple Deprivation (IMD, computed at small-area level using participants’ residential postcodes).

Chalder Fatigue Scale (CFQ)
^12,13

The CFQ is a reliable 11-item scale of fatigue severity designed for use in hospital and community settings, and which has been validated in clinical and non-clinical samples^12,13. The questionnaire comprises two subscales – physical and mental fatigue. Using a bimodal scoring system, total scale scores range from 0-to-11 and are calculated as the sum of item scores in which four response options of increasing severity (e.g., from ‘less than usual’ to ‘much more than usual’) are assigned values of 0, 0, 1, and 1. Total scores ≥4 indicate ‘case-ness’ (a term which means the score is severe enough to be regarded as a clinical case)¹⁷. Using a Likert-style scoring system (where item responses options are assigned values of 0, 1, 2, and 3), total scale scores range from 0-to-33, with higher scores indicating greater fatigue severity. The Likert-style scoring system is not typically used to define case-ness; therefore our primary analysis focused on the bimodal system with a supplemental analysis using the Likert-style system^18,19,20.

Single-item fatigue assessment

All questionnaires contained several single-item assessments of a broad range of symptoms, including “Are you experiencing unusual fatigue/tiredness?” – with three response options (“No”, “Mild fatigue”, “Severe fatigue – I struggle to get out of bed”). This item has not previously been validated, but previous CLoCk cohort publications identified a high prevalence of fatigue according to this item^1,8, necessitating further exploration of its validity.

Statistical analysis

We characterised fatigue cross-sectionally at each follow-up using descriptive statistics and longitudinally across all follow-ups using linear mixed-effect models.

Descriptive analysis

To characterise profiles of fatigue (research question one) using the standard CFQ bimodal scoring system, we described the total and subscale scores, individual item scores, and the proportions meeting case-ness^12,13 at each follow-up. We report Cronbach’s α at each follow-up as a measure of reliability.

We compared characteristics of participants that met CFQ case-ness at least once during follow-up (ever-cases) to those who never met case-ness threshold (never-cases). Characteristics included age at infection, sex at birth, ethnicity, IMD quintile, and whether participants reported retrospectively at enrolment: having learning difficulties at school and/or an Education Health and Care Plan (EHCP) before the pandemic (the latter indicating a need for extra learning support in school); often feeling very tired in early March 2020; and unusual fatigue/tiredness as a main symptom at testing in January-March 2021. We compared the proportion that met (vs did not meet) the research definition²¹for PCC 3-months post-infection. As per previous studies^8,15, this definition was operationalised as (i) experiencing ≥1 symptom from a pre-specified list of 21 symptoms (including an ‘other’ option) and (ii) ‘some’ or ‘a lot of’ problems with mobility, self-care, doing usual activities, having pain/discomfort, or feeling very worried/sad/unhappy as measured using the EuroQol Five Dimensions Youth scale²².

To assess the validity of the single-item assessment (research question two), we first explored the relationship between CFQ case-ness and single-item assessments via cross-tabulation. Using CFQ case-ness as a benchmark, we then combined ‘severe’ and ‘mild’ single-item responses (as done in previous studies using CLoCk data^1,8) and calculated sensitivity, specificity, Youden’s J, positive and negative predictive values at each time-point. In a supplementary analysis, we compared these metrics using just ‘severe’ single-item responses.

Longitudinal analysis

To investigate fatigue over time (research question three), we used linear mixed-effects regression to model trajectories in fatigue as assessed using the CFQ. For the primary analyses, we used the total score derived from the bimodal scoring system (and used the Likert-style scoring system in supplementary analysis)^12,13. We also investigated trajectories in the mental and physical fatigue subscale scores.

The total CFQ score at 3-, 6-, 12-, and 24-months post-infection was our modelled outcome. Our initial model included time since infection, a constant-term, and a participant-level random intercept only. Time was defined as the number of days between the baseline test and questionnaire completion at each follow-up, divided by 30.25 for interpretation in monthly units. We included time as a linear term and explored whether model fit was improved by including other functional forms (square, square root, cube, inverse). Including these forms led to limited improvement according to the Akaike information criterion values compared to the linear model (all differences <9.5; see Supplementary Table 1). Therefore, we retained the more parsimonious linear model and estimated the predicted mean fatigue trajectory with 95% confidence intervals.

To explore if fatigue trajectories varied by participant characteristics, we sequentially added (to the above-described model) explanatory variables, including both fixed main effects and interactions with time. Each variable was tested in a separate model. Explanatory variables were age, sex at birth, ethnicity, IMD quintile, and binary indicators for learning difficulties at school and/or EHCP status, frequent pre-pandemic fatigue, unusual tiredness/fatigue as a main symptom at acute infection, and fulfilment of the PCC definition 3-months post-infection. Likelihood ratio tests were used to compare models with and without the relevant interaction terms to determine if their inclusion significantly improved model fit. For each initial trajectory model (bimodal and Likert-style scoring), we undertook a series of diagnostics which are described in Supplementary Methods and illustrated in Supplementary Figures 1–2.

All analyses were conducted using R (version 4.4.0)²³ in RStudio. Linear mixed-effects models were constructed using the lmer function from the lme4 package²⁴. Almost all questions were compulsory in the CLoCk questionnaire and, therefore, within the analytical sample there was no missing data by design. Data from CLoCk are publicly available via the UK Data Service (ID: 9203)²⁵. All analyses were pre-specified and exploratory, we therefore did not correct for multiple testing.