Study population
A prospective cohort study was performed and is reported according to STROBE guidelines.19 Patients were recruited from four Dutch GP practices in the Nijmegen region (east of the Netherlands). These practices are members of the family medicine network (FaMe-Net), which is a network of GPs with years of experience in registration and coding according to the International Classification of Primary Care (ICPC).20 Patients from these practices understand that their electronic health records (EHRs) may be used for scientific research.
At the very beginning of the pandemic, when there was no test capacity, FaMe-Net used a specific code (ICPC R83) for patients with a suspicious clinical picture of COVID-19.21 All patients with the ICPC code R83 between 1 March and 31 May 2020 were invited to participate. For every patient with suspected COVID-19, a matched non-COVID-19 patient was invited to participate. The participants were informed that the aim of this study was to investigate the long-term health status of people who had experienced COVID-19 compared with people who had not. They were matched based on sex, age (plus or minus 3 years), and GP practice. The patients had to be at least 18 years of age. The matched non-COVID-19 patients visited the GP during the same period for any complaint, except respiratory complaints or fever. The participants provided informed consent and were aware of the subject of the study.
Measures
The primary outcome in this study was persistent fatigue, measured using the Checklist of Individual Strength 8R (CIS8R). This standardised and validated questionnaire has been used for healthy individuals as well as those with respiratory diseases.22 The CIS8R has good internal consistency (Cronbach’s alpha of 0.92 in patients with rheumatoid arthritis) and good reproducibility (intraclass correlation coefficient 0.81).23 The severity of fatigue was measured with eight questions about the intensity of fatigue in the past 2 weeks on a seven-point Likert scale. The sum score ranges from 8 to 56, which is categorised as ‘no fatigue’ (CIS8R sum score <27) or ‘fatigue’ (CIS8R sum score ≥27).22,24
Based on fatigue scores at 3, 6, and 15 months, the primary outcomes were defined as follows:
‘no persistent fatigue’: ‘no fatigue’ scored on at least one of the follow-ups; and
‘persistent fatigue’: ‘fatigue’ scored on all three follow-ups or at two follow-ups with missing data from the other occasion.
Patients who reported ‘fatigue’ at one follow-up but did not participate on both other occasions, or those who did not participate in any of the surveys, were coded as ‘missing’.
Based on what was known about other post-infectious diseases, information was also collected on sociodemographic, lifestyle, and vulnerability factors. Sociodemographic factors included sex, age, education level, and marital status. Lifestyle factors included BMI, smoking, and alcohol use. Vulnerability factors included ‘neuroticism’, life events, resilience, perceived personalised GP care, comorbidities, medication use, and frequency of contact with a GP.25
Neuroticism was measured using the Eysenck Personality Questionnaire Revised-Short Form (EPQR-S),26 life events using the Brugha questionnaire,27 resilience using the Sense of Coherence-13 (SOC-13) questionnaire,28 and perceived personalised care provided by their GP using the Person Centered Primary Care Measure (PCPCM).29 Higher scores indicated higher levels of neuroticism, more life events, stronger resilience, and higher perceived personalised care from their GP.
Data on smoking, alcohol use, comorbidities, medication use, and frequency of contact with a GP were collected from the EHR. For contact frequency, the number of contacts with a GP in the year before inclusion in the study was used. This variable was made dichotomous, using a cut-off of ≥13 contacts, as this was the 80th percentile. For the number of medications, all active medications were summed. These included all the Anatomical Therapeutical Classification codes that were used during the inclusion period (from 1 March 2020 to 31 May 2020). Medication use was also transformed into a dichotomous variable, using a cut-off of ≥5 medications, which is often used to define polypharmacy. For comorbidities, all patients’ chronic comorbidities were summed (Supplementary Box S1).
Analysis
The data were cleaned and analysed using RStudio.30 Using a multilevel generalised mixed-effects logistic regression (glmer) analysis, the odds ratios (OR) and corresponding 95% confidence intervals (CIs) for the association between the groups (suspected COVID-19 or non-COVID-19) and persistent fatigue were calculated. Levels were added for GP practice and matched patien31t-pairs to correct for clustering and matching. A sensitivity analysis was performed, excluding patients in the non-COVID-19 group who tested positive during the study. Thereafter, univariate multilevel analysis was performed in the suspected COVID-19 group only to determine prognostic factors for persistent fatigue. Individuals with missing data were excluded analysis by analysis. Spearman’s rho was used to check the correlation between the statistically significant prognostic factors.
Multivariable multilevel analysis (forward conditional) was performed, including significant prognostic factors from the univariate analyses, after checking for multicollinearity. Finally, the analysis was performed in the whole study population with interaction terms between the groups and prognostic factors to determine whether the associations between the prognostic factors and persistent fatigue differed between the groups. In the case of significant interaction effects, stratum-specific ORs were calculated and presented. For the multilevel models a P-value of 0.05 was used as a cut-off. For the interaction terms a less strict P-value of 0.10 was used.31
