COVID-19-related sick leave among 4,721 NHS workers in England and its link to long COVID symptoms

We analysed data from NHS CHECK, a longitudinal survey distributed to all HCWs (clinical and non-clinical), students, and volunteers in 18 NHS Trusts across England [20]. NHS CHECK began in 2020 to study the impact of COVID-19 on the health and wellbeing of NHS staff. Trusts were invited to participate via direct emails to senior leadership teams and purposively selected to offer diversity in geographical location, urban versus rural settings, and acute versus mental health Trusts [20]. The Trusts circulated emails promoting the survey to all eligible staff via existing group emails. Staff support teams/leads, Chief Nursing Officers, Medical Directors, occupational health departments, trade union representatives, and wellbeing hub users were also engaged to promote the study. NHS CHECK was discussed during team briefings, advertised via screen savers on Trust computers, and included in Trust newsletters, news items on Trust intranet websites, and closed social media groups.

Data for NHS CHECK were collected across four waves. Baseline data were collected between April 2020 and January 2021. Follow-up surveys were distributed to participants approximately 6, 12 and 32 months post-baseline. Responses across each wave were linked by participants' email addresses, allowing for longitudinal exploration. To account for attrition between survey waves, a sub-sample of new participants was recruited at 12 months. In total, the NHS CHECK cohort consists of 24,137 HCWs, including 22,554 who began at baseline and 1,583 at 12 months. A full overview of the NHS CHECK methodology and the information collected in the surveys, please refer to the NHS CHECK protocol [20] and pre-registered protocol for this sub-study [21].

Collected on the 12 and 32 month follow-up surveys, we have data on the self-reported total number of sick days and episodes of sickness absence related to a COVID-19 infection. At 12 months, participants were asked to report sickness absence related to any past COVID-19 infection, while at 32 months, participants were asked to only report COVID-19-related absence in the preceding year, ensuring both measurements were unique. Number of sick days was our primary outcome for descriptive analyses.

We used long-term sickness absence as our outcome when looking for predictors of sickness absence. To do this, we created a binary variable using the total number of sick days and episodes, and followed NHS policy to define long-term sickness absence as 4 + consecutive weeks off work [22]. We chose to create this variable as we believed it would be difficult to meaningfully apply our findings to practice if number of sick days was the outcome. Most HCWs reported one episode of absence (58.4% at 12 months and 63.3% at 32 months), so were classified as requiring long-term sickness absence if they reported 20 + days off work. For the remainder who reported multiple sickness episodes, we did not have an objective measure if any lasted for 4 + consecutive weeks. Instead, for each additional sickness episode they reported, we incrementally increased the total number of sick days required to meet caseness, ensuring that at least one sickness episode exceeded 20 + days, i.e. those who reported two sickness episodes required a minimum of 39 sick days to meet caseness, those who reported 3 episodes required a minimum of 58 sick days, etc.

We define participants as having LC if they reported COVID-19 symptoms lasting for 12 + weeks [1]. At 12 and 32 month follow-up, HCWs were asked to report previous COVID-19 infections and those who reported one were then asked to specify which of 24 common COVID-19 symptoms they experienced and for how long. HCWs were also asked to self-report if they had received a formal LC diagnosis from a healthcare professional. We classified HCWs as having LC if they reported experience of any COVID-19 symptom for 12 + weeks [1], or reported a formal diagnosis of the condition. We created an additional variable for formal diagnosis of the condition to allow us to explore the relationship between formal LC diagnoses and sickness absence. Among HCWs we classified as having LC, the most common symptoms were fatigue (51.4%), difficulty concentrating (36.7%), insomnia (35.2%), depression/anxiety (33.0%), and shortness of breath (28.6%) [23].

Variables were included based on input from our PPIE group, and knowledge of the literature [5, 24] and of the variables available for analysis in the NHS CHECK cohort. Demographic covariates in our analyses included age, sex, ethnicity, relationship status and pre-existing respiratory illnesses (i.e. asthma or chronic obstructive pulmonary disease (COPD)). We included symptoms of common mental disorders, measured by the 12-item general health questionnaire (GHQ-12) [25], and used the recommended cut-off score, where scores ≥ 4 indicate probable common mental disorders [25]. The robustness of the GHQ-12 measure has been established even where response options differ slightly from the original (e.g. in this study, there was a small typographical error in one response option of one GHQ-12 item) [26, 27]1. Occupational covariates included job role, perceived support from managers and colleagues, contract type, reported thoughts about leaving the NHS, income, contact with COVID-19 patients, perceived access to personal protective equipment (PPE), and confidence in workplace infection control policies. All covariates were collected in the baseline survey, except for reported thoughts about leaving the NHS, which was collected at both follow-up timepoints.

We pre-registered a statistical analysis protocol on the Open Science Framework [21], and this paper addresses research questions four and five. Using population-level data provided by each participating Trust, the cohort was weighted using a raking algorithm based on the age, sex, ethnicity, and job role profile of the workforce to maximise representativeness. To complete the weighting, missing data were imputed using the kth nearest neighbour algorithm using R v4.0.2 [28]. These imputed data were only used to complete weighting. All percentages in the results have been weighted.

We conducted descriptive statistics on the predictor variables and self-reported sickness absence. During descriptive analyses, we removed two HCWs from the 12 month sample as the number of sickness days they reported were deemed likely errors, as 560 and 817 exceeded the maximum number of days we thought possible for a participant to have acquired by our 12 month follow-up period. When calculating weighted percentages for each predictor, we accounted for the proportion of missing data (since HCWs who did not respond to these questions are still included in subsequent analyses thanks to the use of multiple imputation, as described below). To examine differences in self-reported sickness absence between HCWs who either reported LC symptoms or reported a formal LC diagnosis and those who did not, we conducted Mann Whitney U tests.

To examine predictors of long-term sickness absence, we used regression modelling. We used Multiple Imputation via Chained Equations (MICE) to account for missing data in the baseline predictors. A detailed description of the MICE procedure has been provided elsewhere and is believed to be appropriate in imputing realistic values [23]. Two multi-level, multivariable binary logistic regression models were constructed, with imputed data being pooled using Rubin's rules. The first model used only outcome data collected at 12 months, while the second used only data collected at 32 months. We included only HCWs who reported LC symptoms in these models, and examined data from both timepoints separately, given the notable change in the proportion of HCWs reporting long-term sickness absence. This contributed to the models being under-powered [21]. The models were multi-level as we grouped participants by Trust to account for clustering. In addition, we included a 'burden period' variable in the models, i.e., periods of time at baseline when the reported number of COVID-19 cases and deaths were higher or lower, which likely influenced pressure on HCWs and may have influenced the predictors [20]. Presented odds ratios are adjusted for all variables in the model. As a sensitivity test to examine our categorisation of long-term sickness absence, we re-ran the regression models using data from only those who reported one episode of sickness absence. Data analysis was completed using Stata v.18.5 [29].

A patient and public involvement and engagement (PPIE) group was set up as part of this study. The group consisted of 16 HCWs with experience of LC from a range of demographic and professional backgrounds. The first group meeting in May 2023 was to discuss the project aims and analyses plans. At the second meeting in October 2023, we discussed the initial findings.

Of the 5,246 HCWs who self-reported a previous COVID-19 infection at 12 and 32 month follow-up, 4,721 also self-reported information about sickness absence related to their COVID-19 infection (90.0% of those with a previous COVID-19 infection, and 19.6% of the total NHS CHECK cohort). Table 1 presents demographic information for the total NHS CHECK cohort and the subsamples under investigation in this study, while Table 2 presents occupational information. This includes 2,200 HCWs at 12 months and 2,906 HCWs at 32 months, with 385 HCWs reporting COVID-19-related sickness absence at both timepoints. Of the 4,721 HCWs in the sample for this analysis, 75.1% were female, 81.9% were white, 75.7% were married or lived with a partner, and 83.7% did not report a pre-existing respiratory illness. Just under half of HCWs with a previous COVID-19 infection screened positive for symptoms of probable common mental disorders at baseline. Of occupational factors, 8.1% of the sub-sample were doctors, and the remainder were roughly evenly split between nursing, other clinical, and non-clinical staff. Most HCWs perceived at least moderate support from managers (69.7%) and colleagues (80.3%), 89.4% were on a permanent contract, and 66.5% reported thoughts about leaving the NHS. See Supplementary Tables 1a-b and 2a-b for more detailed breakdown of HCWs' demographic, health and occupational information by those who did and did not report LC symptoms at each timepoint.

At 12 months, 1,945 of the 2,200 HCWs (89.5%) reported taking time off work due to a COVID-19 infection, while 2,361 of the 2,906 (84.6%) reported the same at 32 months. Table 3 gives information on sickness absence reported by these HCWs. At both timepoints, Mann-Whitney U tests found very strong evidence that HCWs who reported LC symptoms attributed more days off work to COVID-19 infection compared with HCWs who did not report LC symptoms (p < 0.001). Similarly, HCWs with a formal LC diagnosis attributed more days off work to COVID-19 at both timepoints compared to HCWs who did not (p < 0.001). At 12 months, 20.8% of HCWs with LC symptoms also reported long-term sickness absence due to COVID-19, with 34% of HCWs with a formal LC diagnosis reporting the same. The proportion of HCWs requiring long-term sickness absence reduced from 12 months to 32 months, as only 4% of HCWs with LC symptoms and 13.3% of HCWs with a formal LC diagnosis required it at 32 months.

Of the 2,657 HCWs who reported a COVID-19 infection at 12 months, including those who did not report sickness absence, 1,164 (43.8%) also completed the 32 month survey. Of these, 1,145 HCWs (98.6%) reported if they still worked in the NHS at 32 months, while 992 (85.2%) reported if they were currently absent from work due to ill health (see Table 4). Most HCWs (n = 1,113, 95.4%) were still working in the NHS, even if they had moved to a different Trust, while a minority (n = 32, 3.2%) reported that they had left and the remainder (n = 19; 1.4%) did not respond. There were many reasons for choosing to leave the NHS, with only six reporting that this decision was related to COVID-19. Similarly, the majority were not currently absent from work (n = 950, 80.5%). Of those who were off sick (n = 45, 3.7%), only nine said their absence was related to COVID-19, LC, or a common LC symptom.

After demonstrating that sickness absence was higher among people who reported LC symptoms, we explored predictors of long-term sickness absence among HCWs with LC symptoms. See Table 5 for the results of these analyses. At 12 months and adjusting for all variables, we found strong evidence (p = 0.001) that HCWs with a pre-existing respiratory illness had over 3 times the odds of reporting long-term sickness absence compared to those without (adjusted odds ratio, aOR: 3.45, 95%CI 1.91, 6.23). We found weak evidence (p = 0.030) of a similar association (aOR = 2.79) at 32 months. At 12 months, we found strong evidence that HCWs aged 41 and older had between 3 and 5 times the odds of long-term sickness absence compared to those aged 30 and younger (41–50 age group aOR: 3.32, 95%CI 1.47, 7.52; 51–60 age group aOR: 3.96, 95%CI 1.38, 11.34; 61 and older age group aOR: 4.99, 95%CI 1.70, 14.61). At 32 months, however, we only found weak evidence (p = 0.048) of an association among HCWs aged 41–50 (aOR: 3.62, 95%CI 1.01, 12.96). At 12 months only, we found moderate evidence (p = 0.020) that non-clinical HCWs had lower odds than nursing staff of reporting long-term sickness absence (aOR: 0.43, 95%CI 0.21, 0.86). As a sensitivity test, we reran the model with only HCWs who reported one episode of sickness absence and found that no new statistically significant associations were observed. With the exception of being aged 41–50 at 32 months, we also found strong evidence of the aforementioned predictors being associated with increased odds of requiring COVID-19-related long-term sickness absence, though we note that the aOR increased for all (aOR among non-clinical HCWs at 12 months decreased as the sole association < 1.00).

The primary strength of this study was its longitudinal nature, which allowed us to examine predictors of sickness absence using data that were collected at least one year before staff reported data on COVID-19 infections. Additionally, using data provided by each of the Trusts, our sample was weighted to better represent the population from which they were drawn. Finally, our PPIE group helped to ensure that this study addressed issues pertinent to those experiencing the condition, e.g. by advising on what predictors we should include in our analyses, and shared their lived experiences which informed our interpretation of the results.

The results of our study must also be considered within some limitations. Firstly, data on sickness absence and on COVID-19 symptoms and their duration were self-reported. This may have influenced results, as these were not objective measures and may have been biased. For example, a minority of HCWs indicated that they did not take any sickness absence due to their COVID-19 infection, despite clinical guidance at both data collection timepoints mandating that all people with a COVID-19 infection complete an isolation period. This included clinical and non-clinical HCWs who could not reasonably work from home and should require time off work when infected with COVID-19. We cannot be certain if this is a biased measurement or if they are true estimates from HCWs who chose not to or were not allowed to take time off work. Secondly, we did not have an objective measure of whether HCWs required 4 + consecutive weeks off work at once. To overcome this, we approximated the number of HCWs who took long-term sickness absence using the total numbers of days off work and the total numbers of sickness episodes they reported. It may have been, however, that some HCWs who did take long-term sickness absence may not have been correctly categorised. Our sensitivity analysis, however, indicated that similar associations were present between our predictors and outcome when only including HCWs who reported one sickness episode.

Our research was not able to examine every factor which may have an impact on COVID-19-related sickness absence. For example, evidence suggests vaccination against COVID-19 significantly may reduce adult's risk of developing LC [41], which in turn likely lowers their risk of requiring time off work. Similarly, studies suggest that infections related to the Omicron variant of the COVID-19 virus were associated with lower prevalence of LC and symptom severity compared to the Alpha and Delta variants, and that different variants are associated with different LC phenotypes [42], indicating that those infected by the Omicron variant may also require less sickness absence. While these variables are not available in NHS CHECK, we note that future studies should explore them, as they may interact or confound predictors that we have identified, as well as other unseen variables such as time since last vaccine, vaccine type (e.g. mRNA or other), or number of COVID-19 infections. A complete view of LC is important to understanding the mechanism(s) behind it.

Additionally, when looking for predictors of long-term sickness absence, we chose to analyse the data collected at 12 or 32 months separately given noticeable differences in reporting on sickness absence between both timepoints. This resulted in our regression model being under-powered [21], as evidenced by our wide confidence intervals. We recommend future studies to explore this in more detail with a sufficiently powered sample, which would allow for more robust findings. We also did not examine the severity of LC symptoms, and so are unable to examine if symptom severity was related with sickness absence, which previous research has explored [31]. Finally, while we offer an insight into absenteeism in our sample, we do not offer insights into other aspects of work participation, such as phased return to work and reasonable adjustments, presenteeism, or medical retirement due to LC. Future waves of NHS CHECK data collection will seek to include indicators on these aspects when examining the impact of COVID-19 and other illnesses on work participation.