Fig 1.
COVID-19 surveillance data for England.
Monthly observations where months before the dashed blue vertical line were used for developing the regression models. (a) confirmed PCR-positive COVID-19 tests; (b) ONS estimated new COVID-19 infections per 100,000; (c) new COVID-19 hospitalisations; (d) ONS average testing positive for COVID-19 each day. Note that ONS data were not published before July 2020 and the ONS stopped publishing incidence estimates after July 2022 [20].
Fig 2.
Overall sickness absence rates in NHS England staff by month.
The yellow, blue and green lines indicate the overall sickness absence rates by month for 2020, 2021, and 2022 respectively. The dark grey line shows the mean sickness absence rate for a fixed month in 2015-2019, and the light grey region highlights the minimum and maximum rate during this period.
Fig 3.
Overall sickness absence rates in NHS England staff by month, broken down by reason.
The main reason behind a member of staff’s sickness absence was recorded in the electronic staff record. This timeseries shows the monthly observations of sickness absence rates from January 2015 until the end of September 2022. Each colour indicates the proportion of the overall monthly sickness absence rate attributed to either S13 cold/cough/flu (pink), S15 chest and respiratory problems (dark yellow), S27 infectious disease (purple), S10 anxiety/stress/depression/other psychiatric illnesses (red), or any other reason (white). Multimedia Appendix 1: Additional Figures provides a more detailed version of this figure, giving a breakdown for all reasons (Fig A1).
Fig 4.
Sickness absence rates in NHS England staff by month for a specific reason.
Each panel shows the trend over time for (a) absence in the mental health-related category; (b) absence for the COVID-19-related category (S13 cold/cough/flu, S15 chest and respiratory problems, or S27 infectious disease).
Table 1.
Regression models estimating the COVID-19 related sickness absence rate.
Coefficients were estimated using data between July 2020 and December 2021. Each numbered row in the table indicates a different model. Columns 2-5 contain the regression coefficients (top) and their corresponding standard error (bottom), with the significance of the coefficient indicated by the number of asterisks. The following asterisk system is to indicate the significance of values:
; *
; **
; ***
.
Fig 5.
Univariate models of COVID-19-related absence.
The panels in each row contain results for a different univariate predictor; new COVID-19 hospitalisations (a-c), ONS estimated COVID-19 incidence (d-f), confirmed PCR-positive COVID-19 tests (g-i), ONS estimated average COVID-19 positivity (j-l). The panels in each column show a different visualisation for a univariate model. (a,d,g,j) Scatterplots of predictor against the absence rate, including the fitted regression line. (b,e,h,k) Timeseries of modelled values (dashed grey) with 95% confidence interval (grey, shaded), forecasts (dashed blue) with 95% prediction interval (blue, shaded), and observed sickness absence trend (solid black). (c,f,i,l) Scatter plot of modelled values against the observed values. Black line is the theoretical line of equality (modelled = observed).
Fig 6.
COVID-19 related absence as a multivariate regression model of new hospitalisations and ONS estimated COVID-19 positivity.
(a) Timeseries of modelled values for 2020–2022 (dashed grey) with 95% confidence interval (grey, shaded), predictions for 2022 (dashed blue) with 95% prediction interval (blue, shaded), and observed sickness absence trend (solid black). (c) Scatter plot of modelled values against the observed values. Black line is theoretical line of equality (modelled = observed).
Fig 7.
SARIMA models of mental health-related sickness absence.
(a) Time series showing the observed trend (solid black line) and predictions. We used a SARIMA model trained with 2015–2019 to predict the 2020–2021 trend (dashed blue line, with the light blue shaded region indicating prediction interval). Another model was trained with 2015–2021 data to predict the 2022 rates (dashed red line, with the light red shaded region indicating prediction interval). (b) Barplot showing the difference between the observed trend and the predictions from the SARIMA model trained on 2015–2019 data (shaded blue bars). Bars above the horizontal zero line (solid black) are months where the observed trend was higher than the model predictions. The dotted line indicates the difference between the observed trend and the 95% prediction interval.
Fig A1.
Overall sickness absence rates in NHS England staff by month, broken down by reason.
The main reason behind a member of staff’s sickness absence was recorded in the electronic staff record. This timeseries shows the monthly observations of sickness absence rates from January 2015 until the end of September 2022. Each colour indicates the proportion of the overall monthly sickness absence rate attributed to each main reason for absence.
Fig A2.
Diagnostic plots to test assumptions of the multivariate linear regression model of new hospitalisations and ONS estimated COVID-19 positivity.
(a) Residuals vs fitted values; (b) Normal Quantile-Quantile (Q-Q) plot; (c) Scale-Location Plot; (d) Plot of residuals against the COVID-19 hospitalisations predictor; (e) Plot of residuals against the ONS average COVID-19 positivity predictor.