The psychosocial impact of the COVID-19 pandemic on 4,378 UK healthcare workers and ancillary staff: initial baseline data from a cohort study collected during the first wave of the pandemic

Study design and participants

The sample for NHS CHECK includes all staff working in participating NHS Trusts, including medical, nursing, midwifery, allied health professionals, support, administrative, management, and students who volunteered or were fast-tracked into clinical roles during the COVID-19 pandemic. In this paper we report the results from the initial lockdown period of the pandemic (April-June 2020) during which there were ongoing high levels of COVID-19 hospital admissions and mortality. At that time, the study had been launched across three NHS Hospital trusts in South East London serving two of the Greater London boroughs that were in the top ten for highest rates of COVID-19 related deaths during this period.[8]

Procedures

We identified eligible participants via the Trusts’ human resources (HR) systems. We used the Trusts’ existing group emails to distribute communications to NHS staff explaining and promoting the study. A dedicated NHS CHECK recruitment email was sent by senior management explaining the study, with a link to the study website. A variation of this email was sent repeatedly during baseline recruitment. We also used existing team-based local contacts established through in-reach staff support, for example staff support teams/leads, chief nursing officers, medical directors, occupational health departments, trade union representatives, and wellbeing hub users. NHS CHECK was promoted during team briefings, included in Trust newsletters, news items on Trust intranet websites, closed social media groups and screen savers on Trust computers etc.

Data collection and materials

A survey was developed with input from the NHS CHECK Patient and Public Involvement (PPI) advisory group and we tested the acceptability of the questions, materials, and data collection procedures with this small informal reference group of front-line staff (psychologists, managers, administrators, security staff, intensivists, and trainees) and refined it accordingly. Participation in the baseline survey of NHS CHECK involved completing a short (5-10 minutes) online questionnaire that was hosted using Qualtrics questionnaire software, with the option to complete an additional, longer (10-15 minutes) questionnaire if they wished. Participant information provided on the study website reinforced that participation was voluntary and confidential. Participants provided online informed consent before starting the online survey. NHS CHECK was launched on the 24^th of April 2020, 5 weeks after the initial lockdown in the UK began (23/03/20).

Statistical analysis

Aggregate-level population data were provided by participating Trust HR departments including information on the number of employees and the age, sex, and ethnicity composition of the workforce. From this, we calculated a response rate for each Trust. Response weights were generated using a raking algorithm based on age, sex and ethnicity. For weighting purposes only, missing demographic data were imputed using multiple imputation. Missingness was no more than 6%, 2%, and 2% for age, sex, and ethnicity, respectively, before imputation.

The analysis proceeded in five stages. First, we described the representativeness of the sample by comparing composition in terms of age, sex, and ethnicity with the HR population data using frequencies and percentages. Second, we summarized survey participants with appropriate descriptive statistics for each variable (frequencies and weighted percentages for categorical variables, mean and standard deviation for continuous variables). Third, we described differences between individuals who completed both the ‘short’ and ‘long’ baseline surveys versus those completing only the ‘short’ survey. Fourth, we summarized the weighted prevalence of the primary (GHQ) and secondary outcomes (GAD-7, PHQ-9, AUDIT, PCL-6, MIES, and CIS-R) for all participants and stratified by socio-demographic and occupational factors. Finally, we used multivariable binary logistic regression models to explore relationships of socio-demographic and occupational factors (age, sex, ethnicity, role), as well as moral injury with the primary and secondary outcomes, adjusting for all covariates in the models. All analyses were conducted using Stata 16.

Ethical approval

Ethical approval for the study was granted by the Health Research Authority (reference: 20/HRA/210, IRAS: 282686) and local Trust Research and Development approval.

Demographic composition of the sample in comparison to population

The demographic characteristics of the sample are described in Table 1. In all data presented here, frequencies are unweighted, and proportions are weighted to adjust for Trust size and non-response (Trust-level data provided by Trust HR: see Supplementary Table 1 for full weighted and unweighted proportions). There were some statistically significant differences between those who completed only the short survey (n=2,212), and those who completed both the long and short survey (n=2,166), with men, those identifying as being from racial and ethnic minority groups, those born outside the EU (including UK), and doctors, other clinical staff, and non-clinical staff being less likely than nurses to complete both surveys (see Supplementary Table 2 for full details).

Mental health measures

Table 2 shows the prevalence of probable CMDs, measured by the GHQ, with 58.9% (95%CI 58.1, 60.8) of participants meeting the cut-off score. In terms of secondary outcomes, 23.2% (95%CI 21.3, 25.8) met the cut-off score for probable anxiety (GAD-7), 27.3% (95%CI 25.3, 29.4) for probable depression (PHQ-9), 10.5% (95%CI 9.2, 11.9) for probable alcohol misuse (AUDIT), 30.2% (95%CI 28.1, 32.5) for probable PTSD (PCL-6), and the sample had a mean score of 15.5 (95%CI 15.1, 16.0) on the moral injury scale (MIES). A full breakdown of each measure by all demographics is given in Supplementary Table 3.

In the past two months, 8.5% (95%CI 7.3, 9.8) of participants had considered taking their own life, while 2.0% (95%CI 1.4, 2.7) had attempted suicide, and 3.0% (95%CI 2.3, 3.9) had harmed themselves (see Table 3 for full details).

Logistic regression models

The multivariable logistic regression models showed that the likelihood of meeting the GHQ cut-off score, indicating probable CMDs, was 0.98 (95%CI 0.97, 0.99) times lower for each increased year of age, adjusting for all other covariates. Males were 0.7 (95%CI 0.5, 0.9) times less likely to have probable CMDs than females. Nurses were 2.2 (95%CI 1.5, 3.1) more likely to have probable CMDs than doctors, with other clinical and non-clinical staff also more likely to have probable CMDs than doctors. Those reporting high exposure to potentially morally injurious events were 2.6 (95%CI 2.0, 3.4) times more likely to report symptoms of probable CMDs than those reporting low exposure (Table 4).

These patterns were similar when exploring our secondary outcomes of interest, with older participants and men being less likely to have probable anxiety and PTSD than younger participants and women. Black HCWs were less likely to have probable depression compared to their white peers, while Black, Asian, and other racial and ethnic minority groups were less likely to engage in alcohol misuse than white HCWs. Doctors were less likely to report symptoms of probable anxiety, depression and PTSD compared to all other groups of HCWs. Those with higher exposure to moral injury were more likely to have probable depression, alcohol misuse, and PTSD symptoms. Full details are shown in Table 4.

Statement of principal findings

Our study found a high prevalence of adverse mental health outcomes amongst a sample of London- based HCWs during the first wave of the COVID-19 pandemic. Of particular note, we found concerning numbers of HCWs who reported symptoms of PTSD, and that higher exposure to moral injury was associated with worse outcomes. We identified groups at higher risk of experiencing poor mental health included women, younger staff, and nurses.

Strengths and limitations

The NHS CHECK multi-site study provides findings from three different NHS Trusts, including acute and mental health services. The sites are located in London, which had a high regional case burden of COVID-19 during the first wave when recruitment commenced. This study represents a large sample size (n=4,378) relative to existing literature, with similar demographic characteristics to the overall NHS workforce in terms of age and sex [17], while London Trusts typically have a higher proportion of staff from racial and ethnic minority groups than other areas of the country, which is represented in our weighted data. We had detailed population-level HR information and were able to estimate response rates and the extent of non-response bias. These are notable methodological strengths compared with existing literature which have largely not reported response rates or assessed for related bias.[19] This study was uniquely inclusive in having gathered data from all staff employed by the NHS Trusts, including non-clinical, ancillary, lower-paid, and temporary staff. To our knowledge, this is also the only study that considers moral injury in the context of COVID-19 in HCWs.

However, there are some limitations. Firstly, despite substantial recruitment efforts as outlined above, the response rate was 12% and therefore it is inevitable that findings are open to selection bias, with those for whom the survey had greatest salience (i.e. those who were distressed) probably being most likely to participate. We hypothesise that the participation levels may be reflective of HCW burden related to COVID-19, such as work pressures and research/survey fatigue at this time, as well as longer term decline in HCW participation rates in surveys as observed over recent decades.[20] We found evidence that participants from some demographic and occupational groups, notably those from racial and ethnic minority groups and men, were less likely to participate. Although we took steps to weight the data for such non-participation, differential participation may lead to inaccuracies in prevalence estimates and hampers generalisability of the results. These concerns are compounded by differential response between our short and long surveys, however due to the time limitations on HCWs it was necessary to provide the option to complete only the short survey.

Secondly, it has also been shown that occupational specific surveys of any given group give rise to higher prevalence of mental health distress compared to population studies that contain sufficient numbers of the same occupational group for analysis, suggesting a possible contextual or framing effect.[21] Thirdly, given that all three sites were based in London, we have not captured associated national geographical variations.. Finally, the data was at this stage cross-sectional, meaning we were unable to look at whether symptoms persist, although as noted above, this is preliminary data, and 15 additional sites and longitudinal data will follow. Unfortunately, as this study began in response to the COVID-19 pandemic, we do not have pre-COVID baseline data with which to compare our results to, which is a significant limitation.

Comparison with literature

Unanswered questions and future work

The wider NHS CHECK study, from which the data presented here were taken, is collecting data at baseline, with consent from participants to contact them for follow-up data at six-monthly intervals thereafter, dependent on how the pandemic evolves over time. We have expanded our study population to ensure a broader, more representative geographical spread, and our study now includes 18 sites across England and Wales. In particular, the elevated rates of PTSD identified warrant further investigation in the ongoing study and larger data-set analysis. As data from more sites become available, we will also explore the impact of organisational and social factors on the experiences of staff and, mental health outcomes (e.g. working practices, perceived support from colleagues and managers, COVID-19 impacts and experiences, and staff support programmes available and used).

Further, although we have used validated mental health questionnaires, questionnaire-based surveys of mental health are case-finding, rather than diagnostic tools, which favour sensitivity over specificity. Occupational self-report measures commonly overestimate rates of “caseness”, i.e. individuals who are likely to meet true diagnostic criteria and might thus require specific interventions.[21] Collecting information about distress at a time of great stress means that high proportions of participants reporting difficulties might be expected, however this is not necessarily indicative of being mentally ill. In particular, we need to know whether these symptoms persist and whether they are associated with a reduction in ability to function at work and home, and other meaningful outcomes, for example, help seeking, reduced desire to remain in their post. Hence we will be extending the study in a two stage psychiatric epidemiology design,[40] using standard psychiatric diagnostic interviews administered by phone with a sub-sample of participants, in order to distinguish distress from disorder.

We anticipate finishing baseline data collection in December 2020, with follow-up data collection starting at six-months post-baseline for each individual participant. We will continue to increase response rates through extra engagement tools, including: new patient and public involvement (PPI) initiatives, further work with trade union representatives, offering incentives to participants, and ensuring paper surveys are available in addition to the online version. We will also focus on engaging with racial and ethnic minority groups to address the relatively reduced participation rate and completion of the longer component of the survey by forming partnerships with those experienced in such engagements. Inclusion of an ethnicity-focused module at eight months will help capture inequalities in mental health and occupational outcomes.

Implications

HCWs are experiencing a high burden of adverse mental health outcomes during the COVID-19 pandemic and some demographic and occupational groups are more vulnerable than others. The mental health offering to HCWs must therefore consider the overlap of these factors. For example, our findings indicate that young female nurses may be more vulnerable to adverse mental health outcomes and therefore may benefit from targeted support.

Despite the high levels reported by our study, and others, there remains uncertainty about whether HCWs have worse COVID-19 related mental health outcomes compared to the general population. We urge additional research that emphasises quality and methodological rigor, namely with use of standardised diagnostic interviews with a defined sampling frame. In doing so, the mental health offer will be better able to avoid pathologizing ordinary distress responses, identify those most at risk, and provide necessary support.