All studies published between January 1, 2019, and December 31, 2021, that reported on the mental health of frontline HCWs affected by the COVID-19 pandemic, such as insomnia, stress, anxiety, and depression, were identified using EMBASE, MEDLINE, PsycINFO, Cochrane Central Register of Controlled Trials, Cochrane Database of Systematic Reviews, Web of Science, and ERIC. Government databases websites, conference proceedings, and medical society websites were also searched (independently performed by J.T. and W.Q.L.). The investigators screened the reference lists of identified articles using the approaches recommended by the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA; Moher et al., 2009). Computer-based searches used terms related to the mental health of frontline HCWs during the COVID-19 pandemic (see Table 1). Studies which published in peer-reviewed journals that reported data on frontline HCWs and used a validated method to assess insomnia, stress, anxiety, and depression were also included.

Inclusion criteria must be that the study population is frontline healthcare workers in COVID-19 affected countries or areas. Only high-quality studies evaluating the prevalence rates of specific mental health problems such as insomnia, stress, anxiety and depression are eligible for inclusion. However, mental health problems such as post-traumatic stress disorder, psychotic disorder and obsessive-compulsive disorder were excluded because of the limited number of high-quality studies. Additionally, broad terms such as “psychological distress and psychological abnormality” were excluded as they can be difficult to quantify.

The investigators (J.Z. and W.Y.) independently extracted the following information from each article using a standardized form: study design, country, survey period, specialty, sample size, average age, diagnostic methods, screening instrument outcome, and reported prevalence of insomnia, stress, anxiety, and depression. To eliminate studies involving the same population or multiple identical publications. The quality of non-randomized studies was assessed by the modified version of the Newcastle-Ottawa Scale (NOS; Stang, 2010). This scale assesses the quality of the study through five dimensions: sample representativeness, sample size, comparability between respondents and non-respondents, ascertainment of insomnia, stress, anxiety, and depression, and statistical quality (Supplementary Table 1). Studies were judged to be at low risk of bias (≥3 points) or high risk of bias (<3 points). The third reviewer (N.Z.) adjudicates all discrepancies.

Prevalence estimates of insomnia, stress, anxiety, and depression were calculated by pooling the study-specific estimates using random-effects meta-analysis. It accounted for between study heterogeneity. The 95% confidence intervals of the studies were calculated by the Clopper Pearson method, which considered the asymmetry. Study heterogeneity was assessed by standard χ²-tests and I² statistics. Heterogeneity values ≥75% indicated considerable heterogeneity. The characteristics of different study levels were grouped, and hierarchical meta-analysis and meta-regression were carried out (van Houwelingen et al., 2002). The impact of individual studies on the estimation of overall prevalence was explored by sensitivity analysis. Bias secondary to study effects was studied by funnel diagram and Egger's test. All analysis were performed using R Foundation for Statistical Computing (version 4.1.1; Computing, 2022). The statistical test of all studies was two-sided and used a significance threshold of P < 0.05. This study is registered on PROSPERO, number CRD42021253821.

The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.

The development of the research question was informed by the mental health status among frontline HCWs during the COVID-19 pandemic. Patients were not advisers in this study, nor were they involved in the design, recruitment or conduct of the study. Results of this study will be made publicly available through open-access publication where study participants may access them.

Seventeen cross-sectional studies (n = 11,727), one cohort study (n = 4,804), and one RCT study (n = 482) involving 17,013 individuals were included in this study (Figure 1). The study involved 12 countries, including one in North America, one in South America, five in Europe, four in Asia, and one in Africa. Sixteen studies recruited participants from first frontline physicians and nurses, while three recruited participants exclusively from physicians. There were an average of 895 participants per study, ranging from 98 to 4,804. Thirteen studies assessed insomnia using the Insomnia Severity Index (ISI), 2 used the Pittsburgh Sleep Quality Index (PSQI), and one used the Sleep Condition Indicator (SCI). Five studies assessed stress using the Depression, Anxiety and Stress Scale (DASS-21), two used the 10-item Perceived Stress Scale (PSS-10), one used the 22-item Impact of Event Scale–Revised (IES-R), one used the Effort Reward Imbalance (ERI), and one used the 4-item Primary Care PTSD screen (PC-PTSD). Nine studies assessed anxiety using the Generalized Anxiety Disorder 7-item (GAD-7), five used the Depression, Anxiety and Stress Scale (DASS-21), two used the Goldberg depression and anxiety scale (GADS), one used the Effort Reward Imbalance (ERI). Eight studies assessed for depression using the Patient Health Questionnaire-9 (PHQ-9), five used the Depression, Anxiety and Stress Scale (DASS-21), two used the Goldberg Depression and Anxiety Scale (GADS), two used the 2-item Patient Health Questionnaire (PHQ-2), one used the Hospital Anxiety and Depression Scale (HADS), and 1 used the Patient Health Questionnaire-8 (PHQ-8). All studies were evaluated by the NOS quality assessment criteria, two studies received five points, six received four points, five received three points, and six received two points (Table 2, Supplementary Table 2).

A meta-analysis of insomnia or insomnia symptoms among frontline HCWs during the COVID-19 pandemic reported by 16 studies showed a summary prevalence of 42.9% (7,068/15 413 individuals, 95% CI, 33.9–51.9%). There was significant evidence of between study heterogeneity (Q = 2,106.2, P < 0.001, τ² = 0.03, I² = 99.0%). A sensitivity analysis showed that no individual study had an impact on the overall prevalence estimate of more than 3% (Supplementary Table 3). The estimated values were stratified by screening instrument and outcome definition. Summary prevalence estimates ranged from 43.5% of the ISI (6,121/13,893 individuals, 95% CI, 35.4–51.6%, Q = 1,149.3, τ² = 0.02, I² = 99.0%) and 49.3% of the PSQI (914/1,365 individuals, 95% CI, −8.6–107.2%, Q = 185.1, τ² = 0.17, I² = 100.0%; Table 3).

According to study-level characteristics, there were statistically significant differences in prevalence estimates between multi-center studies [6,504/13,655 (49.0%; 95% CI, 37.7–60.4%)] and non-multi-center studies [564/1,758 (32.1%; 95% CI, 23.6–40.7%)] (Q = 0.02, P = 0.02). Studies were further stratified by continent or region. Studies performed in the South America [780/1,059 (73.7%; 95% CI, 70.9–76.4%)] has the highest prevalence, which is significantly different from those performed on other continents (Q = 47.1, P < 0.001). When studies were stratified by study period and individual specialty, prevalence estimates were not statistically significant differences (P > 0.05; Table 7A). By NOS criteria, studies with more thorough descriptive statistics reporting [2,968/5,183; 54.2% (95% CI, 37.5–70.9%)] had higher prevalence estimates than those with less comprehensive descriptive statistics reporting [4,100/10,230; 34.2% (95% CI, 27.0–41.3%); Q = 4.7, P = 0.03]. The estimated prevalence was not significantly different when studies were stratified by sample representativeness, size, comparability between respondents and non-respondents, ascertainment of insomnia, or total NOS score (P > 0.05; Table 8A; details of the NOS appear in Supplementary Table 1).

There were 10 studies on the prevalence of stress or stress symptoms among frontline HCWs during the COVID-19 pandemic. The meta-analysis pooling of the prevalence estimates was 53.0% (3,284/5,976 individuals, 95% CI, 41.1–64.9%), with significant evidence of between study heterogeneity (Q = 897.7, P < 0.001, τ² = 0.04, I² = 99.0%). A sensitivity analysis showed that the impact of individual studies on the estimation of the overall prevalence estimate was <6% (Supplementary Table 4). The estimated values were stratified according to screening instrument and outcome definition, showing that summary prevalence estimates ranged from 78.3% for the PSS-10 (6,121/13,893 individuals, 95% CI, 67.5–89.0%, Q = 5.8, τ² = 0.01, I² = 83.0%) and 41.1% for the DASS-21 (1,453/3,298 individuals, 95% CI, 28.6–53.7%, Q = 234.6, τ² = 0.02, I² = 98.0%; Table 4).

According to the study-level characteristics of the study period, study design, individuals‘ specialty, and continent or region, prevalence estimates were not significantly different (P > 0.05; Table 7B). Prevalence estimates of studies with more sample representative reporting [3,207/5,869; 50.9% (95% CI, 38.3–63.6%)] were lower than those with less sample representative reporting [77/107; 72.0% (95% CI, 63.6–80.4%); Q = 7.4, P = 0.007] when evaluated by NOS criteria. There was a significant difference (Q = 4.3, P = 0.04) between the prevalence estimates of studies with lower valid of ascertainment [261/647 (40.3%; 95% CI, 36.6–44.0%)] and more valid of ascertainment [3,023/5,329 (54.4%; 95% CI, 41.6–67.2%)]. The prevalence estimates were not significantly different when studies were stratified by comparability between respondent and non-respondent, sample size, descriptive statistics reporting, or total NOS score (P > 0.05; Table 8; details of the NOS appear in Supplementary Table 1).

The meta-analysis of anxiety or anxiety symptoms among frontline HCWs during the COVID-19 pandemic reported by the 19 studies showed a summary prevalence of 43.0% (6,509/17,013 individuals, 95% CI, 33.8–52.3%), and there was significant heterogeneity between studies (Q = 3,019.0, P < 0.001, τ² = 0.04, I² = 99.0%). A sensitivity analysis showed that the impact of individual studies on the overall prevalence estimate was <7% (Supplementary Table 5). According to stratification by instrument and outcome definition, summary prevalence estimates ranged from 44.0% for the GAD-7 (2,644/6,195 individuals, 95% CI, 31.6–56.5%, Q = 818.0, τ² = 0.04, I² = 99.0%), 43.2% for the DASS-21 (1,373/3,298 individuals, 95% CI, 34.1–51.5%, Q =98.3, τ² = 0.01, I² = 96.0%), and 46.4% for the GADS (835/1,214 individuals, 95% CI, −12.8–105.6%, Q = 341.3, τ² = 0.18, I² = 100.0%; Table 5).

According to study-level characteristics, the prevalence of studies performed in South America [810/1,059 (76.5%; 95% CI, 73.9–79.1%)] was higher than that on other continents, and the difference was statistically significant (Q = 152.9, P < 0.001). When studies were stratified by study design, period, and individuals specialty, the prevalence estimates of anxiety were not significantly different (P > 0.05; Table 7C). There were no statistically significant differences in prevalence estimates of six-dimensional stratification when evaluated by NOS criteria (P > 0.05; Table 8C; details of the NOS appear in Supplementary Table 1).

A meta-analysis of depression or depressive symptoms among frontline HCWs during the COVID-19 pandemic reported by the 19 studies yielded a summary prevalence of 44.6% (7,452/17,013 individuals, 95% CI, 36.1–53.1%), with significant difference between study heterogeneity (Q = 2,410.8, P < 0.001, τ² = 0.04, I² = 99.0%). Sensitivity analysis suggested that no individual study had an impact on the overall prevalence estimate of more than 5% (Supplementary Table 6). According to stratification by instrument and outcome definition, prevalence estimates ranged from 46.3% for the PHQ-9 (2,896/6,097 individuals, 95% CI, 32.0–60.5%, Q = 975.6, τ² = 0.04, I² = 99.0%), 45.8% for the DASS-21 (1,485/3,298 individuals, 95% CI, 37.0–54.5%, Q =105.5, τ² = 0.01, I² = 96.0%), 55.5% for the GADS (904/1,214 individuals, 95% CI, 5.2–105.7%, Q = 173.9, τ² = 0.13, I² = 99.0%), and 39.8% for the PHQ-2 (1,849/5,461 individuals, 95% CI, 24.1–55.5%, Q = 62.4, τ² = 0.01, I² = 98.0%; Table 6).

The estimated values were stratified according to study-level characteristics. The prevalence estimates between multi-center studies [5,286/10,353 (52.6%; 95% CI, 42.7–62.6%)] and non-multi-center studies [2,141/6,660 (33.3%; 95% CI, 23.7–42.9%); Q = 7.5, P = 0.006] were statistically significance. Studies were further stratified by continent or region. Studies performed in the South America [858/1,059 (81.0%; 95% CI, 78.7–83.4%)] had the highest prevalence, which was significantly different from those performed on other continents (Q = 187.8, P < 0.001). The prevalence estimates were not significantly different when studies were stratified by study period and individual specialty (P > 0.05; Table 7D). When assessed by NOS criteria, studies with more thorough descriptive statistics reporting [3,005/5,183; 56.5% (95% CI, 45.3–67.7%)] had higher prevalence than those with less thorough descriptive statistics reporting [4,422/11,830; 37.7% (95% CI, 28.9–46.5%); Q = 6.7, P = 0.01]. There were no statistically significant differences in prevalence estimates between studies stratified by sample representativeness, sample size, comparability between respondents and non-respondents, ascertainment of depression, or total NOS score (P > 0.05; Table 8; details of the NOS appear in Supplementary Table 1).

Visual examination of the funnel plot reporting on insomnia, stress, anxiety, and depression revealed minimal asymmetry (Supplementary Figures 1–4), suggesting that there was no significant publication bias (P = 0.75; P = 0.69; P = 0.16; P = 0.51 using the Egger test).

The systematic review and meta-analysis of 19 studies involving 17,013 frontline HCWs in 12 countries during the COVID-19 pandemic demonstrated that 42.9% (range, 33.8–52.3%) and 53.0% (range, 41.1–64.9%) of workers reported insomnia and stress, and 43.0% (range, 33.8–52.3%) and 44.6% (range, 36.1–53.1%) screened positive for anxiety and depression, respectively. This data is significantly higher than the prevalence of anxiety (31.9%), depression (33.7%), and stress (29.6%) among general population population during the COVID-19 pandemic (Salari et al., 2020), and is similar to the prevalence of depression (45.0%), anxiety (47%), and insomnia (34%) among patients with COVID-19 (Deng et al., 2021). Compared with the small pandemic of SARS in 2002 and MARS in 2012, COVID-19 has a greater psychological impact on HCWs (Boden et al., 2021; Magnavita et al., 2021). This is closely related to the unknown characteristics of COVID-19, the high risk of exposure to infection, and multiple work pressures. However, Magnavita et al. (2020b) pointed out that the frequency of anxiety and depression among HCWs during the early stage of the COVID-19 pandemic was not higher than the commonly recorded during periodic checks in the years preceding the epidemic, which may be related to the severity of the epidemic in the region or the type of work involving front-line or non-front-line workers. These findings are concerning given that the development of mental health among HCWs has been linked to an increased long-term risk of future mental health diseases.

It is important to note that all participants were assessed through online self-report inventories rather than the gold-standard diagnostic clinical interviews. The sensitivity and specificity of these instruments for estimating symptoms of insomnia, stress, anxiety, and depression vary substantially (Supplementary Table 7). In the evaluation of insomnia and stress, the sensitivity of the ISI (Morin et al., 2011; 99%, 95% CI, 97–100%) and IES-R (Lee, 2012; 86%, 95% CI, 67–94%) was higher than that of the others. However, that of ESS (Siegrist et al., 2014; 76%, 95% CI, 63–86%, 20%) and PC-PTSD (Li et al., 2019; 57%, 95% CI, 45–68%) are the least. When estimating anxiety symptoms, the specificity of the HADS-A (Hitchon et al., 2020) was 78% (95% CI, 69–85%), which was lower than that of the others. In contrast, the PHQ-9 (Williams et al., 2002) has high sensitivity (88%, 95% CI, 74–96%) and specificity (88%, 95% CI, 85–90%) for diagnosing major depression and has been proven to be comparable with the administered assessments by clinicians, whereas the GHQ (Williams et al., 2002) has low specificity (66%, 95% CI, 57–74%). In addition, all self-report measures were conducted online because of the COVID-19 pandemic which may affect the accurate evaluation of symptoms.

The sub-analysis revealed potentially important differences, such as study period, specialty and region. The prevalence of stress and depression is higher in 2020 and that of insomnia and anxiety is higher in 2021. This probably reflects the already established psychological problems of the population that have changed with the evolution of COVID-19 (Simon et al., 2020; Kola et al., 2021). HCWs changed from stress and depression during the outbreak of the COVID-19 pandemic to mild psychological problems of sleep and anxiety with the normalization of COVID-19 prevention and control. In the HCWs specialty, the prevalence of psychological problems of physicians is higher than that of other HCWs. This is related to the working hours and strength of the physicians at the frontline of COVID-19 (Elbay et al., 2020). By region, the prevalence of HCWs in South America is higher, while the prevalence of anxiety and depression in North America is lower, which is related to local economic level, contact tracing, isolation and management, and other measures (Fitzpatrick et al., 2020; Goularte et al., 2021).

Furthermore, according to NOS criteria analysis, variation in sample size contributed significantly to the observed heterogeneity in the study. Studies with fewer participants usually produced more extreme prevalence estimates of stress, indicating publication bias. However, for insomnia, anxiety, and depression, studies with more participants usually produced more extreme prevalence estimates. These differences were partly captured by the NOS score, which assessed the risk of bias in each study. Studies with a higher risk of bias produced higher prevalence estimates of insomnia and stress, while studies with a lower risk of bias produced higher prevalence estimates of anxiety and depression. These findings may be related to the heterogeneity between studies on study design (i.e., multicenter vs. non multicenter), online surveys, study period of COVID-19, positions of frontline HCWs, and severity of regional pandemic.

Most opinions show that psychological strategies are the mediating factor to change these outstanding problems, although the stressors (COVID-19 pandemic) cannot be changed (Mediavilla et al., 2022). Importantly, increasing the so-called social support barriers (i.e., factors that increase the use of social support, even when available) might also contribute to improving the mental health of HCWs (Thoresen et al., 2014). With mounting evidence suggesting an association between reported discrimination against COVID-19 and poor mental health outcomes among HCWs, mental health strategies at the community level could take the form of anti-stigma campaigns (Taylor et al., 2020). At the policy level, addressing various common problems reported by HCWs, including increased workload, shortage of protective equipment, or lack of standardized operating procedures, may also enable HCWs to help reduce the negative consequences of mental health problems (Erquicia et al., 2020).