The psychological impact of COVID-19 on Chinese healthcare workers: a systematic review and meta-analysis

Relevant records that were published until June 10, 2020 were searched in the databases of Medline, PsycINFO, EMBASE, the Cochrane Library (including Cochrane Database of Systematic Reviews), and main Chinese databases including Sinomed, the China National Knowledge Infrastructure (CNKI), and WanFang data. Preprint articles published on Medrxiv and SSRN servers, as well as the Google Scholar, and the daily updated WHO COVID-19 database were also included. The search strategy for Medline was provided in the supplementary materials. The language was restricted to English, Chinese or German. Furthermore, the reference lists from reviewed articles were searched to identify and retrieve relevant articles.

We included studies that evaluated the psychological impact of COVID-19 on HCWs in China, which ought to include at least one of the following five target dimensions: PTSS, anxiety, depression, sleep disturbance or profession-related burnout. To be included into the meta-analysis, studies should use measurements that were proved to be valid to measure at least one of our target dimensions. Therefore, studies were excluded if they only measured general distress using tools of the General Health Questionnaire (GHQ-12) [6], or used non-validated self-designed questionnaire [7] or one single question of “what has been your mental attitude since COVID-19 outbreak” [8].

In order to ensure the study quality, we only included Chinese articles that were published in journals which are incorporated in the Chinese Science Citation Database (CSCD).

Both cross-sectional studies and interventional studies aligning with our criteria would be included, provided that, for the latter, the baseline level of psychological impact was extractable. Surveys investigating both HCW and other populations were included only if the data on HCW could be extracted separately.

Studies with neither data of our interest nor extractable data were excluded. When papers contained post-hoc analyses of an already included study, data was combined into one data set [9‐12].

The methodological quality of the included studies was assessed using an 11-item checklist recommended by Agency for Healthcare Research and Quality (AHRQ) [13]. An item was scored ‘0’ if it was answered ‘NO’ or ‘UNCLEAR’; if it was answered ‘YES’, then the item scored ‘1’. Article quality was assessed as follows: low quality = 0–3; moderate quality = 4–7; high quality = 8–11.

Primary outcomes include the prevalence of moderate to severe psychological impact, i.e. its five dimensions, PTSS, anxiety, depression, sleep disturbances, and profession-related burnout. The secondary outcomes include the prevalence of mild to severe psychological impact, and the severity of the psychological impact which were reflected through continuous variables.

The severities of each dimension were defined according to the validated cut-off values of each measure.

Data extraction was then independently carried out by three authors (NNX, YP and JL). To compute the prevalence rate, both the number of confirmed cases and the total amount of participants was extracted. For primary outcomes of the prevalence of moderate to severe cases, only participants who indicated symptoms above the cut-off for moderate symptoms were classified as burdened. Participants with no or mild symptoms were classified as not burdened. Continuous outcomes were analysed using the number of participants, the mean, and the standard deviation (STD). Missing STD values were calculated from reported confidence intervals (CI), standardized errors (SE), or p values. When none of the above data were reported, study authors were contacted via e-mail for further information.

As some studies report the prevalence rates of mild, moderate, severe symptoms plus the mean and STD of the same scales, these studies were included in the pooled analyses for both primary and secondary outcomes.

Statistical heterogeneity was tested using the I² statistic and the natural approximate chi-square test [14]. I² values above 50% indicated high heterogeneity. High heterogeneity was indicated by p values smaller than 0.10. The random effects model was used for heterogeneous data. The rate of prevalence was pooled by the inverse variance method. For continuous data, because of the variance in measurements across studies, the reported values were first transformed into the standardized values with a range from 0 to 100, according to the possible ranges of each questionnaire. Then the standardized values were pooled and compared between different subgroups. The likelihood of significant publication bias was assessed by both Begg’s test and Egger’s test. In addition, funnel plots were provided as a visual tool for publication bias. The Stata (15.1) [15] and the metan package [14] was used for statistical analyses.

Subgroup analyses were planned a priori to investigate potential moderators influencing the psychological distress of HCW, and thereby, to assess the sources of high heterogeneity. Therefore, both primary and secondary outcomes were compared by frontline HCW (yes/ no), gender, occupation and work location. Subgroup analysis was performed when there were at least two comparisons included. Further meta-analysis was performed in different subgroups. Meta-regressions were not performed here due to the partially small number of studies.

Sensitivity analyses were conducted to test the reliability of primary outcomes, which were performed by excluding studies with low quality both individually and altogether.

Overall, 85 full text articles were obtained and assessed for eligibility (Fig. 1). Of these, 44 studies met the inclusion criteria for the review, and were included in the meta-analysis [10, 11, 16‐57].

44 studies with a total of 65,706 participants were included (see Table 1), with 76.7% being women. Apart from the studies that did not report the specific occupation of HCW, a total of 19,316 (33.0%) doctors, 35,644 (60.9%) nurses and 3552 (6.1%) technicians and administrative staff were investigated. As shown in Table 1, most studies were conducted from early to mid-February, at the height of the COVID-19 epidemic in China. Regarding the locations of the studies, 12 were conducted in Wuhan city, 3 in Hubei province, 16 in other provinces in China, and another 13 at multiple centres nationwide or in unknown areas. Twenty-four of the included studies were published in English, whereas the remaining 20 studies were published in Chinese. No articles in German were identified.

All selected articles were assessed for methodological quality. According to the criteria of AHRQ, only two studies were of high quality, 26 studies were of moderate quality, and 16 studies were of low quality (see Table 1 and detailed information in Supplementary Table 1).

All assessment tools and the according cut-off values employed to measure the psychological impact are listed in Table 2.

The prevalence of mild to severe PTSS was not pooled due to the lack of data. The prevalence of mild to severe levels of anxiety, depression and sleep disturbances was estimated as 31% (95% CI 25–37%), 37% (95% CI 32–42%) and 39% (95% CI 25–52%), respectively (Supplementary Figs. 10–12).

Reflected by continuous data, the psychological impact was pooled and compared between different subgroups. Take the pooled ES of both frontline and non-frontline HCW for an example, on a standardized range between 0 and 100, all dimensions of psychological impact, including PTSS [SMD = 37.0 (95% CI 28.2–45.8)], anxiety (SMD = 43.7 (95% CI 39.6–47.8)], depression [SMD = 37.4 (95% CI 29.4–45.5)], and sleep disturbances [SMD = 36.2 (95% CI 25.1–47.3)] fall into a mild to moderate range.

By excluding all studies with low methodological quality, the prevalence of moderate to severe PTSS, anxiety, depression, and sleep disturbance was slightly higher, which was estimated as 32% (95% CI 26–39%), 21% (95% CI 15–27%), 16% (95% CI 14–18%), and 21% (95% CI 8–34%). The heterogeneity within each outcome remained substantial, with the I² lowered by 0–1.6%.

Results from Egger’s test showed potential publication bias concerning the prevalence of moderate to severe PTSS (p = 0.009), anxiety (p = 0.003), and depression (p < 0.001), but the Begg’s test did not reveal risk of publication bias in the prevalence of PTSS (p = 0.23). Neither test indicated potential risk in the prevalence of sleep disturbances (p = 0.254 and 0.221), or the severity of PTSS (p = 0.640 and 0.436), anxiety (p = 0.079 and 0.315), depression (p = 0.616 and 0.767) or sleep disturbances (p = 0.366 and 0.734) as reflected in continuous data. Funnel plots were provided in the Supplementary Figs. 29–36.

Forty-four studies with a total of 65,706 HCW were included. The period of studies spanned the end of January to early April 2020, at the height of the COVID-19 epidemic in China. Despite the great heterogeneity of the studies, there is strong evidence that about a third of the clinic staff showed at least one dimension of psychological symptoms. More pronounced symptoms like moderate and severe level of PTSS, anxiety, depression, and sleep disturbances, were found in 27%, 17%, 15%, and 15% of the participants examined, respectively.

Compared with results of the latest China mental health survey [61], the psychological burden was significantly elevated. However, the increased values differed little from the values reported for the general population in China in the above-mentioned months [4, 62, 63].

Compared to studies from other countries: a multi-centered study in Singapore and India investigated frontline HCW. In this study, only 2.2% were screened positive for moderate to extremely-severe stress, 8.7% for anxiety, and 5.3% for depression. However, the prevalence of physical discomforts was as high as 33.4% [64]. It was speculated that somatic symptoms were used to represent emotions in this situation. A study conducted during an early peak of COVID-19 in New York City also found very high positive screens for psychological symptoms as follows: 57% for acute stress, 48% for depressive symptoms, and 33% for anxiety symptoms [65].

In addition, after the data of our meta-analysis were collected, similar studies were piled up and have provided further evidence in professional-related burnout and post-traumatic growth. For example, a study showed that the burnout thresholds in disengagement and exhaustion were met by 79.7% and 75.3% of respondents [66]. Another large-scale survey of frontline nurses in China reported that 13.3% of them experienced trauma, and 39.3% experienced post-traumatic growth [67]. In a tertiary hospital of a highly burdened area of north-east Italy, 38.3% HCW showed high emotional exhaustion and 46.5% showed low professional efficacy [68].

Similar to our results considering the prevalence of psychological burden, previous evidence from China, Germany, and worldwide has shown that mental problems were more pronounced in female participants, nurses, and frontline health professionals, especially those in the departments for infectious diseases, fever clinics and intensive care units, which was in line with their exposure level and proximity to COVID-19 patients [3, 10, 58].

However, contrary to our expectations, the subgroup comparison results differed between the binary and the continuous outcomes, i.e. reflected by continuous data, basically all subgroups were comparable, and that HCW in the Hubei province even had higher levels of anxiety and depression than those in Wuhan city. One possible explanation was that since most HCW did not have severe symptoms, so that the effect sizes from higher percentages of confirmed cases were diluted. Such results also remind us to not neglect the mental health of “non-frontline” healthcare providers. In addition, during that period, it is possible that Wuhan got most of the assistance at the very beginning, and healthcare providers in other cities of the Hubei province, like Huanggang city, were suffering under more severe anxiety and depression [69].

Similar to the COVID-19, several viral epidemics have occurred in the past 20 years, such as the SARS, the A/H1N1 influenza pandemic, the Middle East respiratory syndrome (MERS), and the Ebola virus disease. According to a review, in these large-scale viral outbreaks, HCW at high risks of exposure also had greater levels of both acute or posttraumatic stress (OR 1.71) and psychological distress (OR 1.74) [2]. Another similar review found that 11–73.4% of HCW reported PTSS during outbreaks, whereas depressive symptoms were reported in 27.5–50.7%, insomnia symptoms in 34–36.1%, and severe anxiety symptoms in 45% of HCW [70].

Compared with these results, our findings on the prevalence of psychological symptoms, especially the moderate and severe cases, in HCW from China under the COVID-19 were at the lower end. A possible explanation for the lower psychological distress of HCW in China could be the relatively low mortality rate, the quick control of the epidemic in China and available experiences acquired from previous pandemics, like the SARS in 2003.

In response to the crisis, as early as Jan 27, 2020, the National Health Commission of China published a national guideline of psychological crisis intervention for COVID-19 to provide multifaceted psychological protection of the mental health of medical workers [71]. Mental health experts across the country responded quickly to form psychosocial crisis intervention teams, and offered both online and face-to-face psychological counseling, hotline services, and online platforms with psychological self-help information, such as mindfulness and relaxation techniques [69]. Based on previous experiences, they were also expected to look after the needs of teams that were newly formed in the course of the Corona-related restructuring or that have come into conflict situations due to stress overload.

However, the implementation of psychological intervention services encountered obstacles, as Chen et al. pointed out [72]. Even though medical staff showed signs of psychological distress, they denied problems and refused psychological help. Therefore, interventions were adjusted to focus more on fulfilling their basic needs, such as providing more places to rest, guaranteeing food and daily living supplies [72]. According to existing evidence, prevention efforts such as screening for mental health problems should be provided in a proper way [73].

In some studies, psychological support demonstrated protective effects. For example, the Balint group, which was developed by Michael and Enid Balint, is a small group of clinicians who meet regularly to discuss cases from their practices, with a focus on the doctor–patient relationships. In Iran, the Balint groups were found to help healthcare workers to better cope with psychosocial stressors by improving participants' insight into their experience and by facilitating group learning on the doctor-patient relationships [74].

Our review has several limitations. First, we only focused on studies conducted in China. Future reviews with studies from other countries that are affected by the virus at a later time period may show different exposure profiles. In particular, different health care systems will have an influence on the severity of mental stress and coping strategies. Second, the heterogeneity of the studies was high, perhaps due to the different assessment scales and its respective cut-off scores. We have tried to compensate for this heterogeneity by differentiating the moderate and severe cases from other subclinical symptoms, as well as by examining the severity reflected through continuous data. Subgroup analyses were also carried out and revealed that the frontline work, gender, occupation, and location could partly explain the high heterogeneity among the different studies. Third, even though we tried to include more high-quality studies by setting criteria for the journals in which they were published, the methodological quality of most studies was assessed as being moderate, and even low. For one, given the special background, most of the studies were carried out online so that no response rate could be provided, and the representative of the sample could not be guaranteed. Moreover, risk factors were seldom inquired, such as previous mental disorders or stressful live events, which could be the confounding factors of the current psychological distress. In addition, we did not set a criterion for the minimum time period when assessing the study quality, even though the AHRQ checklist was used to assess whether the study period was reported. It is also one of our limitations that we did not look up the subsequent studies that have referenced the studies included. Last, our analyses showed that a publication bias was likely, pertaining to the prevalence of PTSS, anxiety, and depression.

Future high-quality research remains necessary to explore the impact of COVID-19 on profession-related burnout of HCW, its long-term impact, and post-stress or post-traumatic growth, i.e., the positive psychological change experienced after the struggle with COVID-19, such as to identify meaning in interpersonal relationships, to change priorities, and to have a richer spiritual life.

To improve the methodological quality, researchers should pay further attention when designing the study, and should especially consider how to improve the representativeness of the sample, and the validity and reliability of assessment tools, and how to assess and control for potential confounding factors more sufficiently.