We systematically searched published articles on the Web of Science, Embase, PubMed, Cochrane Library, Scopus, ScienceDirect, SinoMed, China National Knowledge Infrastructure (CNKI), and Wanfang Database without language restrictions from January 1, 2020, to October 20, 2022. Based on the Medical Subject Headings, we used subject terms and free-text terms for the literature search. The subject terms used were mindfulness, COVID-19, health care worker, anxiety, depression, and stress. The search strategies for each database are shown in Multimedia Appendix 1. Furthermore, to prevent the omission of relevant literature, we screened the references of the included studies to find clinical studies that met the inclusion criteria.

The relevant studies were independently searched by 2 authors (QZ and HH). If there were any disputes, the help of a third author (J-mY) was sought. Based on the PICOS (population, intervention, comparison, outcomes, and study design) principles [27], we established the following inclusion criteria for the study:

Studies that did not meet the above requirements were excluded.

Two authors (HY and Y Long) independently extracted the data from the included studies into an Excel (Microsoft Corp) spreadsheet and then summarized the data in a table. If there were any inconsistencies, a third author (J-mY) was involved, and a unified opinion was formed.

The data extracted included (1) first author and year of publication; (2) country of participants; (3) occupational roles of participants; (4) groups and sample size, sex, and age of participants; (5) outcome measurement time points; (6) duration of intervention; (7) main outcomes; and (8) outcome assessment scale.

Two authors (HY and Y Long) independently completed the methodological quality assessment of the included studies using the Physiotherapy Evidence Database (PEDro) scale. If they did not agree to the scoring of the same study, a session was held to resolve the issue. If the problem remained unresolved, a third author (J-mY) reached a consensus. There were eleven items on the PEDro scale: (1) eligibility criteria and source, (2) random allocation, (3) concealed allocation, (4) baseline comparability, (5) participant blinding, (6) therapist blinding, (7) assessor blinding, (8) adequate follow-up (>85%), (9) intention-to-treat analysis, (10) between-group statistical comparisons, and (11) point and variability measurements. It is important to note that the scores on the PEDro scale are cumulative from items 2 to 11, excluding item 1, with a minimum score of 0 and a maximum score of 10. Based on the scores of each study, we classified study quality into four levels: (1) poor, scores less than 4; (2) fair, scores between 4 and 5; (3) good, scores between 6 and 8; and (4) excellent, scores between 9 and 10 [28].

Two authors (Y-bZ and Y Luo) independently assessed the risk of bias in the included studies using the Cochrane risk of bias tool [29]. Any disagreements were resolved at the end of the discussion. There are seven items in the bias risk table: (1) random sequence generation (selection bias), (2) allocation concealment (selection bias), (3) blinding of participants and personnel (performance bias), (4) blinding of outcome assessment (detection bias), (5) incomplete outcome data (attrition bias), (6) selective reporting (reporting bias), and (7) other bias. Each item was classified as “low risk,” “high risk,” or “unclear risk.” Furthermore, if a sufficient number (n≥7) of studies were included, a funnel plot analysis was performed to assess publication bias.

Based on the GRADE (Grading of Recommendations Assessment Development and Evaluation) criteria [30], 2 authors (J-mY and HY) assessed the level of evidence for the main outcome. The criteria are applicable to randomized trials and observational trials. For randomized trials, the initial quality grade for the outcome was high. The level of evidence was then appropriately downgraded based on the assessment of 5 dimensions, namely risk of bias, inconsistency, imprecision, indirectness, and publication bias. According to the results of the assessment, the level of evidence was categorized into 4 grades: “high,” “moderate,” “low,” and “very low.”

Review Manager (version 5.3; The Cochrane Collaboration) was used to perform the statistical analysis. In this study, we extracted the postintervention data for statistical analysis because the baseline differences between studies were not significant. For continuous data, we entered the pre-extracted mean, SD, and sample size into statistical software. For the same outcome, if the method of assessing the outcome was the same across studies, the mean difference was used to estimate the effect size and the 95% CIs to indicate the CI range; if not, the standard mean difference (SMD) was used. In all analyses, the magnitude of heterogeneity was analyzed according to the I² statistic, with I²<50% indicating low heterogeneity and I²≥50% indicating high heterogeneity. If the P value of the heterogeneity test was less than .05, the random-effects model was used; if the P value was greater than or equal to .05, the fixed-effects model was used [31]. In addition, subgroup analysis was used to look for sources of heterogeneity and to explore whether the results were the same for subgroups under different conditions; sensitivity analysis was performed to verify the stability of the results.

After a systematic search of 8 databases, we retrieved 449 articles, including 13 in PubMed, 46 in Web of Science, 3 in Embase, 124 in the Cochrane Library, 19 in Scopus, 206 in ScienceDirect, 19 in SinoMed, 13 in CNKI, and 6 in the Wanfang Database. In addition, 3 papers came from the citation search. First of all, after eliminating duplicate articles by EndNote software (Thomson Corp), 387 articles remained. Next, we read the titles and abstracts, and 359 articles were removed, with 28 remaining. We then read the full text, and 18 articles were excluded, leaving 10 articles. All literature selection processes were done independently by 2 authors using EndNote software. The list of excluded studies and the reasons for their exclusion are shown in Multimedia Appendix 2. As a result, 10 randomized controlled studies were ultimately included in the study [32-41] (Figure 1).

Of the 10 articles, 5 were in English [32-36] and 5 were in Chinese [37-41]. For convenience, all the Chinese references have been translated into English. Three studies recruited only nurses [34,37,41], and the remaining studies recruited 2 or more occupations. The total duration of the intervention was 8 weeks in 3 studies [36,38,39], 4 weeks in 3 studies [37,40,41], 16 days in 1 study [35], and 2 weeks in 1 study [32]. The duration of the intervention was the number of sessions in 2 studies: 7 sessions in 1 study [33] and only 1 session in another study [34]. Three studies assessed anxiety and depression used the SAS and SDS, respectively [37,40,41]; 3 studies assessed anxiety and depression used the 7-item Generalized Anxiety Disorder Scale and Patient Health Questionnaire-9, respectively [35,38,39]; 1 study assessed depression, anxiety and stress used the Depression, Anxiety, and Stress Scale-21 [32]; 1 study used the Hospital Anxiety and Depression Scale (HADS) to assess depression, and the HADS and the Short Health Anxiety Inventory (SHAI) to assess anxiety, respectively [33]; 1 study assessed depression and anxiety used the Goldberg Depression Scale and Goldberg Anxiety Scale, respectively [36]; 1 study assessed stress used the State Anxiety Inventory-I [34]; and 1 study assessed stress used the Perceived Stress Scale [35] (Table 1).

The financial support and disclosures of interest in the included studies are shown below. Six studies indicated that they were funded [32,33,35,36,39,40], and 5 claimed no conflict of interest between the authors [32-36]. The remaining studies did not mention financial support or conflicts of interest.

One study asked participants to use PsyCovidApp (a self-managed and self-guided psychoeducational mobile-based intervention with no therapist support) for a self-management psychological intervention based on cognitive behavioral therapy and mindfulness therapy [32]. One study divided the participants in the experimental group into 21 WhatsApp groups for 7 intervention sessions. The content of the 7 sessions was selected according to standard cognitive-behavioral and mindfulness techniques. The duration of each session varied from 45 to 90 minutes of communication [33]. One study divided the intervention group into 9 subgroups, each of which underwent a single session of mindfulness-based breathing and music therapy for approximately 30 minutes [34]. One study asked participants to practice 15 minutes of brief mindfulness meditation daily for 16 days [35]. The experimental group in 3 studies used MBSR training [36,39,40], 2 studies used MBSR combined with general psychological intervention [37,41], and 1 study used mindfulness meditation combined with positive psychological intervention [38]. Details of the interventions for the experimental and control groups of each study are shown in Multimedia Appendix 3.

The PEDro scores and score details of each study are presented in Table 2. The PEDro scores for the 10 studies were as follows: 1 study scored 4 (quality grade: fair) [35], 1 study scored 5 (quality grade: fair) [36], 6 studies scored 6 (quality grade: good) [34,37-41], 1 study scored 7 (quality grade: good) [33], and 1 study scored 8 (quality grade: good) [32]. All studies had a random allocation, baseline comparisons, between-group comparisons, and point and variability measures. Two studies referred to the random assignment of participants by individuals who did not participate in the study (concealed allocation) [32,33]. Blinding was not mentioned in any of the remaining studies, except for one study that used double-blinding [32] and 3 studies that used single-blinding [33,34,36]. Two studies did not have adequate follow-up numbers (<85%) [35,36]. All participants in 5 studies completed the experiments according to the established study protocol [37-41].

Figure 2 [32-41] shows the risk of bias results of the 10 studies based on the Cochrane risk of bias tool. Three studies did not mention the specific method of random sequence generation (high risk of bias) [35,37,39]; 4 studies blinded the assessors [32-34,36] and only 1 study stated that it blinded the participants (low risk of bias) [32]; 2 studies used allocation concealment (low risk of bias) [32,33]; and outcome data of 2 studies were incomplete (high risk of bias) [35,36]. No studies had selection bias or other bias (low risk of bias). As a result, the overall bias was low for 1 study [32], high for 4 studies [35-37,39], and unclear for other studies.

Of the 10 studies, 9 investigated the effect of web-based mindfulness-based on anxiety and depression [32,33,35-41], and 3 measured the effect on stress [32,34,35]. Among them, the means and SDs of data from 2 studies could not be extracted [38,39], so for anxiety and depression, only data from 7 studies were pooled. The data were meta-analyzed using the SMD because the assessment scales used varied across studies. For anxiety and depression, we used a random-effects model to estimate effect values because the P value for heterogeneity was less than .05; for stress, we used a fixed-effects model because the P value was greater than .05.

The results of forest plots showed that the web-based mindfulness-based interventions significantly reduced anxiety (SMD=–0.63, 95% CI –0.96 to –0.31, P<.001, I²=87%), depression (SMD=–0.52, 95% CI –0.77 to –0.26, P<.001, I²=75%), and stress (SMD=–0.20, 95% CI –0.35 to –0.05, P=.01, I²=58%) among frontline health care workers during the COVID-19 pandemic (Figure 3). Furthermore, although we did not combine data from the studies of Zou et al [38] and Huang et al [39], their results both suggest that web-based mindfulness-based interventions reduced anxiety and depression in health care workers.

First, we performed subgroup analyses for anxiety and depression according to the duration of the intervention. The duration of intervention for the 2 subgroups was 7 to 16 days and 4 to 8 weeks, respectively. For anxiety, both subgroups showed that the web-based mindfulness-based interventions significantly reduced the anxiety levels of health care workers (P<.05). For depression, mindfulness-based interventions lasting 4 to 8 weeks significantly reduced depression levels (SMD=–0.73, 95% CI –0.91 to –0.54, P<.001, I²=6%), but interventions lasting 7 to 16 days did not (SMD=–0.08, 95% CI –0.23 to 0.08, P=.33, I²=0%). It is important to note that for depression, the heterogeneity was 0% and 6% for the 2 subgroups, suggesting that the heterogeneity may come from the duration of intervention. (Multimedia Appendix 4)

Second, we performed subgroup analyses for anxiety and depression according to the types of scales. The 2 subgroups were SAS or SDS and other scales, respectively. The results showed that both subgroups reduced the anxiety and depression levels of health care workers (P<.05). It is worth noting that for anxiety, the heterogeneity was 0% for both subgroups, suggesting that the heterogeneity perhaps mainly comes from the types of scales. (Multimedia Appendix 5)

By removing the included studies one by one, we performed sensitivity analyses for anxiety, depression, and stress. The results showed that the effects of web-based mindfulness-based interventions on anxiety and depression were very stable, with P values consistently less than .05. However, for stress, the result became meaningless (P>.05) when the study by Fiol-DeRoque et al [32] or Yıldırım et al [34] was removed.

By generating funnel plots, we assessed the publication bias for anxiety and depression. For anxiety, the funnel plot was asymmetrically distributed on both sides of the null line, suggesting a possible publication bias. For depression, the funnel plot was symmetrically distributed, suggesting a low likelihood of publication bias. For stress, the funnel plot analysis was not performed due to the limited number of studies. (Figure 4)

The assessments of anxiety, depression, and stress are shown in Table 3. For anxiety, the funnel plot was asymmetric on both sides of the null line, so the level of evidence was downgraded by one level due to publication bias. For stress, there was a high risk of selection bias and attrition bias for 1 study, so the level of evidence was downgraded by one level for a serious risk of bias. For anxiety, depression, and stress, the heterogeneity was relatively high (I²>50%) and the 95% CIs were wide, so the level of evidence was reduced by 2 levels due to serious inconsistency and imprecision. Consequently, the level of evidence for depression was low, and for anxiety and stress, it was very low. That said, we have low levels of evidence that the web-based mindfulness-based intervention reduced depressive symptoms in frontline health care workers during the COVID-19 pandemic, yet only very low levels of evidence that it reduced anxiety and stress.

This study aimed to investigate the impact of web-based mindfulness-based therapy on the symptoms of common mental disorders among frontline health care workers during the COVID-19 pandemic. After a comprehensive search, 10 randomized controlled studies that met the inclusion criteria were included. We found that web-based mindfulness-based intervention may promote a reduction in the symptoms of anxiety, depression, and stress among frontline health care workers. This suggests that even during the COVID-19 pandemic, in an environment of fear and depression, web-based mindfulness-based strategies may provide some reduction in the symptoms of anxiety, depression, and stress in some circumstances.

The latest data released by the World Health Organization show that as of June 14, 2023, as many as 767.9 million patients have been diagnosed with COVID-19 worldwide, with 6.94 million cumulative deaths and 286,078 new cases in the last 7 days [42]. The global form of the COVID-19 pandemic remains severe. Health care workers on the front lines of the pandemic are under immense psychological stress and are subject to many common mental disorders, such as post-traumatic stress disorder, anxiety, and depression [43-45]. Moreover, this phenomenon is more evident in nurses than in other medical staff members [5,6]. Health care workers are the most important bulwark in the fight against COVID-19, and they have worked hard to prevent the spread of the pandemic and save the lives of patients. Undoubtedly, their physical and mental health is an important guarantee of victory in the fight against the pandemic. Therefore, while fighting the COVID-19 pandemic, it is crucial to find appropriate intervention strategies to reduce the symptoms of common mental disorders among health care workers.

Mindfulness-based therapy is a psychological intervention that addresses stress, pain, and illness through mindfulness meditation. During the mindfulness-based practice, the participant’s attention is primarily focused on the breath, emotions, thoughts, and body sensations. The practice of mindfulness is generally initiated by focusing on the breath. During the breathing process, an effort is made to feel the state of being in that moment and to focus attention on that state. Focusing on the breath is the foundation of mindfulness-based practice, which promotes self-control, increases personal awareness, and reduces the effects of stress factors [34]. The mechanism by which mindfulness-based interventions reduce the symptoms of common mental disorders in individuals may lie in enhanced signals from brain regions (eg, frontal cortex and hippocampus) involved in emotion regulation in patients who receive mindfulness-based training [46,47]. Mindfulness-based intervention is acceptable and suitable for different age groups, such as older individuals [48], young adults [49,50], and children [51]. It is also suitable for different social groups, such as students [52], pregnant women [53], patients with cancer [54], and health care workers [55,56]. Given the recommendations about maintaining social distance and the overloaded work schedule of health care workers during the COVID-19 pandemic, attending face-to-face psychological counseling is very inconvenient for them. Therefore, tele-counseling is a better option in this situation. Compared to face-to-face interventions, web-based interventions can reduce the risk of COVID-19 infection for health care workers, saving them valuable time so that they can better focus on patient assistance. Although previous studies have shown the effectiveness of web-based mindfulness-based interventions in managing mental disorders among health care workers [14,21,22], there was no global outbreak of COVID-19 at the time, and the daily stress faced by health care workers was likely to be relatively low. In the context of the COVID-19 pandemic, can web-based mindfulness-based interventions still play a beneficial role for health care professionals? Our findings suggest that web-based mindfulness-based interventions could continue to have positive effects on the management of common mental disorders among health care workers, even during the COVID-19 pandemic. This may help reduce the symptoms of mental disorders among frontline health care workers in the fight against such outbreaks.

It is important to note that the pooled heterogeneity of anxiety, depression, and stress is high. Using subgroup analysis, we determined that this heterogeneity stems mainly from the duration of intervention and the types of scales. For example, the study by Ghazanfarpour et al [33] used 2 scales (HADS and SHAI) to assess participants’ anxiety; however, the results of the HADS showed that web-based mindfulness-based interventions reduced anxiety levels in health care workers, but the SHAI did not. In addition, Santamaría-Peláez et al [36] showed that when the web-based mindfulness-based interventions lasted up to 4 weeks, there were no significant differences in anxiety and depression levels between the experimental and control groups; however, when the interventions were conducted for up to 8 weeks, the differences between the groups became significant. In addition, we noted that the meta-analyses had wide CIs, indicating poor precision. Therefore, the results should be interpreted with caution.

Furthermore, we noted that the study by Santamaría-Peláez et al [36] had a 3-month follow-up for participants, and the results of the follow-up were the same as those at the end of the intervention. For example, at 4 weeks, the difference between groups was not significant, nor was the difference at follow-up; at 8 weeks, the difference between groups was significant, as was the difference at follow-up. This suggests that the web-based mindfulness-based interventions may have a long-term effect on participants’ mental regulation. However, there are very few follow-up studies about the effects of web-based mindfulness-based interventions on health care workers’ mental disorders, suggesting that future researchers should focus on the follow-up effects of mindfulness-based interventions.

The strength of this study is that it is the first systematic review and meta-analysis of the impact of web-based mindfulness-based interventions on anxiety, depression, and stress among frontline health care workers during the COVID-19 pandemic. We focused not only on the immediate effects after the web-based mindfulness-based intervention but also on the follow-up effects after the intervention to explore the long-term effects of the web-based mindfulness-based intervention.

This study has several limitations. First, the use of different scales to assess anxiety, depression, and stress in different countries is a major source of heterogeneity. Due to the high heterogeneity and wide CI of the results in the meta-analyses, we should be cautious about the interpretation of the results. Second, possibly owing to the limitations of the pandemic, some studies had short intervention durations, ranging from a single session to 8 sessions, which leaves a lack of evidence for the long-term effects of web-based mindfulness-based interventions on their efficacy in reducing the symptoms of common mental disorders among health care workers. Third, we analyzed both postintervention and follow-up data, which may bring about some bias.

Given the current trend of the pandemic, regular epidemic prevention and control are the main focus areas. This also suggests that the majority of health care workers need to be prepared to fight the pandemic for a long time. Therefore, safeguarding the mental health of health care workers has long been an exceptionally important matter. It is valuable to develop an easy-to-use app for health care workers to relieve stress and experience relaxation at work. In addition, researchers should make a transition from the short-term to the long-term effects of web-based mindfulness-based therapy on reducing symptoms of common mental disorders in frontline health care workers, with appropriate attention to follow-up effects after the end of the intervention.

This study systematically reviews the effectiveness of web-based mindfulness-based interventions in the management of common mental disorders among frontline health care workers in the fight against the COVID-19 pandemic. The results showed that web-based mindfulness-based psychotherapy may be effective in reducing the symptoms of anxiety, depression, and stress among health care workers, even during the COVID-19 pandemic, but benefits are likely to be mild on average, and more research is needed before a clear recommendation can be made. In the future, randomized controlled studies with high quality and long intervention times are required to further justify this conclusion.