Mental health symptoms during the COVID-19 pandemic in developing countries: A systematic review and meta-analysis

Background This systematic review aims to 1) summarize the prevalence of anxiety, depression, distress, insomnia, and PTSD in the adult population during the first year of the COVID pandemic in developing countries and 2) uncover and highlight the uneven distribution of research on mental health in all developing countries across regions. Methods Several literature databases were systemically searched for meta-analyses published by September 22, 2021, on the prevalence rates of mental health symptoms in developing countries worldwide. We meta-analysed the raw data of the individual empirical results from the previous meta-analysis papers in developing countries in different regions. Results The prevalence rates of mental health symptoms were summarized based on 341 empirical studies with a total of 1 704 072 participants from 40 out of 167 developing countries in Africa, Asia (East, Southeast, South, and West), Europe, and Latin America. Comparatively, Africa (39%) and West Asia (35%) had the worse overall mental health symptoms, followed by Latin America (32%). The prevalence rates of overall mental health symptoms of medical students (38%), general adult students (30%), and frontline health care workers (HCWs) (27%) were higher than those of general HCWs (25%) and general populations (23%). Among five mental health symptoms, distress (29%) and depression (27%) were the most prevalent. Interestingly, people in the least developing countries suffered less than those in emergent and other developing countries. The various instruments employed lead to result heterogeneity, demonstrating the importance of using the well-established instruments with the standard cut-off points (eg, GAD-7, GAD-2, and DASS-21 for anxiety, PHQ-9 and DASS-21 for depression, and ISI for insomnia). Conclusions The research effort on mental health in developing countries during COVID-19 has been highly uneven in the scope of countries and mental health outcomes. This meta-analysis, the largest on this topic to date, shows that the mental health symptoms are highly prevalent yet differ across regions. The accumulated systematic evidence from this study can help enable the prioritization of mental health assistance efforts to allocate attention and resources across countries and regions.


Data sources and search strategy
This study is built upon the existing meta-analyses; we contacted their authors to ask for their coding data for aggregation, since relevant meta-analyses already exist at the scale of individual countries or regions. We searched PubMed, Embase, PsycINFO, Web of Science, medRxiv, and Google Scholar in English for meta-analyses on mental health symptoms of the key adult populations during COVID-19 from January 1, 2020 to September 22, 2021. For example, the following Boolean operators on three sets of keywords were used in Web of Science: • (ALL= ((2019-nCoV OR 2019nCoV OR COVID-19 OR SARS-CoV-2 OR (Wuhan AND coronavirus)) AND ("depressi*" OR "anxi*" OR "insomnia" OR "sleep" OR "distress" OR "PTSD" OR "post-traumatic stress disorder" OR "mental health" OR "psychiatric"))) AND (TS= "meta-analysis").
The search targeted meta-analyses that focused on the prevalence of anxiety, depression, distress, insomnia, and PTSD in specific regions or countries during COVID-19. Meta-analyses that did not specify regions or countries were excluded. When multiple meta-analyses existed on the same region, the most comprehensive analysis was chosen. We contacted the authors of these meta-analyses to request their original coding data. Figure  S1 in the Online Supplementary Document details the flowchart of the overall search process.

Selection criteria
To be included in this review, the evidence must have studied the prevalence of at least one mental symptom outcome (eg, anxiety, depression, distress, insomnia, and PTSD) of adult populations such as frontline health care workers (HCW), general HCWs, general adult population, medical students, and general adult students in any developing countries based on the definition of [11] during the COVID-19 pandemic, published in English.
We excluded empirical studies using the following criteria: 1. Population: children, adolescents, or specific niche adult populations such as COVID-19 patients, inpatients, or other patients, adults under quarantine, or pregnant/postpartum women in developing countries 2. Methodological approaches: non-primary studies such as reviews, qualitative or case studies, interventional studies, interviews, or news reports 3. Measurements: non-validated mental health instruments (i.e., self-made questionnaires) or instruments without a validated cut-off score to calculate a prevalence rate (i.e., STAI, SCL-90 for anxiety and depression).

Selection process and data extraction
The coding data from the original meta-analyses were reviewed and recoded based on a single pre-developed coding protocol to ensure the consistency and comparability of the results [10]. If the authors of the existing meta-analyses did not share the coding data, we identified their original empirical studies, independently extracted the relevant data into a coding book based on the same coding protocol [10] and crosschecked their coding. We assessed the eligibility of each study by reading its full text to remove redundant empirical studies and code relevant information such as the authors and year of the study, title, publication status, sample locations, date of data collection, sample size, response rate, population, age (mean, SD, min, and max), gender proportion, instruments, cut-off scores used, the prevalence/mean/SD of the mental health outcome, and other notes or comments. In cases where the two coders disagreed, a lead researcher checked the study independently to determine its coding. The lead researcher integrated all the coding information and reviewed the key information such as mental outcomes, instruments, outcome levels, prevalence, population, sample, and regions.
To consistently analyse the data, we verified the independence of mental health symptoms and samples. For instance, if a study used more than one instrument to measure a mental health outcome, we reported the results based on the most popular instrument. We used the three typical cut-off levels of mental health symptoms (above mild, above moderate, and severe) as standards for reporting the prevalence above mild, moderate, and severe levels. If an empirical study reported the prevalence rates differently from the three-level norm with cut-off points, such as at overall level, we converted the prevalence rates into above mild, above moderate, or severe based on the typical cut-off points of the instruments used.

Assessment of bias risk
The Mixed Methods Appraisal Tool (MMAT) was used as a quality assessment tool [12][13][14]. Two reviewers independently assessed scores (ranging from 0 [low] to 7 [high]) for the studies using the tool dictionary and guidelines, crosschecked their coding, and resolved disagreements. Studies were categorized as high, medium, or low quality based on the score of >6, 5-6, or <5, respectively.

Statistical analysis
A random-effects model was used (the metaprop package in version 16.1 of Stata) to compute the pooled estimates of outcome prevalence between populations by assuming that these studies were randomly selected from their targeted populations [15].
Given the high degree of heterogeneity of the true differences in the effect sizes [16], we ran a meta-regression to regress the prevalence upon outcomes (five types of mental health symptoms), severity of outcome (above mild/above moderate/severe), five major population groups (frontline HCWs, general HCWs, general population, adult students, medical students), and on continents or regions, sample size, research design, and study quality. Given the size of Asia, which contains 60% of the world population, we used the sub-continental regions of Asia (Central, East, Southeast, West, and South). The other continents were not subdivided due to being smaller both in terms of populations as well as the number of conducted studies, so that the regions do not contain too few samples.
The meta-analytical results of our study enable the prediction of prevalence rates while accounting for multiple factors at the same time, thus offering a superior model over prior meta-analyses, which accounted for predictors separately [17,18]. Hence, based on the results of meta-regression, we predicted the prevalence rates of anxiety, depression, and insomnia symptoms at mild above, moderate above, and severe for frontline HCWs, general HWCs, general population, and general students in the seven regions. Due to small sample size, we did not predict the prevalence rates on distress and PTSD or medical students. The statistical significance is taken at the 95% confidence interval level.
The DOI plot and the Luis Furuya-Kanamori index [19] were constructed to assess publication bias [20,21]. We used event ratio as the primary effect measure for the pooled estimates.

Study screening
The search generated smaller elementary meta-analyses on mental health symptoms during COVID-19 [9,10,18,[22][23][24][25][26][27][28][29], and we were able to obtain the original coding results from seven of them. The aggregation resulted in a total of 461 studies, 341 of which were unique studies that fit the criteria for this meta-analysis of developing countries ( Figure S1 in the Online Supplementary Document). More than 80% of the studies covered anxiety and depression symptoms (45.36% and 38.87%, respectively). Just over one-tenth investigated insomnia symptoms (11.03%); few studies investigated PTSD (2.65%) and distress (2.51%). The studies reported the prevalence rates using cut-offs at the "mild above" (38.87%), "moderate above" (37.82%), and "severe above" (21.84%) level of symptom severity.
We next break down the studies by country and region. PTSD -Posttraumatic stress disorder *A study may include multiple independent samples. An independent sample in a study may report anxiety, depression, and insomnia at the levels of mild above, moderate above, and severe. Hence, the total number of prevalence rates is larger than the total number of independent samples. †Two studies included studies from different regions. and South Asia countries. Overall, our analysis contains seven regions: Africa, East Asia, Europe, Latin America, South Asia, Southeast Asia, and West Asia.
The 341 papers employed a wide arrange of instruments to assess mental health (

Major issues from findings of the key study characteristics
Our analysis reveals several widespread issues in mental health research during COVID-19, such as a wide array of used instruments, inconsistent reporting of prevalence rates, inconsistent use and reporting of cut-off points, varied cut-off values for determining the overall prevalence as well as the severity, and other issues on reporting standards and terminologies. Table 3 summarizes popular instruments used for measuring the five mental health symptoms with their primary cut-off point and different variants. Table S2 in the Online Supplementary Document summarizes the full list of instruments used by the individual studies included by this meta-analysis. All these issues may contribute to the heterogeneity and confusion in accumulating evidence.

A myriad of instruments:
The individual studies on mental health research during COVID-19 employed a wide variety of instruments with varying degrees of popularity and validity, making it challenging to compare or accumulate evidence.

Admixed outcome severity level:
The individual studies reported the prevalence rates at a range of symptom severity. First, the studies use different terminologies when reporting the overall prevalence rates. The overall   prevalence rate could indicate the percentage with moderate symptoms or above, or mild symptoms or above (eg, [62]). Even worse, many studies did not specify if the overall prevalence rate used cut-off at the level of above mild or above moderate. Second, some studies use other terminologies, such as "extremely severe" [63], "very severe" [64], or "very high" [65], "moderate-severe" [66], "moderate to severe" [67,68], "moderately severe" [69], and "poor" (40), making it even more challenging to categorize symptoms. We manually recoded all the studies that indicated their cut-off scores.
Clarity on the cut-off points used to determine severity: Some studies employed non-standard or unusual cut-off scores [70], at times without referencing validation studies that supported the use of those special cutoff scores (eg, [50,71]). Some studies did not report the cut-off score used or did not provide any references [72,73], making the comparison and accumulation difficult. All cut-off variants of the 5 mental health symptoms are listed in Table 3.
Pooled prevalence rates of mental health symptoms Table 4 reports the pooled prevalence rates of mental health symptoms by subgroups of population, outcome, severity, and region. The meta-analyses generally found mental health symptoms to be highly prevalent yet different across regions. Comparatively, Africa had the worst overall mental health symptoms (39%), followed  Among different populations, medical students had the worst overall mental health symptoms (38%), followed by general students (30%) and frontline HCWs (27%). Adults suffered most from distress symptoms (29%), followed by depression (27%) and anxiety (25%). Overall, a whopping 43% of adults in developing countries suffered from mild above mental health symptoms, 21% suffered moderate above, and 8% severe mental health symptoms.
The results of subgroup analyses of popular instruments of mental health symptoms show the various instruments lead to different results ( Table 5). While the prevalence rates of anxiety measured by GAD (27%) and DASS (29%) are relatively close, they are significantly different from those measured by SAS (7%), HADS (39%), and BAI (17%). The prevalence rates of depression differ significantly among studies with different measurements, specifically, PHQ (30%), DASS (26%), HADS (33%), SDS (12%), and CES (36%). The prevalence rates of distress measured by DASS (30%) and CPDI (27%) are significantly higher than those measured by K6 (18%). The prevalence rates of insomnia measured by ISI are 22%, which was significantly different from those measured by PSQI (29%) and AIS (44%). While the prevalence rates of PTSD are 15% when measured by PCL by 36% measured by IES. At least partially due to the popularity and standardized usage, the anxiety symptoms measured by GAD-7, GAD-2, and DASS-21 appear more comparable than those measured using other measurements. Similarly, the depression symptoms measured by PHQ-9 or DASS-21 and the insomnia symptoms by ISI appear to be more comparable.

Meta-regression on the prevalence of mental health symptoms
To better explain the heterogeneity of the prevalence of mental health symptoms, Table 6 reports the results of a meta-regression analysis. The meta-analytical model explained over 51% of the variance of mental health CI -confidence interval, PTSD -posttraumatic stress disorder, HCW -health care worker *The total independent samples are larger than the number of studies because some studies included multiple samples. †The total sample sizes are larger than the total sample of the 404 independent samples because one sample can assess multiple mental health outcomes. symptoms among these studies (R-squared = 51.8%, tau 2 = 0.15). The prevalence rates of depression (P = 0.012) are significantly higher than those of anxiety (reference). The prevalence of severe mental health symptoms is significantly lower than that of moderate mental illness (reference) (P < 0.001), which is in turn significantly lower than that of mild mental illness (P < 0.001). The prevalence rates of general HCWs and the general population are significantly lower (P < 0.001) than those of frontline HCWs (reference) (P < 0.001).
The prevalence of mental health symptoms of African adults was significantly higher than in South Asia (reference) (P = 0.001), which in turn was significantly higher than in East Asiam(P < 0.001) and Southeast Asia (P < 0.001) yet not significantly different from West Asia, Europe, and Latin America (P > 0.05). The prevalence rates reported by studies with larger sample size are significantly lower than those of studies with smaller sample size (P = 0.018). The prevalence rates of mental health symptoms in emergent and other developing countries were significantly higher than those in the least developing countries (reference) (P = 0.012). Analyses of studies with a higher quality rating (P = 0.06), publication status (P = 0.72), and research design (P = 0.31) did not predict significant prevalence rates. Table 7 shows the predicted prevalence rates of mental health symptoms by populations, outcomes, severity, and regions by the meta-analytical regression model. Figure 2 and Figure 3 show the predicted prevalence rates of depression and anxiety symptoms in different countries or regions, respectively.

Sensitivity analysis
Our meta-analytical model considered several factors, such as publication status (insignificant), sample size (significant), and article quality score (insignificant). Furthermore, excluding each study one by one from the meta-analytical model did not significantly alter the findings. A visual inspection of the sensitivity plot, however, revealed that there is significant asymmetry. Figure S2 in the Online Supplementary Document reports the DOI plot in combination with the Luis-Kanamori (LFK) index, which has higher sensitivity and power than a funnel plot [74,75]. LFK index scores of ±1, between ±1 and ±2, or ±2 indicate "no asymmetry", "minor asymmetry", and "major asymmetry" respectively, and hence the LFK index of 5.61 represents major asymmetry. Therefore, the presence of publication bias is likely for mental health prevalence studies under COVID-19.

Meta-regression findings
Thanks to a large number of samples in developing countries overall, we were able to conduct meta-regression to account for multiple predictors at the same time to enable better prediction of the prevalence of each mental health symptom. The accumulated evidence shows that several predictors are significantly associated with prevalence rates of mental symptoms during COVID-19, including the severity and type of mental symptoms, population, region, sample size, and study characteristics.
The severity of mental symptoms, largely unaccounted for in prior meta-analyses, was found to contribute greatly to the heterogeneity of prevalence rates; hence, future research on mental health needs to break down and pay special attention to the severity and specify its level. The significant differences revealed by this study call for more meta-analyses on varying levels of severity to provide evidence for practitioners relevant to their concerns.
Among the mental health symptoms examined, distress and depression generally had the highest prevalence rates. Our findings suggest that practitioners need to pay more attention to distress and depression of various populations under the COVID-19 pandemic [76][77][78].
While not significantly higher than frontline HCWs, general adult students and medical students suffered more than general HCWs and the general population. More than a two-third of studies investigated general HCWs and the general population to generate more meta-analytical evidence, which suggests that policymakers and health care organizations need to further prioritize frontline HCWs and students in this ongoing pandemic. Medical (including nursing) students [79] are worthy of special attention.

A mental health research agenda during COVID-19
Our systemic review and meta-analysis uncovers several widespread problems in the individual papers that impede evidence accumulation. We offer a few concrete suggestions for focusing research and reporting future mental health studies for authors, editors, and reviewers (Table 8), to improve the quality of mental health studies and to facilitate evidence accumulation in future meta-analyses. To make results consistent and comparable, we strongly suggest researchers to use standardized scales with well-established cut-off points (see Table 3 for the popular instruments to measure mental health symptoms and their cut-off points and Table 5 for the results of subgroup analyses on instruments). Table 8. A list of recommendations for conducting and reporting future mental health research studies Outcome and instrument 1) Study health outcomes that have higher prevalence rates, eg, distress 2) Use the well-established instruments with the standard cut-off points listed in Table 3.

Severity of the symptoms
3) Report more levels of severity of symptoms and the cut-off points used 4) Specify the meaning of overall prevalence, whether above mild or above moderate 5) Specify the cut-off values used with the reasons/references Characteristics of the samples 6) Report the sampling dates 7) Report the age/gender of the participants 8) Report participant rate Population 9) Separate and focus on frontline HCWs from general HCWs 10) Separate and focus on general adult students and medical students Study design 11) More future research using cohort designs

Study limitations and future research
This meta-analysis has a few limitations. First, there may be some bias because all studies were English. Second, 96.48% of studies included in this meta-analysis were cross-sectional surveys, and we call for more cohort studies to examine the effect over time [80]. Third, the validity of our findings rests upon the quality and reporting of the original studies. As discussed before, individual mental health studies varied in their usage of instruments, cut-off scores, the use of cut-off scores to define mental symptoms, and the reporting standards. For example, the overall prevalence refers to "above the cut-off of mild" in some studies, yet "above the cutoff of moderate" in others. Worse, many studies report the overall prevalence without specifying which/how cut-off scores are used. While we focused on the severity, the cut-off points, and the ways in which individual studies used this information, various approaches contribute to additional noise and variance in the analysis. It is also possible that the diagnostic systems might need to be adjusted across contexts, but such adjustments need to be carefully validated and reported. Fourth, we are limited in examining linear effect, and future research may examine nonlinear effect, as past research has shown age and distance to epicentre may have nonlinear effect on mental health [8,81,82]. Lastly, various classification schemes and terminologies exist on "developing countries" exist, such as low-and middle-income countries (LMIC), newly industrialized countries, emerging markets, third world countries, etc. and future research may use our data to analyse based on other classification schemes.

CONCLUSION
Since the COVID-19 pandemic started in November 2019, hundreds of studies have documented the mental health of major populations by the key mental outcomes and varying levels of severity across the world. This systematic review and meta-analysis synthesized the evidence on the prevalence rates of mental health symptoms in developing countries under the COVID-19 pandemic. We hope this meta-analysis reveals and synthesizes not only the accumulative evidence on mental health research but also reveals key directions for this important research stream.