Abstract
Background Mental health problems (MHP) in COVID-19 patients and survivors were anticipated already during early stages of this pandemic. We aimed to synthesize the prevalence of anxiety, depression, post-traumatic and general distress of major virus epidemics since 2002.
Methods In this systematic review and meta-analysis, we searched PubMed, PsycINFO, and Embase from 2002 until April 14, 2021 for peer-reviewed studies reporting prevalence of MHP in adults with laboratory-confirmed or suspected SARS-CoV-1, H1N1, MERS-CoV, H7N9, Ebolavirus, or SARS-CoV-2 infection. We included studies that assessed MHP with well-validated and frequently used scales. A three-level random-effects meta-analysis for dependent sizes was conducted to account for multiple outcome reporting. We pooled MHP jointly and separately for mild or moderate-to-severe severity by acute (one month), ongoing (one to three month), and post-illness phase (longer than three months). A meta-regression was conducted to test for moderating effects. PROSPERO registration: CRD42020194535.
Findings We identified 59 studies providing a total of 187 effect sizes. Range for sample size (n=14-n=1002), females (22-79%), and mean age (32-72 years). MHP prevalence was higher for mild (35·5-46·3%) compared to moderate-to-severe MPH (17·3-22·3). MHP, in general, decreased from acute to post-illness from 46·3% to 38·8% and for mild and moderate-to-severe from 22·3% to 18·8%, respectively. We found no evidence of moderating effects except for non-random sampling and H1N1 showing higher proportions.
Interpretation MHP decreased over time but were still on a substantial level at post-illness. This highlights a need for rapid access to mental health care and rehabilitation planning in affected individuals.
Introduction
A large body of evidence from COVID-19 and earlier epidemics such as SARS-CoV-1 and Ebola raised concerns regarding long-lasting neurological and psychological problems in infected individuals.1–4 Yet, the pathological processes in which COVID-19 leads to detrimental psychoneurological effects are still largely unknown and are subject to extensive ongoing research. Pathophysiological mechanisms including immune response, vascular damage, detrimental effects of critical illness and side effects from treatments may play an important role.5 Psychological problems (thereinafter mental health problems[MHP]), are at least partly attributable to sociodemographic and environmental factors. Infected individuals may suffer from life-threatening complications with uncertain survival or recovery, social isolation, and reduced access to social support, all of which can lead to loneliness and MHP.6,7 Affected individuals may face compromised access to care due to escalating case numbers and overwhelmed health services. Post-illness, survivors may face ongoing symptoms such as reduced physical functioning, fatigue, social and economic issues such as stigmatization with a refusal of services, and reduced working abilities.1,3,8,9 Mental health problems may also be aggravated by a history of pre-existing mental health problems.10
As treatment possibilities improve, there is a shift towards physical and psychological long-term effects in affected individuals. Estimating the magnitude of long-lasting mental and physical problems, including mechanisms and risk factors, is critical to estimate individual, societal and economic costs and to facilitate treatment and rehabilitation planning.11 Results from original studies and reviews estimating MHP prevalence values can be limited in estimating the burden of disease due to a large heterogeneity in study quality, like assessment methods used.1,12
We are not aware of a systematic evidence synthesis including patients and survivors of major infections disease epidemics in the last two decades that include severe acute respiratory syndrome coronavirus 1 (SARS-CoV-1), swine flu (H1N1), Middle East respiratory syndrome coronavirus (MERS-CoV), avian influenza (H7N9), Ebolavirus, and severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). In the present meta-analysis, we aimed to estimate the overall MHP prevalence and the prevalence proportion of anxiety, depression, post-traumatic stress, and general distress with validated and widely used scales in suspected and laboratory-confirmed patients and survivors. We used a meta-regression to investigate potential moderating effects on the overall MHP proportions.
Methods
Search strategy and selection criteria
The study was registered with PROSPERO (CRD42020194535) and is reported in adherence to PRISMA guidelines.13 As compared to the registered protocol, we focused on infected individuals/survivors and excluded original studies on the general population and health-care workers due to the excessive number of systematic reviews already available for these populations.
We searched PubMed, PsycINFO, Embase and Google Scholar for peer-reviewed studies in the range from January 1, 2002 to April 14, 2021. Reference lists of all eligible studies and topic relevant reviews were screened to identify studies that were potentially missed. We used a broad set of keywords (appendix p 2) related to epidemics of interest and frequently used and established assessment instruments found by our last review,12 as defined in the inclusion section. Inclusion criteria were: (a) peer-reviewed articles using a quantitative methodology; (b) published in the languages Dutch, English, French, German or Spanish; (c) providing MHP prevalence values assessed by any versions of the impact of event scale (IES), the Center for Epidemiologic Studies Depression Scale (CES-D), the Patient Health Questionnaire/Generalized Anxiety Disorder scale (PHQ/GAD), the Hospital Anxiety and Depression Scale (HADS), and the General Health Questionnaire (GHQ); (d) adult patients/survivors (≥ 18 years) with suspected or confirmed SARS-CoV-1, H1N1, MERS-CoV, H7N9, Ebolavirus or SARS-CoV-2 infection. Exclusion criteria were: (a) subgroups of patients/survivors including psychiatric patients, marginalized individuals, people with chronic physical conditions; (b) articles not providing prevalence (appendix p 3).
After removing duplicates electronically and manually, two authors (SJZ, PK) identified studies meeting the inclusion criteria based on title and abstracts independently and blinded to each other’s decisions. Any study that met the inclusion criteria was inspected independently and blinded in full-text by two authors (SJZ, CA, CB, PK) for closer inspection. Agreement on the eligibility of full-text was 90.3% (Cohens Kappa: 0·79 [95% CI 0·69-0·90]). Discrepancies were resolved through discussion.
Data extraction and coding
A standardized form was used to extract study data and quality (appendix p 4). Data was extracted and checked by two authors for each included study (SJZ, CA, CB, PK). Variables extracted for descriptive and/or moderator analyses were first authors, year of publication, country, world-region (China, Asia excl. China, Europe, Africa, America), study design (cross-sectional, longitudinal, intervention), sampling method (random/complete sampling, non-random/unknown sampling strategy), response rate, epidemic, sex ratio, mean/median age, proportion with a history of a mental health conditions, proportion in need of intensive care unit, proportion of health care workers, proportion with higher education, treatment received (inpatient, outpatient, or miscellaneous), and months follow-up defined as the time elapsed since treatment or discharge coded as acute (≤ 1 month), ongoing (1-3 months), or post-illness (> 3 months) based on a recently proposed recording system.14 Outcomes were prevalence proportions defined as the number of positive classified cases by assessment instrument divided by sample size. We calculated the overall MHP prevalence excluding the GHQ as a too generic measure of distress and MHP by domains including anxiety (GAD and HADS scale), depression (PHQ, CES-D, HADS-D scale), post-traumatic stress (IES-Scale), general distress (GHQ scale), and somatization (PHQ-15 Scale). Prevalence values were further stratified by follow-up timepoint and severity defined as at least mild symptoms or at least moderate-to-severe symptoms cut-off by assessment instruments (appendix p 5).
Appraisal of the evidence
Quality was appraised independently by SJZ, CA, CB, and PK using eight items of the nine criteria version of the Joanna Briggs Institute Critical Appraisal Checklist for Studies Reporting Prevalence Data.15 Each item was rated with yes, no, or unclear and covered sampling frame, sampling/recruitment, sample size, subjects and setting description, coverage, standardized procedures, transparent statistical analyses, and response rate. We excluded the item on valid assessment methods as we only included studies using well-established assessment scales (e.g., PHQ). To the best of our knowledge, there is no published recommendation of weighting and scoring.15 We therefore binarized each quality item into yes/no (no or unclear) and calculated the quality achieved in percent (possible range from zero to eight out of a maximum of eight). We classified at least seven points (>87%) as good to excellent, five to six (63-75%) as moderate, less than four (≤50%) as poor. Discrepancies were resolved through discussion. Average inter-rater agreement across all items ranged from 80·0 to 96·7% (Cohens Kappa; 0·75 [95% CI 0·58-0·92] to 0·93 [95% CI 0·83-1·0]).
Data Analysis
We conducted a random-effects meta-analysis and meta-regression for dependent and non-dependent effect sizes for point prevalence values and moderators of overall MHP and by MHP domains separately including anxiety, depression, post-traumatic stress, general distress, and other MHP such as somatization. All estimates were separated by severity (mild or moderate-to-severe) (appendix p 5) and by follow-up time. In the case of dependent effect sizes for studies reporting multiple outcomes in the same participants, a three-level mixed-effects model was fitted (taking withing study variation into account).16 To pool prevalence with 95% CIs, each study’s proportion estimate was transformed by the Freeman-Tukey double-arcsine transformation for meta-analysis and back-transformed after model fitting.17 We used I2 to determine heterogeneity for analysis in non-dependent effect sizes. In dependent effect sizes, the distribution of total variance (%) attributed to between and within-study variance was estimated.18 Meta-regression was conducted on arcsine transformed proportions due to better statistical properties as compared to untransformed proportions. Restricted maximum likelihood was used for parameter estimation. The threshold for significance was set to p<0·05. Statistical analyses were conducted in R (version 4.0.3) using the metafor package.19 Sensitivity analyses were performed by the investigation of influential effect-sizes using DFBETAS and Cook’s distance.19
To analyze the data, we conducted the following a-priori defined steps. First, we estimated the overall pooled prevalence, including all effect sizes (excluding general distress as a generic measure of general distress) and by MHP domains separately. All estimations were stratified by mild or moderate-to-severe severity and acute, ongoing and post-illness follow-up timepoint. Second, a meta-regression on the overall mild and moderate-to-severe MHP values was conducted including follow-up timepoint, sex, age, education, history of a mental health condition, health-care workers, duration of hospitalization, intensive care treatment, type of treatment, response rate, sampling method, world region, and epidemic type as moderators. All moderators were tested individually while adjusting for follow-up timepoint in months. A complete model to test all moderators jointly was not possible due to the substantial missingness patterns.
Results
Study Characteristics
The systematic search yielded 3304 records, of which 59 (61 samples) were included in the analyses (Figure 1). The number of individuals ranged from 14 to 1’ s002, with a proportion of females ranging from 22% to 79% and mean/median age from 32-72 years. A total of 187 effect sizes for mild and moderate-to-severe MHP prevalence values were reported, with one study that assessed samples in different countries20 and two that had overlapping samples.21,22 Studies covered China (n=20), Asia excluding China (n=14), Europe (n=17), Africa (n=6), and America (n=3). 42 (71%) studies investigated SARS-CoV-2, eight (14%) SARS-CoV-1, four (7%) MERS-CoV, four (7%) Ebolavirus, and one (2%) H1N1, while no study covered H7N9 (Table 1). Time elapsed since treatment/data collection ranged from 0 to about 40 months, with 116 effect sizes providing data for acute, 35 for ongoing, and 36 post-illness stages, respectively. Studies rarely reported complete data on all descriptive or moderator variables leading to a substantial amount of missing data. While female/male ratio was provided regularly (>98%), other variables such as (history of a psychiatric condition, percent health care workers) were not regularly provided. Overall, 18 studies showed an excellent, 15 studies a moderate and 26 a poor quality on the appraisal scale (Table 1 and appendix pp 6-7).
Characteristics of included studies reporting mental health problems in virus disease patients and survivors (n=59)
PRISMA Flow chart of included studies reporting on mental health problems in virus disease patients and survivors
Pooled prevalence of mild or moderate-to-severe MHP including all domains jointly and separately by all timepoint of follow-up are shown in Figure 2 and Table 2. Studies contributing to summary effects are shown in the supplementary material (appendix pp 8-13). Mild problems occurred at a substantially higher rate as moderate-to-severe MHP. We found a general downwards trend from the acute to ongoing/post-illness stage. This trend was also seen in the meta-regression including follow-up time as a moderator. The overall MHP significantly decreased from acute to the post-illness stage for mild (Coeff.arcsine −0·18 [95% CI −0·31 to −0·06], p=0·0038) as well as for moderate-to-severe (Coeff.arcsine −0·15 [95% CI −0·25 to −0·04], p=0·0080) severity. However, this trend was not significant for the ongoing vs. acute stage (appendix p 14). Meta regression including mild or moderate-to-severe domains of anxiety, depression, and post-traumatic stress as outcomes, separately (data not shown), yielded a significant decrease in moderate anxiety from acute to post-illness only (Coeff.arcsine −0·14 [95% CI −0·26 to −0·02], p=0.0228).
Pooled prevalence proportions of mild or moderate-to-severe mental health problems including all domains jointly and separately at acute, ongoing, and post-illness stage
Pooled prevalence proportions of mild or moderate-to-severe mental health problems including all domains jointly and separately at acute, ongoing, and post-illness stage
MHP=mental health problems, Mod.-sev.=moderate-to-severe. Meta-analysis was conducted only where at least two effect sizes were available. The analysis was conducted with a random effects meta-analysis in the case of independent effect sizes or a three-level random effects model in the case of dependent effect sizes. We used double arcsine transformation for variance stabilization. Displayed are the back-transformed estimates in percent.
Aside from a decreasing time trend, prevalence values were still substantial over all domains at post-illness phase for mild 38·8% [95% CI 33·6−44·1] and moderate severity 18·8% [95% CI 13·4−25·0]. Likewise, prevalence proportions for mild severity by domain at post-illness ranged from 33·5% [95% CI 24·8−42·7] to 43·5% [95% CI 34·1−53·1] for anxiety and post-traumatic stress, respectively. Moderate-to-severe MHP ranged from 12·1% [95% CI 5·3−21·1] to 39·1% [25·7−53·2] for anxiety and post-traumatic stress, respectively. Data for pooling general distress was available only in the ongoing stage. Pooling for somatization, and general distress (acute, post-illness), was not possible since only single effect-sizes were available (appendix. p 13).
Generally, there were many more effect sizes available for the acute phase, especially for depression and anxiety and less for the ongoing or post-illness phases (Table 2). To emphasize, the estimates for mild or moderate post-traumatic stress were based on very little data (two effect sizes), while the majority of estimators showed substantial heterogeneity with a significant part of variance that could be attributed to between and/or within-study heterogeneity (dependent effect sizes). Sensitivity analyses did not change the interpretation.
A meta-regression was conducted to test for moderating effects on mild or moderate MHP severity including all domains jointly. All moderatos were tested separately and adjusted for follow-up time in months (appendix p 14). We found no evidence of moderating effects for sex, age, education, history of a psychologic condition, health care workers, duration of hospitalization, treatment in ICU, type of treatment, response rate, and world-region. In contrast, we found some evidence for lower mild MHP proportions in random vs. non-random sampling methods (Coeff.arcsine −0·16 [95% CI −0·28 to −0·04], p= 0·0083) and higher proportions of moderate-to-severe MHP in H1N1 vs. SARS-CoV-2 (Coeff.arcsine 0·39 [95% CI 0·003 to 0·78], p=0·0483). However, the contrast between H1N1 and SARS-CoV-2 should be interpreted cautiously since it is based on very little data.
Discussion
We conducted a systematic review and meta-analysis that explored the prevalence proportions of acute, ongoing and post-acute mental health sequelae after infection with SARS-1, MERS, Ebola, H1N1 or SARS-Cov-2 viruses. We included 59 studies providing a total 187 effect sizes in the analysis. In general, we found that mild psychological problems had higher prevalence proportions than moderate to severe conditions. We also found that acute infections were associated with higher prevalence proportions of mental health problems than post-illness. This, however, was not found uniformly across all time-points and mental health problem severity groups.
The overall picture suggests that any mild and moderate-to-severe psychological conditions will be experienced by 39% and 19% of infection survivors for longer than 12 weeks. Likewise, anxiety, depression, and post-traumatic stress, and general distress were substantial beyond the acute phase. Combined, these results show that a considerable proportion of infection survivors will suffer from mental health problems severely for a longer time.
Our results support meta-studies on earlier epidemic outbreaks. A meta-analysis with Chinese publications on SARS-1 found a decrease of mental health problems over time but reported at 12 months post-hospital discharge a level of average distress above population norms.77 A systematic review on post-Ebola virus disease studies gathered publications that found considerable depression prevalence and other psychological sequelae in Ebola survivors.78
Our results provide more precision than earlier meta-analytical results from COVID-19 meta-analyses. A not yet peer-reviewed study on various long-term sequelae of COVID-19 reported prevalence of 12% for anxiety and 13% for depression.9 It is, however, unclear at what time this was assessed. A meta-analysis on early neurological and neuropsychiatric studies (published until July 2020) reported point prevalence values for anxiety of 15.9% and of 23.0% for depression.1 Our results similarly fall in that range. However, milder mental health conditions are somewhat more prevalent and can have detrimental effects as well.
The moderator analysis has shown that post-illness COVID-19 mental health sequelae are not fundamentally different compared to consequences of earlier virus epidemics. While the information on the post-viral health detriments was available from earlier infection outbreaks, it is somewhat astonishing that the risks of longer-term conditions were overlooked in the early days of the pandemic and that it took quite some time to receive the science and media attention it now has.79
The moderator analysis has also shown effects for the time point and for the sampling method in particular. This suggests that methodological details and study quality are very important for assessing the contribution of single studies. In general, and as reported from earlier epidemiological studies on mental health effects of epidemics,80 the study quality has to be evaluated as mixed. While about 56% reached a moderate to high study quality about 44 per cent showed a poor quality. Studies frequently suffered from non-transparency regarding sampling frame, recruitment, sample size and provided only a poor description of the study sample and setting and showed low response rates. Furthermore, at least one third showed an unclear or insufficient coverage of the identified sample. In contrast, studies predominantly assessed MHP in a standardized way and clearly described how prevalence was calculated.
Our findings highlight the importance of mental health interventions generally, but at an acute stage of infection specifically. First, clear information about the disease (e.g., infection rates, quarantine, vaccination) is not only important to address uncertainty and fear but also to improve mental health literacy within the population.81 Therefore, public communication should also integrate virus- and pandemic-related mental health issues. Furthermore, multidisciplinary mental health support (that includes psychiatrists, clinical psychologists, mental health nurses and other professions) should be delivered at an early stage.82 Access to mental health interventions could be supplemented by digital health online and smartphone technologies if face-to-face treatment is limited.81,83,84 A clinical screening for psychiatric symptoms would ideally be an integrative element already within the acute stage. Therefore, mental health awareness appears to be an important aspect within primary care and emergency departments in particular. Moreover, social support for impaired individuals should be strengthened. This could be supported by prevention strategies that include community-based collaboration among education and employment services, families and housing, and voluntary work.81
Strengths and Limitations
The main strength of this research is that it is a large systematic review and meta-analysis that encompasses the major virus outbreaks within the last 20 years. Moreover, this research differentiates between diagnoses, time-points and moderators that allows a comprehensive view regarding the prevalence values of MHP.
This research also has some limitations. Firstly, most studies within this systematic review and meta-analysis lack representativeness. Often, patients with unstable conditions or those within ICU units, or those not hospitalized were not included. This selection bias potentially led to an underestimation/overestimation of the prevalence values. Further, many original studies were generally of poor or moderate quality with incomplete data or a lack of random or complete sampling. This might be due to the urgent need for conducting such studies in a pandemic situation. Secondly, this meta-analysis methodology was limited by the fact that further well-validated instruments (e.g., State-Trait Anxiety Inventory) were not included in this research. The meta-regression should be interpreted cautiously since many studies did not provide data on all moderators. Furthermore, it was not possible to include all hypothetical moderators. Specifically, the physical disease severity and the burden of late physical effects were not covered as this was often not reported within studies.
Conclusion
This systematic review and meta-analysis explored the prevalence proportions of acute, ongoing, and post-acute mental health problems after a major virus outbreak within the last 20 years. Generally, mild mental health problems had higher prevalence proportions than moderate to severe conditions. Moreover, most mental health problems had a higher prevalence proportion at an acute infection stage compared to a post-acute stage. Our findings further underline the importance of the study quality that is not often given within the original studies. Therefore, guidelines aiming to advice assessment and reporting acute and post-illness MHP in a standardized way are urgently needed. Overall, this research highlights the fragility of mental health after infection from a pandemic virus. Additionally, it emphasizes the need for the early provision of mental health interventions that follow long-lasting post-viral mental health sequelae.
Data Availability
Relevant data is provided in the manuscript.
Conflict of interest
We declare no competing interests.
Author Contributions
SJZ and DR conceived and designed the study. SJZ, CB, CA, and PK conducted the literature search and selection, data extraction, and appraisal of study quality. SJZ conducted the statistical analyses. SJZ, PK, DR and AIL wrote the first draft of the manuscript. All authors contributed to data interpretation, manuscript revisions and approved the final version of the manuscript.
Funding
No funding.
Acknowledgements
No acknowledgements
References
