History for some or lesson for all? A systematic review and meta-analysis on the immediate and long-term mental health impact of the 2002–2003 Severe Acute Respiratory Syndrome (SARS) outbreak

The review protocol was registered with the PROSPERO registry (ID: CRD42020183812).

This review included original studies written in English or Chinese from January 2003 till May 2020 that reported on psychological distress, psychiatric symptoms or diagnoses, and health-related quality of life (HRQoL) among SARS patients, healthcare workers and the general public in the five key outbreak regions (Mainland China, Hong Kong, Canada, Taiwan and Singapore) using standardised measures or clinical assessments by clinicians. Quantitative studies, including observational studies and interventional studies, were included. To ensure comprehensiveness, studies published in peer-reviewed journals or disseminated via other channels, such as conference proceedings or thesis databases, were included, provided full texts of the articles were available for assessment. Case reports and qualitative studies were excluded.

Major electronic medical and social science publication databases (Pubmed, Medline, EMBASE, Web of Science Core Collection and PsycInfo) were searched using combined search terms that covered SARS and a wide range of mental health conditions. Dissertation databases were searched for unpublished studies. Databases of publications in the Chinese language (e.g. CNKI and National Central Library Online Catalog) were searched using Chinese search terms (see full search terms in Additional file 1). Duplicates were first removed automatically and then manually. The titles and abstracts were screened independently by two reviewers (SWC and OWW), and any disagreement was settled by consensus. Full text articles were reviewed independently by two reviewers (SWC, EKW or SSC) for their eligibility, and any disagreement was also settled by consensus. We conducted a snowball search by identifying additional potentially relevant studies from citation lists of eligible articles. We also contacted corresponding authors of eligible articles through email and asked for their suggestions as to relevant studies and grey literature. SWC checked for overlapping of reported data during the data extraction process. Decisions to exclude articles during the data extraction process were jointly made by three reviewers (SWC, EKW and OWW).

Data extraction from each article was performed by SWC and one of the two reviewers (SWC, EKW, OWW) in duplicate using a standardised extraction form. Basic characteristics of the studies and data reporting mental health outcomes of relevance, including any psychiatric diagnoses and well-defined measurements of psychiatric symptoms, psychological distress and HRQoL, as well as selected predictors of these outcomes (defined after pilot exploration of the eligible studies) were extracted. Data from studies containing data of selected predictors that were not included in their analyses or results were still extracted, and basic univariate analysis was applied where appropriate. Interventional studies were treated as cross-sectional studies, from which only their baseline assessment data was extracted.

Based on the availability of studies that could be harmonised, the outcomes we were able to include in the meta-analysis were:

The statistics extracted from the studies for meta-analysis were either prevalence or mean (standard deviation (SD)). Estimates for studies that reported median and interquartile range (IQR) were converted to mean and SD [13]. Some studies did not report prevalence rates but reported only mean (SD) or median (IQR). For these studies, we employed Monte Carlo simulations to estimate the proportion of the outcome based on appropriate cut-off points for the tool used to measure the outcome. We assumed normality of the distribution when mean (SD) were reported or the data were symmetrical. We used a random effects model to estimate pooled prevalence rates for the above outcomes. We then used a random effects model to compute pooled estimates of the mean scores of HRQoL (mental health, role emotional and social functioning domains from the 36-item Short Form Survey (SF-36)) among patients.

In assessing the overall certainty of the body of evidence from the review, risk of bias in individual studies, inconsistency across studies and publication bias were considered. Risk of bias in individual included studies was independently evaluated by two reviewers (SWC, EKW or OWW) using the Study Quality Assessment Tools by the National Heart, Lung and Blood Institute [14]. Each study was rated as having poor, fair or good quality, which indicates significant, moderate and low risk of bias at the study level, respectively. For assessment of inconsistency across studies, we used the I² statistic to assess heterogeneity between the studies: values of 25, 50, and 75% were used to categorise the degree of heterogeneity into low, moderate and high, respectively [15]. To investigate sources of heterogeneity, we conducted subgroup and sensitivity analyses for outcomes that had sufficient sample sizes. We conducted subgroup analyses by time (relative to the epidemic or discharge from the hospital). Sensitivity analyses were conducted by age (≤40 years vs > 40 years), study design, and outcome measurement tool. We assessed potential publication bias through visualisation of funnel plots and Egger’s test for outcomes with a sample size ≥10. All analyses were conducted using STATA 14.0.

The reporting of this study followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (see PRISMA checklist in Additional file 4).

The major electronic databases yielded 6195 titles. Additional searches from citation lists of included studies and recommendations from authors of included studies yielded an additional 789 titles. After removing 4463 duplicates, 2521 titles and abstracts were screened. Two hundred ninety-four full-text articles were reviewed and eventually 80 studies were included in the systematic review (Fig. 1) (see Additional file 2 for the full list of excluded studies). Thirty-three studies primarily reported mental health outcomes of SARS patients, 28 reported outcomes of healthcare workers of affected regions, and 19 reported outcomes among the general public. Sample sizes ranged from 10 to 10,511. Forty-seven studies were cross-sectional in nature, and only 12 studies used clinical interviews as an assessment tool. Seventeen studies were published in the Chinese language (Table 1A, B & C); and three were unpublished postgraduate theses.

Studies reporting on mental health outcomes of non-infected healthcare workers of SARS-affected regions covered the epidemic period and up until 3 years post-SARS [48‐52, 54‐62, 64‐75]. The most commonly reported dimensions were PTSD, depression, and general psychological distress.

The estimated point prevalence of probable PTSD among non-infected healthcare workers from 11 available studies [48, 50, 51, 56, 57, 61, 65, 67, 69, 72, 73] was 26.7% (95% CI: 10.0–43.4%, I² = 99.2%; n = 2791). Subgroup analysis showed that the point prevalence during and after the epidemic was 38.1% (95% CI: 14.5–61.6%, I² = 98.5%; from six studies with total n = 1579), and 11.2% (95% CI: 8.2%-14,2%, I² = 51.1%; from five studies with total n = 1212) respectively. There was a significant difference between the prevalence of the two periods (p = 0.035) (Fig. 4a.

The estimated point prevalence of probable depressive disorder among non-infected healthcare workers from five studies [36, 56, 57, 74, 93] was 27.4% (95% CI: 4.1–50.7%, I² = 98.8%; n = 872), Subgroup analysis showed that the point prevalence after the epidemic was 11.9% (95% CI: 0–24.9%, I² = 95.3%; from three studies with total n = 689). There was an insufficient number of studies to conduct a subgroup analysis of the data from the epidemic period (Fig. 4b).

The estimated point prevalence of significant general psychological distress among non-infected healthcare workers during and after the epidemic was 40.7% (95% CI: 16.1–65.3%, I² = 99.5%; from five studies [49, 50, 62, 63, 65] with total n = 2786), and 32.6% (95% CI: 21.6–43.5%, I² = 91.6%; from five studies [67‐70, 72] with total n = 915) respectively. The prevalence of the two periods were not significantly different (p = 0.556) (Fig. 4c.

Only two studies used clinical interviews as part of their diagnostic tools, and both yielded a low rate of psychological distress among their subjects. Su et al. reported that the prevalence of PTSD and depression among their sample was 0 and 12%, respectively, around the end of the SARS outbreak [56]. The study by Lancee et al., which captured a subset of the sample of the study by Maunder et al. (2006), also reported a low rate of PTSD (1.5%) and post-SARS mental illness episodes (6.8%) 1 to 2 years after SARS [71].

Nineteen studies examined the effect of SARS on the general public [37, 53, 76‐91]. The lack of a pre-SARS baseline and long-term longitudinal data limited the usefulness of many of these studies in assessing the mental health impact of SARS among the general public. Furthermore, the sub-populations in these studies, which included hospitalised patients with non-SARS illnesses, college students, pregnant women and people under quarantine, were too heterogenous. We thus considered it inappropriate to pool the estimates from these studies. Two studies with a pre-SARS baseline showed an increase in anxiety and depressive symptoms during the epidemic [76, 80]. A series of Hong Kong population-wide studies by Leung et al. showed that the trend of the population’s anxiety levels followed the trend of the number of daily new cases [77]. One small prospective cohort study on college students showed that the subjects’ anxiety levels increased markedly in the early phase of the epidemic but gradually reduced over time [53].

It was unclear whether there were any increases in clinically significant outcomes among the public. A Hong Kong study that used longitudinal death registry data reported a significantly increased elderly suicide rate in the year of the epidemic, against a downward trend in previous years, with an unusual peak around the peak of the epidemic. The elderly suicide rate did not return to baseline in 2004 [90]. A study from Taiwan, which used a population insurance database with a high population coverage to study the effect of the global financial crisis of 2008–2009 on the incidence of depression, incidentally found a significant upward deviation from the expected incidence of depression starting from July 2003 (right after the epidemic) [91]. The presentation of this study’s methodology and analysis was limited, however, and the results should therefore be treated with caution.

Age (≤40 years vs > 40 years) was not found to be a significant factor explaining the heterogenicity of the results. The prevalence of PTSD among SARS survivors post-discharge from studies that used clinical interviews and self-reported screening instruments were not significantly different (see Supplementary Figures 4–8).

Due to the diversity of outcome measures and statistical methods, a quantitative analysis could not be conducted. We present the results qualitatively below:

Our results showed that mental health problems were common among SARS survivors. The prevalence rates of anxiety, depression, and PTSD from our analysis are comparable with an earlier meta-analysis that included studies from SARS, MERS and covid-19 [10]. Subgroup analysis of a recent meta-analysis by Yuan et al. on the prevalence of post-infectious outbreak PTSD (which includes studies from SARS) concluded that the prevalence of PTSD was 18.6% within 6 months post-infection, and 28.8% beyond 6 months of infection [12]. However, the different mix of studies in the subgroups (i.e. the fact that more studies in the within-six-months group came from covid-19, while studies from SARS predominated in the beyond-six-months group) makes interpretation of the difference between the subgroups difficult. Our review additionally introduces the first meta-analytic evidence demonstrating the chronicity of the mental health burden among SARS survivors in terms of HRQoL and psychiatric morbidities, including but not limited to PTSD. Despite the very different context between the two novel coronavirus outbreaks, our findings suggest similarities between post-SARS and post-covid-19 neuropsychiatric sequalae: the prevalence of psychiatric morbidities among SARS patients within the first 6 months post-recovery was remarkably similar to that of covid-19 patients in their early post-recovery period, and survivors of both infections reported prolonged excessive fatigue [96]. In the case of SARS, it is intriguing that a substantial proportion of patients still suffered from excessive fatigue years after recovery from SARS, which cannot be explained by the presence of other psychiatric disorders. We have little idea of the biological aetiology of chronic mental health sequelae and fatigue post-SARS, although what we do know is that the SARS-CoV virus can be isolated from the brains of SARS patients [97], just as it is possible to isolate SARS CoV-2 from the brains of covid-19 patients [98]. How the presence of SARS-CoV viruses in the brain affects patients’ mental health, however, is a question that remains to be answered. Overall, while we cannot conclude based on these resemblances that post-covid-19 neuropsychiatric sequalae will follow the same trajectory as that of SARS, our findings suggest that preparations will be needed to address the long-term mental health needs of covid-19 patients.

There are discrepancies among studies concerning healthcare worker status as a risk factor for poor mental health outcomes of SARS patients, and a definitive conclusion cannot be drawn from our analysis. Psychological distress among healthcare workers was high during the epidemic, but the level of psychiatric morbidities returned to a lower level afterwards. While these results are encouraging, one should not underestimate the impact of mental health consequences for healthcare workers on our healthcare systems: Lam et al. reported that 9/43 of healthcare worker patients with a psychiatric diagnosis left health care-related work (21%), as compared to 4.7% among those without a psychiatric diagnosis [43]. Maunder et al. reported that healthcare workers of SARS-hit Toronto had a significantly higher rate of missing work shifts due to stress or illness compared to SARS-free Hamilton since SARS (21.6% vs 12.6%) [72]. Given the large numbers of healthcare workers who are involved in the care of covid-19 patients and/or suffer from the disease themselves, the impact of the mental health morbidities of this population on healthcare systems will be, like the covid-19 pandemic itself, in uncharted waters. The increase in service demand, met with substantial manpower loss, may tip the balance in regions where healthcare resources are already overstretched. In preparing for the aftermath of covid-19, it is noted that occupational exposure and stigmatisation were consistently reported as risk factors for poor mental health outcomes among healthcare workers, suggesting that these will be issues that require close attention. The recent meta-analysis by Kisley et al. has already demonstrated that direct contact with infected patients during novel viral outbreaks (including SARS) is a significant risk factor for higher acute or post-traumatic stress and psychological distress [11], but this study used a restricted definition of exposure (i.e. direct contact) and did not differentiate between its acute and sustained effect. Our qualitative analysis suggests that the effect of working in high-risk venues might persist beyond the epidemic period.

While the evidence converges in suggesting that the general public experienced an increase in anxiety levels during the SARS epidemic, no conclusive evidence can be established regarding enduring or clinically significant mental health effects on the public due to the scarcity of longitudinal data spanning from the pre-SARS to post-SARS period. Other factors also affect the interpretation of available statistics: for example, the psychiatric bed occupancy of the greater Toronto area paradoxically dropped sharply during the early phase of SARS, but this may merely be a reflection of the obstruction of access to mental health services during an infectious outbreak [99]. A point of concern that needs to be highlighted, however, is the report from Hong Kong of the increase in suicide rates among older adults during the SARS epidemic [90]. Although this evidence was from a single study and from only one region, it was methodologically robust. A qualitative study has further identified several common factors among SARS-related suicides in older adults in Hong Kong: fear of contracting SARS, social isolation, disruption of normal social life, and the burden of existing long-term illnesses [100]. These factors are likely also present, if not to a greater extent, in the current covid-19 epidemic.

Stigmatisation was consistently reported to be associated with poor mental health outcomes among SARS patients and healthcare workers. Many SARS patients and their family members, healthcare workers, and residents from SARS-hit neighbourhoods encountered discrimination in various aspects of their lives [37, 64, 101]. The stigma against SARS patients, in particular, persisted years after the outbreak. One account reported that SARS survivors in Hong Kong were denied the opportunity to donate blood 10 years post-epidemic [102]. In the covid-19 context, there have been reports of stigmatisation of covid-19 patients and healthcare workers [103], which is unfortunate but unsurprising. De-stigmatisation appears crucial in mitigating post-covid mental health sequelae, but this would require more than evidence-based public education, as stigmatisation can be unintentionally institutionalised. For example, some researchers have argued that the establishment of special post-SARS clinics has in fact perpetuated SARS-related stigmatisation [104]. This highlights the importance of being mindful of this issue in all aspects of post-covid-19 policy planning.

To the best of our knowledge, this is the first comprehensive systematic review of evidence concerning the effect of SARS on mental health across multiple study populations and timeframes that also captures non-English written studies. This review is mainly limited by the methodologies and quality of the included studies. The studies included were primarily cross-sectional in design, and the association of risk factors and mental health outcomes found in these studies thus cannot be concluded as causal in nature. The few longitudinal studies tended to have small sample sizes. The methodological quality of the studies was generally in the poor to fair range, indicating significant to moderate risk of bias. We did not exclude low quality studies due to the scarcity of studies for specific outcomes, and also because we wanted to represent the available literature comprehensively. The inclusion of low-quality studies in the context of the small number of studies available for data synthesis for each outcome, however, limits the certainty of the results of the review. Another limitation of this study is that, due to the limited numbers of studies included in each meta-analysis and the availability of variables, we could only investigate heterogeneity among studies by mean age, study design, and outcome measurement tool. Since most of the studies used self-rating instruments for measuring outcomes, and the tools used for each outcome were diverse, it is reasonable to suspect that the diverse measurement tools could have introduced heterogeneity to the results of our meta-analysis. We could not investigate whether this was indeed the case, however, due to the limited number of studies available for each outcome measure. Another limitation is that, while we assessed publication bias for outcomes with ≥10 studies, the number of studies included in the funnel plots ranged from 10 to 12, which could limit the power to test for real asymmetry and publication bias. Taking these factors into account, the certainty of the body of evidence of this review is considered to be low.