The Long-Term Mental Health Impact of Disasters: A Systematic Review and Multilevel Meta-Analysis of Longitudinal Epidemiological Studies

Literature Search

Inclusion and Exclusion Criteria

We included articles published in English-language peer-reviewed journals that contained full-text reports of quantitative studies assessing disaster effects. As what constitutes a disaster is not always clear, we defined the term specifically to denote acute, collectively experienced events with sudden onset.¹³ (See also Supplemental Data File 1, https://links.lww.com/HRP/A254.) Following categorization of search terms, we formulated inclusion and exclusion criteria.

Screening Procedure

Titles and abstracts were divided and independently screened by two authors. Disagreements were discussed until we reached consensus. If necessary, a third person was consulted. When in doubt about a title/abstract, the article was retained for the full text stage. For manuscripts based on the same dataset, we included the first published article unless a subsequent paper measured more relevant mental health variables or included more assessment waves. If the more recent paper did not measure all variables assessed in the first published paper, we included both. We only used mental health variables from the initial paper, however, that were not measured in the later one. Studies that did not appear to meet all inclusion criteria when fully reviewed, and studies that did not report useful statistics for the mental health variables (e.g., if the variable was only displayed in graphs but not statistics) were excluded.

Quality Assessment

We conducted an assessment of study quality using an adapted version of the Quality Assessment Tool for Quantitative Studies from the Effective Public Health Practice Project.²¹ This assessment included criteria such as selection bias, study design, data collection methods, withdrawal, and dropout. Additionally, one item from the Newcastle-Ottawa Scale for cohort studies was incorporated to assess outcome quality.²² A study with no poor scores was considered strong, while those with up to two poor scores were deemed moderate. We excluded studies with three or more poor scores. (Supplemental Data File 3, https://links.lww.com/HRP/A255 contains more information on quality assessment per study.)

Data Extraction

From the included papers, we extracted available information on: disaster type and category, year, country, and region of disaster occurrence, population age (i.e., adult, children and adolescents, or all ages), relevant mental health outcomes observed (e.g., PTSD, depression, suicidality), number of assessments conducted (measurement waves), month of measurement (in cases with multiple measurement months, the last month was taken), type of instrument used, total sample size per wave (denominator), and number of participants with a mental health problem per wave (numerator). We calculated prevalence based on the numerator and denominator. The distinction between disorders and symptoms was not always consistent among different authors. We applied the term probable when used by the authors or when the study employed a validated screening or assessment questionnaire with recommended cut-off values. (See Supplemental Data File 6, https://links.lww.com/HRP/A256.) This was especially relevant for PTSD, posttraumatic stress (PTS), and distress. Data extraction started on 26 May 2020. Data were extracted by two authors (MD and AB) and three assistants, and were later double-checked. We clustered countries based on World Bank income categories (high, upper-middle, lower-middle, and low income).

Statistical Analyses

We used a multilevel (meta-analysis) logit-logistic regression model to generate pooled estimates of mental health problems in the disaster timeline.^23–25 This type of analysis enabled us to distinguish between fixed and random effects. Fixed effects capture specific characteristics that remain constant across outcomes, while random effects account for variability and differences among clustered outcomes. Clustering applies to hierarchical levels (observations, measurement waves, studies, and countries), as well as mental health problems and disaster categories. (Figure 1 displays the data’s multilevel structure.)

As a first step, we estimated the effect of several factors on the overall average prevalence of mental health problems (similar to the overall rate in other forms of meta-analysis). As a second step, we estimated average prevalence of mental health problems per disaster category (e.g., earthquake, flood, shooting), disaster type (natural or human-made), country income group (low to high), population type (children/adolescents, adults, all ages), study quality (moderate or high), outcome category (type of mental health issue), and month.

Characteristics of Included Studies

The literature search retrieved a total of 33,205 records to screen. After initially reviewing the records, we identified 783 papers requiring full-text review. Eventually, 71 studies, published from 1990 to 2024, were included for data extraction.^26–96 (See the flowchart in Supplemental Data File 4, https://links.lww.com/HRP/A257.) The average baseline female sex proportion, extracted from 68 studies was 54.03% (95% CI [51.04-57.02]). (See Supplemental Data File 7, https://links.lww.com/HRP/A258.)

The studies addressed disasters in 23 countries from four World Health Organization regions; almost half of the studies focused on China (including Taiwan) and the United States (34); Americas: Colombia (2), Honduras (1), Mexico (1), United States (8); Europe: Armenia (2), Croatia (1), France (2), Greece (1), Iceland (1), Netherlands (4), Norway (5), Spain (2), Sweden (3), United Kingdom (1); Southeast Asia: India (1), Indonesia (1), Nepal (1), Thailand (2); Western Pacific: Australia (1), China (26; 2 of which were from Taiwan, a high-income area), Japan (3), South Korea (2). (See Supplemental Data File 5, https://links.lww.com/HRP/A259 for a world map.) Nearly all studies investigated populations in high-income (35) and upper-middle-income countries (32); three studies addressed disasters in lower-middle-income countries, and one study followed up with survivors across countries from different income groups.

Most of the studied disasters took place in the year 2000 or later (55 of 71 studies; the first disaster took place in 1985, the last in 2017) and the majority concerned natural disasters (48 compared to 23 human-made). The 71 studies included 76 samples, covering multiple disaster categories: earthquakes (29), storms (e.g., cyclones and snowstorms) (9), terrorist attacks (9), tsunamis (8), accidents (6), floods (4), shootings (4), explosions (3), volcanic eruptions (2), fires (1), and landslides (1). The majority of studies were of moderate quality (53). The mental health outcomes differed: PTSD (43), depression (26), PTS (18), other (e.g., agoraphobia, dysthymia) (14), anxiety (11), sleep problems (11), distress (5), panic disorder (4), substance abuse (4), grief (2), and suicidality (2). Since several studies focused on multiple outcomes, the total number of measurement series adds up to 140. As Table 1 displays, these 140 measurement series included 403 observations divided over six waves: the first wave included 140 observations (38.64%; n = 52,937), the second wave included 140 observations (35.65%; n = 48,844), the third wave included 76 observations (14.57%; n = 19,964), the fourth wave included 40 observations (7.28%; n = 9,979), the fifth wave included 6 observations (2.85%; n = 3,911), and the sixth wave included 1 observation (1.00%; n = 1,369). Half (49.88%) of the observations were recorded within 12 months of the disaster and 93.05% within 48 months. (Supplemental Data Files 6, https://links.lww.com/HRP/A256 and 7, https://links.lww.com/HRP/A258 provide additional information for each study.)

Mental Health Problems: Prevalence, Determinants, and Variation

Table 2 presents results from the multilevel analysis. In the fixed effects analysis, the mental health burden in populations confronted with disasters changes across time (p<0.001). With other factors taken into account, prevalence does not differ among natural and human-made disasters, general age, or country income groups, nor is disaster year or study quality an influence. For the random effects analysis, we observed substantial between-measurement and between-study variance, as well as between-disaster categories and mental health outcomes. The pooled prevalence of current or recent mental health problems was 22.1% (95% CI [10.95–39.57]), yet the analysis indicates extensive heterogeneity underlying this estimate, both across and within studies. No prevalence estimates covered a time period longer than 12 months.

Figure 2 provides an overview of the stratified pooled prevalence estimates of post-disaster mental health problems, with 95% CI for each mental health outcome, disaster category and type, population age group, study quality level, and country income group, controlled for the other factors.

Figure 3 illustrates that the time effect (shown in Table 2) does not follow a straight linear trend; rather, it better fits a polynomial trend line. The decrease in the first few years is followed by an increase, peaking between 110 and 220 months (approximately 9–18 years), followed by a rather steep decrease. This graph also shows that most available data cover the first quarter of the timeline.

It was not possible to systematically compare prevalence in disaster-exposed populations to a control group. Only three of the 71 studies in our dataset contain control group measurements (19 of 403 observations).^41,57,89 The analysis demonstrates a higher average prevalence of 27.6% (95% CI: 22.60–32.60; n = 19) in disaster-exposed populations compared to 13.82% (95% CI: 9.03–18.71; n =19) in the control group (Two-Sample T-Test, p < 0.0001). Without adjustment for other factors, this suggests that the pooled prevalence in exposed populations is higher than in unexposed populations.

Limitations

Notable study limitations include restriction to English-language studies, the possibility of missing relevant studies (despite identifying a vast number of records through the search), and exclusion of continuous outcomes from our analysis. Additionally, though our range of mental health problems is similar to earlier reviews,^12,13,15,16 it is not exhaustive (for instance, psychotic disorders are excluded), and certain mental health outcomes, particularly PTSD, may be overrepresented. Moreover, not all assessments constitute clinical diagnoses. It is likely that a portion of study participants already had mental health problems before experiencing respective disasters.¹⁵ The selective nature of outcomes—ranging from clinical diagnoses and probable disorders to trauma-related symptoms and subjective experiences of distress—further complicates the longitudinal comparability of mental health outcomes, even though we found no significant differences among outcome types. In addition, cross-cultural differences and variations in data collection methods, diagnostic criteria, assessment instruments, and assessment timeframes pose potential biases. Likewise, our dataset only includes a portion of countries and regions, represents a limited number of lower-middle-income countries, and lacks low-income countries. (See Supplemental Data File 5, https://links.lww.com/HRP/A259.) This may affect the generalizability of the findings.

Each measured mental health outcome reflects a unique combination of characteristics linked to, among other things, the disaster, affected population, point in time, methodology used, and cultural and socioeconomic context, all contributing to heterogeneity. Our analysis is a first effort in considering multiple factors simultaneously. Nevertheless, we can extend and optimize our approach. Unfortunately, we were unable to include more risk and protective factors in the meta-analysis. In an attempt to include female sex, a confirmed risk factor for post-disaster mental health,^12,13,15 the baseline female sex proportion was not a reliable proxy for follow-up measurements.

Perhaps the most significant limitation, given our objective, was the limited availability of post-disaster measurements exceeding 48 months (6.95% of observations; 16 of 71 studies). Thus, we need to exercise caution in drawing firm conclusions about the shape of long-term trends. In the data on which Figure 3 is based, there are 11 measurements between 72 and 144 months (six to twelve years). In 8 of these, prevalence is higher than before 72 months, producing the second peak. All 8 measurements are from the last wave of their respective measurement series—the studies end at that point, with no follow-up measurements. Beyond 144 months, and up to 300 months (13–25 years), we found 9 observations. Each was preceded by at least 1 measurement before, but not in, the 72–144-month period. In every case, prevalence decreased during this time frame. These particularities are relevant for two reasons. First, we cannot determine how the trend continued in populations contributing to the second peak, since measurement stopped at that point. Second, the few studies with the longest follow-up (up to 300 months) do not sufficiently unpack the prevalence changes in the large middle area of the timeline. Longer-duration measurements with periodic follow-ups are essential to fill gaps and improve the accuracy of estimates. Such longer durations are also necessary to determine more detailed trajectories or response courses, as repeatedly recommended in the literature.^4,8,16,19

There is a need for long-term research across diverse geographies and cultural and socioeconomic contexts, as well as for additional syntheses addressing the identified limitations. These needs extend beyond the scope of this review. While we deliberately excluded pandemics, virus outbreaks, war, and armed conflict, these events significantly affect thousands of people worldwide as well, and studies confirm attendant mental health effects.^7,102,103 Therefore, their long-term influences, which may follow similar patterns to natural and human-made disasters, also warrant careful scholarly attention.

Implications

The accumulated body of knowledge on the mental health burden of disasters provides an invaluable starting point for public health and safety authorities, health care professionals, and other responders to prepare and implement preventive measures, mitigation strategies, or curative interventions across the disaster timeline. Our review, however, urges caution in forming firm general conclusions. Notably, we must make assumptions about differences in mental health risks according to disaster type or category with care. Further, the nature of disaster-related threats and losses, as well as vulnerability profiles of affected populations, can differ greatly across disaster settings, and thus, must be taken into account. Finally, perhaps the most profound practical implication is that, by relying on gradual post-disaster mental health improvement, we may have underestimated the long-term effects. The observed rise in mental health problems years after a disaster, suggests a “twin peaks” phenomenon. Nonetheless, given gaps in data beyond four years, the presence of multiple peaks is poorly understood, requiring validation and conscientious interpretation. In this context, it is notable that the increase identified in the combined longitudinal samples corresponds with findings from a systematic review of PTSD following the 9/11 terrorist attacks in New York City. This study reported declining mental health for responders and residents five to six years post-9/11.⁸

It is important to better understand the combination of factors and the dynamics that contribute to mental health risks at later points in time. A cascading interplay of risk factors, reflected, for instance, in accumulated unmet health care needs, unresolved and new problems, life events, lack of support, and secondary stressors in the aftermath, may account for the increased mental health burden.¹⁷ In the years following a disaster, social relationships can deteriorate,^15,18 and those affected may experience diminished recognition within their personal, professional, and community environments. The explanations described in the literature align with the factors identified by the authors of the included studies, who link them to elevated levels of mental health problems. For example, Tajima and colleagues⁸⁶ highlight major property loss, the death of a colleague, life-threatening danger, and discrimination or slurs as factors that contribute to long-term problem drinking. Goenjian and colleagues⁴⁷ attribute their findings to two post-disaster risk factors: chronic conditions (i.e., newly reported hypertension, diabetes, and arthritis) and adverse conditions in the earthquake zone (including lack of heat, electricity, housing, transportation, and destruction of physical and community infrastructure). Finally, Lima and colleagues⁶⁸ emphasize the difficult living conditions in affected communities: “This may be due to the continued stressful environment in which our subjects lived as they waited to be resettled in neighbouring towns that did not welcome this influx.”

Beyond initial effects and despite extensive restoration efforts (e.g., repair of housing, infrastructure, schools, shops, and businesses), these examples and our review suggest that a substantial portion of the population in disaster-affected communities continues experiencing a mental health burden—largely invisible in scientific reporting until now, and more universal in nature than previously assumed. Such burden is part of the persistent lingering effects of disasters that, as described by Bonanno and colleagues,¹⁵ transcends the individual: “Disasters affect the nature and structure of families. They impact the communities in which they take place, taxing their institutions and sometimes challenging their survival. The impact of a disaster can also ripple through larger segments of a society and sometimes alter the lives of people in remote regions far removed from the geographic loci of the event.” These potentially broad and enduring influences underscore the need for early, proactive, participation-driven recovery approaches. With support from long-term monitoring programs to guide shared decision-making, we can target at-risk communities to help alleviate the mental health burdens of disasters.

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of this article.

The authors would like to express their gratitude to Jonna Lind from the ARQ National Psychotrauma Centre library for her invaluable assistance in designing and conducting the literature search. Additionally, special thanks are extended to Mihaela Cimpian, Josephine Gellersen, Jenny Gerbecks, and Julie Penders for their diligent efforts in screening, data extraction, and quality assessment.