Mental health and disasters

The mental health impacts of disasters are fluid, with the likelihood of negative mental health outcomes fluctuating depending on the severity, scale, and exposure time of the event. Palinkas and Wong (2020) outlined the challenges with documenting and effectively addressing the disparate negative mental health impacts of acute, subacute, long-term, and existential impacts of disasters [3]. Berry et al. (2010) developed a causal framework for understanding the mental health impacts of climate-related disasters based on a literature review [4]. Disasters affect communities and individual physical and mental health both directly and indirectly. Injuries and fatalities – from heat illness, crush injuries, drowning, shock, among others – are examples of direct effects on physical health [5,6]. Further, direct injuries resulting from a given disaster can also contribute to sustained negative mental health outcomes such as post-traumatic stress, depression, and anxiety [7].

In addition to immediate, direct health impacts, disaster-related factors can indirectly exacerbate negative mental and physical health. For example, income loss, especially for those with lower socioeconomic status, can cause stress, anxiety, and depression [8,9]. Increasing economic inequality, driven in part by more frequent and severe disasters resulting from climate-related hazards can further exacerbate the negative mental health impacts these events [10]. Community impacts can mediate both individual mental and physical health outcomes, while physical and mental health are interdependent [4], although somatization of disaster-related trauma is not universally observed across disaster types [11,12].

A range of hazard types can lead to poor mental health outcomes, and these outcomes can occur both acutely and longitudinally. There is not a standard timeframe for assessing mental health impacts of disasters resulting from natural hazards. Most studies assessing mental health following disasters are cross-sectional or repeated cross-sectional studies, with fewer studies assessing impacts longitudinally, and the timepoints considered vary across studies [7]. Across the disaster and recovery cycle, post-traumatic stress, anxiety, and depression increase in populations experiencing wildfires and exposed to wildfire smoke [13,14]. Pietrzak et al., (2012), assessed mental health disorders following Hurricane Ike in Galveston and Chambers counties, Texas, at four time points, and found that post-traumatic stress was most prevalent between 2- and 5-months following the event [15]. Further, the prevalence of all mental health conditions attributed to Hurricane Ike decreased over time. In winter storm Uri, Texans with disabilities, racial and ethnic minorities, and those with prior traumatic storm experiences were more likely to have anxiety or depression six months after the disaster event [16]. Following the 2016 Fort McMurray wildfire in Alberta, Canada, 15% of displaced individuals reported depressive symptoms a year after the wildfire [17]. Sri Lankans from coastal districts who experienced the Indian Ocean tsunami of 2004 were more likely to report partial or post-traumatic stress disorder eight years after the event compared to those without exposure [18].

Interpersonal and self-directed violence and disasters

One of the key considerations for the relationship between disasters and interpersonal violence is the mediating role of external stressors in instigating new or exacerbating existing poor mental health. Mental illness is not, in and of itself, a predictor of violent behaviors; however, in conjunction with other circumstantial factors, such as substance use, disaster exposure, or other external stressors, poor mental health can contribute to violent behaviors [19]. While the relationship between natural hazard exposure and death by homicide or suicide is not well established, it is clear that natural hazard exposure increases perpetration of intimate partner violence and family violence [20] and extreme heat exposure increases a the incidence of various violent behaviors [21]. There are a number of potential confounding and moderating variables for the relationship between disaster exposure and violent deaths, but measurement of these variables is particularly challenging [22]. Matsubayashi et al. (2013) found disparate effects of disasters on suicide rates at the prefecture level in Japan [23]. Rates were dependent on the severity of damage – measured physical and economic losses – resulting from the hazard, and rates varied across age and sex combinations. While findings on the relationship between, and temporality of, violent deaths after disaster exposure are not consistent, negative mental health impacts – including post-traumatic stress, depression, and anxiety – which can contribute to violent behaviors are consistently observed throughout the literature.

Given these complexities, it is critically important to consider the mental health impacts of disasters, including not only violent behaviors but also violent deaths. The aims of this work are 1) to characterize violent deaths in non-disaster and disaster periods and 2) to assess changes in incidence of violent deaths for counties eligible and ineligible for public assistance in non-disaster and disaster periods. By assessing deaths resulting both from suicides and interpersonal violence, this study adds to the literature base for understanding sequelae of violence in disaster contexts.

Study description and population

The National Violent Death Reporting System records homicide and suicide deaths across the United States. Five states were selected for this study – North Carolina, Oregon, Oklahoma, Wisconsin, and Colorado. These states were selected based on 1) the availability of data for the three months before and after a federally declared disaster and 2) the coverage of priority hazard types represented by the instigating hazard events (i.e., flooding, storms, tropical cyclone, fire), and 3) the absence of co-occurring, compounding, or cascading disasters.

The three-month disaster period was selected to measure acute disaster-related mental health impacts rather than more longitudinal impacts outside of the immediate response period. FEMA public assistance eligibility was used as a proxy measure for disaster exposure because federal assistance eligibility requires a community to have loss and damage that surpasses local and state-level capacity to address resulting acute needs, response, and recovery. It is important to note, however, that smaller scale disasters that are not granted federal assistance can still have tremendous effects on impacted communities, and, even within those eligible for public assistance, there is appreciable variation in need. States and associated hazard information are summarized in Table 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. States and associated hazard information.

https://doi.org/10.1371/journal.pone.0337968.t001

Ethics statement

This work did not include human subjects research; data were de-identified with no way of ascertaining the identities or individuals represented in the data set. Therefore, review from an institutional review board was not necessary or required. Secondary, deidentified death record data were used in this study, and oral or written consent could not be given and was not required. Data access was granted March 17, 2024. All findings are reported in compliance with suppression guidelines for ensuring anonymity from the Centers for Disease Control and Prevention.

Measures

Deaths from each state were aggregated at the county level, and deaths were assigned to the individual’s county of residence. The pre-disaster period was defined as the period three months directly preceding the recorded disaster start date in the corresponding federal disaster declaration (Fig 1). The disaster period covered the three months – operationalized as calendar months (see date ranges in Table 1)– following the disaster start date, as defined in the federal disaster declaration. The first date of the disaster period is the date of event onset. This time frame was selected to capture the impacts of acute stress from given disaster events and response as opposed to the longer term stressors related to recovery and rebuilding. Because NVDRS only records deaths, and to ensure that injuries leading to death were accurately captured in the pre-disaster or disaster period, injury date (rather than death date) was used to assign the individual to the period. Using injury date ensures that the causative injury event preceding a violent death occurred in the termporal range of interest. To assess the potential influence of disaster exposure on violent deaths, data were stratified by eligibility for federal disaster public assistance, as defined by federal disaster declarations [24]. Under the Stafford Act, state governors and tribal executives can request federal aid to address materiel, recovery, and economic needs, among others, that surpass local, state, or tribal capacity to address these needs [25]. Upon an event being declared a major disaster or emergency by the President of the United States, public assistance grants are disseminated through the U.S. Federal Emergency Management Agency (FEMA) to aid in community and public agency response and recovery efforts [26]. Disaster declarations specify which counties within a state are eligible for public, individual, or other forms of assistance. For this study, public assistance eligibility was used as a proxy for disaster-affected counties: counties eligible for public assistance were ‘affected’ while counties in the same state that were ineligible for public assistance were ‘unaffected.’

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Schematic showing pre-disaster and disaster periods, normalized to a common timepoint.

https://doi.org/10.1371/journal.pone.0337968.g001

Sociodemographic variables were used to describe deaths. To account for small cell counts and suppression, race (White only/ other), sex (male/ female), and marital status (married or partnered/other) were included as binary variables. Education was measured as a categorical variable (high school or less/ post-high school education/ missing). Age was treated as a continuous variable. Because the unit of analysis for this study was the aggregate of affected and unaffected counties, sociodemographic characteristics are reported to contextualize the underlying population but are not included in regression models.

Analysis

Individual-level frequencies and proportions representing the underlying demographics of deaths recorded in the NVDRS were calculated. To compare deaths among those eligible for public assistance and those not eligible in pre-disaster and disaster periods, chi-square p-values are reported for categorical variables (i.e., race, sex, marital status, and education), and Kruskal-Wallis chi-square p-values are reported for median age and interquartile range. Crude incident rates were calculated with 95% confidence intervals (S1–S3 Tables).

Model specification for incidence rate ratios

This study uses an ecological pre-post (i.e., pre-disaster and disaster period) design with within-state internal controls (i.e., affected an unaffected counties). Incidence rate ratios comparing violent deaths in disaster periods to pre-disaster periods were calculated using Poisson regression with a population denominator offset and log link, using the GENMOD procedure in SAS v9.4 (Cary, NC). Overdispersion was assessed by comparing the mean and variance of the count data, and Poisson regression was an appropriate modelling selection. Incident rate ratios and 95% confidence intervals are reported for all states together, as well as stratified results for each state individually.

Unit of analysis

Counties were aggregated into those ineligible and those eligible for public assistance, and results are stratified by these units of analysis.

Predictor variables

For all models, disaster period (yes/no) was treated as a fixed effect. When controlling for state to generate overall rate ratios comparing non-disaster and disaster periods, residence state was treated as a fixed effect.

Outcomes

Models were specified with all deaths, suicide deaths only, and other violent deaths only, as outcomes of interest. ICD-10 codes corresponding to intentional self-harm or sequelae of intentional self-harm were used to determine suicide deaths. All other deaths were considered ‘non-suicide violent deaths.’ All violent deaths include any death recorded in the NVDRS database, including both suicide and other forms of violent deaths. Other violent deaths include homicides, deaths in individuals killed by law enforcement officers in the line of duty, violent deaths of undetermined intent (including manslaughter), and unintentional firearm deaths.

Limitations

In addition to the limitations and implications on study results previously discussed, there are other important limitations to consider. Not all states have reported to NVDRS, and the years with reporting coverage are not standard across states. Further, coverage within reporting states is not necessarily inclusive of all counties, with some states only reporting for a subset of their population. In selecting states for this study, we considered all of these factors, as well as the presence of a federally declared disaster within the timeframe of available data.

For the purposes of this study, we were interested in assessing the rate of deaths resulting from both interpersonal and self-directed violence in the acute period following disaster events. The relevance of this timeframe was demonstrated by Pietrzak et al., who found the prevalence post-traumatic stress peaked between 2- and 5-months following the event [15]. Some work suggests that deaths resulting from self-directed violence occur at a longer-term interval, with increased rates 2-years following exposure to a disaster event [28,42]. As reporting on violent deaths improves, temporal aspects should be considered within the larger framework of understanding the effects of disasters on violent deaths, particularly as disaster recovery complexity extends recovery periods and focuses on building resilience.

This study aggregates data at the county level. To reduce potential biases in exposure classification, we assigned deaths to the county of residence, not county of injury or death. For example, a resident of one county could have died in a neighboring county ineligible for public assistance dependent upon the location and accessibility of skilled trauma centers. A more granular understanding of individual-level disaster exposures could provide different or more nuanced findings. Known confounders of the relationship between disaster exposure and violent deaths at the individual-level were not available and/or comparably measured at the county-level for all counties in the study. Future work should not only account for potential aggregate-level confounders but also consider confounders or effect modifiers that differ between suicide and other violent deaths individually.

We used eligibility for federal public assistance to determine if a county was disaster affected. However, the effect of a disaster on a community is considerably more nuanced than the presence or absence of a federal disaster declaration. For example, populations within a community could still experience highly disparate disaster impacts from one another, underscoring the importance of capturing disaster exposure on a more granular scale. Different classifications to better capture the nuance of disaster exposure should be explored in future studies that aim to assess the causative relationship of disasters on suicide and violent death mortality rates. Additionally, more complete reporting on decendent’s disaster exposure within the NVDRS reporting framework could greatly improve the feasibility of assessing the impact of disasters on violent deaths at the individual-level.