1. Introduction
Climate change is increasing population exposure to weather-related hazards, such as extreme precipitation events, storms and flooding [
1,
2]. In the UK, the government has identified flooding as posing particularly high risks to people and communities [
3]. Alongside damage to homes and businesses, there is disruption to domestic utilities and transport links [
4]. Exposure to extreme weather events such as flooding results in ‘psychological casualties’ [
5], with significant impacts on mental health.
Capturing the mental health impacts of extreme weather events such as storms and floods presents many research challenges. Large representative surveys are required to compare the mental health of exposed and non-exposed populations in analyses with measures of other factors (for example, an individual’s financial circumstances) that may explain associations between storm or flood exposure and mental ill-health. A review of evidence on the mental health of flood-exposed populations found that the large majority (77%) of studies did not include a comparison group and noted a lack of attention to potential confounders like socioeconomic status [
6]. Additionally, robust studies require detailed measures of people’s mental health, including treated and untreated mental health disorders, based on clinically validated diagnostic criteria. Only well-resourced health surveys are equipped to collect such information for large population surveys [
7].
This study uses a large, national and representative mental health survey in England to investigate (i) the social profile of those with recent experience of storm- and flood-related damage to their home and (ii) whether the experience independently predicts common mental disorder.
2. Materials and Methods
The Adult Psychiatric Morbidity Survey (APMS) is funded by England’s Department of Health and Social Care, commissioned by NHS Digital (Leeds, England), and conducted by the National Centre for Social Research and Leicester University. It is the primary source of information on the mental health of people living in England [
8]. The most recent APMS (
n = 7525) ran from May 2014 to September 2015, with fieldwork evenly distributed across this period. For the first time, the survey included the following question: ‘Has your home been damaged by wind, rain, snow or flood in the last six months’; it, therefore, covered the period November 2013 to September 2015. This period included four months, from December 2013 to March 2014, of severe winter storms and extensive flooding in the UK [
9,
10].
The survey is based on a stratified multi-stage random sample of the general population aged ≥16 years in private households. It includes detailed mental health screening tools and assessments to measure treated and untreated mental disorder assessed to the diagnostic criteria set out in the tenth edition of the International Classification of Diseases [
11]. The questionnaire is administered face-to-face in people’s own homes, with the most sensitive information collected by computer self-completion. The overall response rate for the survey (57%) is in line with other national surveys of similar scale [
12]. The analysis used specially developed weighting variables to adjust for known patterns of non-response according to area- and individual-level characteristics; by applying a multiplier to each participant, the sample was rendered representative of the wider population. Ethical approval for APMS 2014 was obtained from the West London National Research Ethics Committee, approval reference number 14/LO/0411. Data use approval for the analysis was secured via NHS Digital. Further methodological information is available elsewhere [
8].
Outcome measure: The outcome variable was the presence of common mental disorder (CMD), based on the CIS-R (Clinical Interview Schedule-Revised), which covers non-psychotic symptoms in the past week [
13]. The CIS-R operationalises standardised diagnostic criteria from the 10th International Classification of Disease [
14]; algorithms were applied to responses to assess for depression, phobias, generalized anxiety disorder, obsessive compulsive disorder (OCD), panic attacks and ‘CMD not otherwise specified’. CMD was defined as the presence of one or more of these six disorders. Our analysis also used variables related to other experiences of distress. Posttraumatic stress disorder (PTSD) was measured by screening positive on the self-completed PTSD Checklist civilian version (PCL-C); it indicates the presence of trauma-related symptoms in the past week and that clinical assessment for PTSD is warranted [
15]. Respondents were also asked about suicidal thoughts, non-fatal suicide attempts and self-harm (without suicidal intent) in both the face-to-face and self-completion sections of the interview. Use of mental health treatment and service use was established via questions about prescribed mental health medication and psychological (talking) therapy around the time of the interview.
Exposure measure: The exposure measure is the experience of damage from wind, rain, snow or flood to the home in the six months prior to interview. This period included the winter floods of 2013–2014 which were ‘one of, if not the most, exceptional periods of winter rainfall in at least 248 years’ [
16]. However, the APMS question did not allow us to distinguish people affected by external flooding from sea, rivers, or surface flow from those affected by escape of water inside the house (e.g., leaking boilers and pipes), so we refer hereafter to ‘storm- and flood-related damage’. Data provided by a leading UK insurance company with broad and representative geographical exposure through a multi-channel distribution route [
17] indicate that, over the 2013–2015 period, weather-related damage accounted for 46% of all claims due to storms, flood, freeze or escape of water, and 62% of all claims costs resulted from these causes. Most weather-related claims were due to storm damage (91%), with 6% due to flood damage and 3% due to freezing. The relative contribution to overall costs of weather-related claims was 79% from storm damage, 19% from floods and 2% due to freezing. Thus, the relative cost of an average flood-related claim is over 3.5 times greater than a storm-related claim. It should be noted these insurance-based data are likely to underestimate the costs of weather-related events; compared to owner occupiers, those in rented accommodation are less likely to have contents insurance (which provides cover for the financial cost of damage to, or loss of, personal possessions located within the home) and those living in more deprived communities are less likely to insure adequately against flood damage [
18,
19].
Covariates: We included individual-level socio-demographic measures: sex, age group (in ten-year bands), broad ethnic groups (White British, White other, Black, Asian, and mixed or other), highest educational qualification and current employment status, together with measures of financial strain (being in serious debt and payment arrears; experience of a major financial crisis in the past six months). Health-related behaviours associated with poorer mental health were also included: harmful drinking (Alcohol Use Disorders Identification Test-AUDIT) [
20], smoking and drug dependence [
21]. General health was measured by self-reported health status (poor, fair, good, very good or excellent), an item from the 12-item Short Form Health Survey (SF12) [
22].
At household-level, we included quintiled equivalised household income, housing tenure and household composition, together with measures of housing conditions that may be related to both storm- and flood-related damage and CMD (presence of mould in the home and whether the participant was able to keep the home warm enough in winter) [
23]. Area-level measures included area deprivation (quintiles of ranked Index of Multiple Deprivation (IMD) scores), which measures relative levels of deprivation in small areas of England called ‘lower layer super output areas’ [
24], and urban/rural location. For the later, we used the government’s standard Rural–Urban Definition for Small-Area Geographies (RUC) classification; the derivation is complex but primarily based on the population density of hectare cells and proximity of other settlements [
25]. To protect anonymity, region of residence was not included in the dataset released for the study.
Statistical analysis: Analysis was carried out using Stata v14, weighted and adjusted for the complex survey design. Descriptive sample profiling is presented in
Table 1 and in the
Supplementary Materials. Logistic regression was used to calculate unadjusted odds and adjusted odds for living in a storm- or flood-damaged home (aim 1) and for CMD (aim 2).
To investigate the social patterning of storm and flood damage (aim 1), we examined bivariate (unadjusted) associations by sex, age group, socioeconomic circumstances (employment status, household income, housing tenure) and area characteristics. Age and other factors with significant bivariate associations with living in a weather-damaged home were then entered into a multiple regression model with area deprivation (
Table 2, Model 1). The final model retained factors significant in Model 1, together with age (
Table 2, Model 2).
To establish whether recent damage to the home from storms and floods independently predicted CMD (aim 2), we began with an overall profile of CMD risk, examining bivariate associations with CMD for storm and flood exposure and for sex, age group, employment status, household income, housing tenure and area deprivation. We also identified other factors that, in these bivariate analyses, predicted CMD. Our first multivariate model (
Table 3, Model 1) included storm and flood-related exposure together with socio-demographic and socio-economic factors that were significant predictors of both CMD and storm- and flood-related damage in bivariate analysis. Storm- and flood-related exposure was then added to this model of CMD risk (
Table 3, Model 2). This final model additionally controlled for factors that were significantly related to CMD in our bivariate analysis and were also known from previous research to be predictors of CMD [
26,
27,
28,
29]. This allowed the independent contribution of storm- and flood-related exposure to the presence of CMD to be estimated after controlling for other relevant factors.
4. Discussion
Climate change is increasing the occurrence of extreme weather events, including storminess (including wind) and the frequency and intensity of floods [
3,
30,
31,
32,
33]. The number of properties in the UK exposed to at least a 1 in 75-year flood risk is predicted to increase by 41% under a 2 °C temperature rise and by 98% under a 4 °C temperature rise [
30]. In response to these threats, policy agencies at local, national and global level are seeking to develop disaster preparedness strategies [
34,
35,
36]. Evidence on the mental health of affected populations is recognized to be an important element of such strategies; studies of populations experiencing flooding have found stronger associations with mental than physical health [
37,
38,
39,
40,
41,
42,
43], with rates of CMDs like depression and anxiety elevated among those whose homes are flooded [
6,
44]. However, much of the evidence derives from studies of flood-exposed areas and not from general population surveys [
6]. In particular, there are few large-scale studies with clinically validated measures of mental health of exposed and non-exposed populations [
35].
Our study helps to address this gap. It is based on England’s major mental health survey (APMS), a large population survey that collects information on people’s mental health using validated tools. The most recent survey was conducted in the year following an exceptional period of flooding. December 2013–March 2014 saw extensive flooding which, compared to other periods of flooding, tended to be of longer duration [
10]. Over 4.2 million flood warnings were issued [
10] and over 10,000 residential properties were flooded. For the first time in the APMS series of surveys, the 2014/15 survey asked participants if they had experienced damage to their home (due to wind, rain, snow, flood) in the six months prior to interview. Using this exposure measure together with the detailed social data collected in the APMS, we investigated the social profile of those who experienced storm- and flood-related damage to their home and whether experience of recent damage independently predicted common mental disorder. As a population survey, the APMS enabled us to compare the mental health of exposed and non-exposed adults in analyses that controlled for other factors that may elevate the risk of CMD (for example, social disadvantage).
We found, firstly, that advantaged groups (owner-occupiers and higher income households) were more likely to report having experienced storm- or flood-related damage to their homes. Information on the location of study participants was not provided in the APMS dataset released for our study. National-level evidence points to an association between deprivation and flood risk in England [
45,
46,
47]; however, this overall pattern masks different social distributions of risk between tidal and fluvial flooding [
48]. Deprivation is associated with greater risk of tidal flooding, an association shaped by the geographical patterning of deprivation; coastal areas, including seaside towns, are home to many of England’s most deprived communities [
48,
49,
50]. Social gradients are flatter for fluvial flood risk and, where gradients are evident, more advantaged groups tend to be at greater risk [
46,
48]. Among the most advantaged fifth of the population (those in the two least-deprived deciles) at risk from river flooding, over half (104,000) live in the South East [
46]. The 2013–2014 floods were regionally concentrated, with the largest number (39%) of flooded homes in the South East [
10], England’s most populous and most affluent region with the highest rates of property ownership [
44,
51,
52,
53].
We found, secondly, that experiencing storm- or flood-related damage in the home increased the risk of having a CMD, over and above the established predictors of poor mental health. As this suggests, our study confirmed the importance of known risk factors for CMD; the predictors evident in our study were: being female, aged 16–24, living in a deprived neighbourhood, being in debt arrears, having poorer general health, and harmful or hazardous patterns of alcohol use. Our study identified storm and flood exposure in the home as an additional risk factor. Our measure did not include those, who while avoiding storm or flood damage to their own home, experienced flood or storm damage in their neighbourhood. Other studies have found that the risk of mental health difficulties is also elevated in this wider group [
44,
54], a heightened risk that may moderate the associations we found between storm and flood damage to the home and poorer mental health.
We note some caveats about our study. England’s floods of winter 2013–2014 had a distinctive regional and meteorological profile. Flooding was concentrated in Southern England, where heavy rainfall resulted in fluvial flooding. In December 2013, rainfall was 154% above the national average for that month and January 2014 was the wettest January on record for Southern England. Soils were quickly saturated and rain ran off the land into rivers that subsequently flooded [
55]. Because the dataset released for our study included only limited area-level data, we could not take account of location in our study. It, therefore, needs to be complemented by national studies of mental health where flood damage relates to different forms of flooding (e.g., tidal flooding) and/or affects different areas, including England’s less wealthy regions and coastal communities [
48,
49,
50].
The APMS is cross-sectional; it, therefore, has no prospective measures of mental health. In consequence, we could not control for prior mental health, and it was not possible to derive causal inference. However, because storms and floods are infrequent, unpredictable and localized events, prospective studies are the exception and, like the APMS, most studies rely on post-exposure measures of mental health [
6]. Our study validates previous work on the mental health impacts of a home being damaged by extreme weather, and provides new evidence on the broader associations of extreme weather damage (beyond flooding) with depression and anxiety disorders. Associations between such exposures and mental health may operate through a range of pathways, including material and financial losses, loss of the psychological security that people derive from their home and locality and disruption of health and social services [
37,
38,
39,
40,
41,
42,
44,
56,
57,
58,
59].
Finally, while large, the sample was too small to examine some variables robustly (for example, ethnic group), to estimate risks for particular CMDs (for example, depression) or to investigate those reporting suicide attempts.
Studies of the mental health of individuals exposed to flooding often rely on area-level measures, for example, patient lists from GP practices and local authority lists of residential addresses in flood-affected areas [
54,
60]. However, flooding can result in population displacement [
6,
54,
60], with a consequent under-reporting of adverse health events and increased healthcare use in those moving out of their home [
61]. Based on individual-level data, the APMS included those displaced from their homes. We found little evidence of population displacement following weather-related damage to the home (0.1%); this suggests that evidence on mental health impacts based on surveys of properties may miss only a small proportion of the flood-exposed population (albeit those most adversely affected).