Survey population

In June 2007, two areas were particularly badly affected, South Yorkshire and Hull, followed by Worcestershire, Gloucestershire and Thames Valley, in July 2007. The HPA health impact assessments were conducted in South Yorkshire (September - October 2007) and Worcestershire (January - February 2008). These areas were chosen because of the feasibility of carrying out the surveys. The timing of the surveys after the floods varied (3 months after the event in South Yorkshire and 6 months after the event in Worcestershire) due to the logistics of obtaining sampling frames and getting the surveys rolled out.

Local public health staff identified areas that had homes affected in varying degrees by the floods and homes that were not affected. The non-affected households were included as controls. In South Yorkshire, the sampling frame was all addresses in one housing estate (n = 347, population 1,500), all addresses in one village (n = 436, population 1,013), all addresses on the Local Authority flooded properties register in one town (n = 626), and a random selection of 1,252 addresses that were not on the flooded properties register. The town had about 5,000 households, population 12,000.

In Worcestershire, we surveyed all addresses in 2 villages (n = 460), all addresses on the Local Authority flooded properties register in two towns (n = 533) and a random selection of 7,995 addresses (of a total of 12,500) that were not on these registers. As the towns were much larger than the villages in terms of population size (29,000 in the towns compared with 950 in the villages), we sampled 15 properties not on the flooded register for every property that was on the flooded register.

Data collection

A letter of invitation to participate in the survey was sent out to 'the occupier' of households by the Directors of Public Health for the area, outlining a choice of three methods to complete the survey, (i) by telephoning a free phone number and completing the questionnaire on the telephone, or (ii) by completing the questionnaire online or (iii) by return of a paper questionnaire using a freepost envelope that was provided. There was no restriction on who should complete the questionnaire. The questionnaire explained the reason for collecting the data and confirmed that any information provided would be treated with the strictest confidence and provided assurance that participation in the survey was voluntary and would not affect the healthcare received in any way. Due to the low response rate in South Yorkshire to the above three methods of data collection the decision was taken to use face-to-face interviews in the two villages in Worcestershire as these areas were small enough for this method of data collection to be feasible. The invitation letter sent to the two Worcestershire villages explained that data collection would be carried out in face-to-face interviews by interviewers who would visit the houses in the area the following week. Reminder postcards were sent out to non-responders after three weeks.

Exposure variables

We recorded flooding (the presence of water in (i) street or garden outside the house; (ii) basement or cellar of the property or below floor level in the ground floor rooms; (iii) above floor level in ground floor rooms), and damage to property (possessions, furniture, kitchen units, documents or decoration or structures).

Psychological exposure

We measured perceived impact of the floods using a five point likert scale of "much better, better, the same, worse, much worse" in response to the following questions: (i) As a result of the flooding how do you expect your financial circumstances to be? (ii) As a result of the flooding, how do you expect house values will be? We also asked about health concerns relating to the flood: 'Do you have any health worries relating to the flood for yourself or your family?'

Incident management variables

The questionnaire included questions about disruption to essential services (such as gas, electricity and water supply) and evacuation (request to evacuate, a request made and refused, a request made and complied with).

Outcome measures

Psychosocial impact was assessed by psychological distress (General Health Questionnaire-GHQ-12, score of 3 or more), [16] probable generalised anxiety (Generalised Anxiety Disorder GAD-7, score of 10 or more), [17] probable depression (Patient Health Questionnaire PHQ-9, score of 10 or more) [18] and probable post traumatic stress disorder (PTSD) (PTSD checklist-shortform, score of 14 or more) [19]. These are standard epidemiologic instruments with validated criteria indicating increased likelihood of mental health symptoms (screening tools).

Statistical analysis

We examined the proportion of respondents who reported 'damage to property and possessions', 'disruption to services' and each of the variables that measured 'psychological exposure', according to the reported level of water in the home. The five-point Likert scale for 'psychological exposure' variables was collapsed to two categories: (i) "much better", "better" or "the same", and (ii) "worse" or "much worse", as there were very few responses in the "much better", and "better" category or in the "much worse" category.

We first examined the univariate relationship between all the exposure variables and mental health symptoms. Multivariable logistic regression was then used to describe the association between water level in the home and mental health symptoms, adjusted for the following confounding variables: age (≤ 35 yrs, 36-50 yrs, 51 + yrs), sex, presence of an existing medical condition and employment status. We included 'area' and 'method of data collection' variables in the model because of the substantial differences in the socioeconomic status between study areas and variation in data collection methods. 'Psychological exposure' variables and 'incident management' variables (evacuation and disruption to essential services) were then added as explanatory variables to this model. All data analyses were carried out using STATA version 10 [20].

Response rates

We mailed 2,627 invitations in South Yorkshire and 8,915 invitations in Worcestershire; 2,265 people responded (38% in South Yorkshire and 14% in Worcestershire). In Worcestershire, the response rate to the postal survey was low at 12%, compared to 58% for face-to-face interviews. Among the respondents who were given the option of either a postal, telephone or web based questionnaire, 12% chose the web based questionnaire, 15% chose the telephone but the majority chose to return postal questionnaires (73%).

Characteristics of responders

The mean age of responders was 50 yrs (s.d.17 yrs) in South Yorkshire and 57 yrs (s.d. 17 yrs) in Worcestershire, The age range for the sample was 16 yrs to 96 yrs. The age profile of responders was similar to the age profile of the sample populations based on 2001 Census data, although people aged under 36 yrs were under-represented and people over 50 yrs were over-represented. A higher proportion of females responded to the surveys (72% in South Yorkshire and 57% in Worcestershire). According to data from the Office for National Statistics (ONS) [21] 52% of the population in the sample areas were female. At least half the respondents reported that they had lived in their area for at least 15 yrs. There was also a higher proportion of responders to the survey who reported they were unemployed (28% in South Yorkshire (ONS data: 20%), 39% in Worcestershire (ONS data: 15%)) or retired (24% in South Yorkshire (ONS data: 16%), 9% in Worcestershire (ONS data: 19%).

Associations between mental health symptoms, socio-demographics and water level in the home

The prevalence of all mental health symptoms was significantly higher among individuals who reported flood water in the home (psychological distress 69%, probable anxiety 48%, probable depression 43%, probable PTSD 22%) compared to individuals who did not (psychological distress 14%, probable anxiety 5%, probable depression 7%, probable PTSD 2%) (p < 0.01 for difference in proportions for each outcome).

Associations between mental health symptoms, 'psychological exposure' and 'incident management' variables

Following adjustment for the perceived impact of the floods and evacuation variables, the odds ratios for 'water above floor level in the home' were reduced to 5.0 (95% CI 3.4, 7.3) for psychological distress, 4.8 (95% CI 3.0, 7.8) for probable anxiety, 2.6 (95% CI 1.6, 4.3) for probable depression, and 3.9 (95% CI 1.9, 7.8) for probable PTSD.

Disruption to essential services was associated with odds ratios of 1.8 - 3.1 after adjustment. Evacuation was associated with psychological distress (OR 1.7 95% CI 1.2, 2.5), but not significantly with the other three mental health symptoms in the adjusted analysis.

Limitations of this study

We found that the risk of mental health symptoms was greater in South Yorkshire compared to Worcestershire. The nature of the floods was similar in these two areas but the population profiles were heterogeneous in terms of area level deprivation. South Yorkshire has a more deprived population compared to the areas surveyed in Worcestershire. Also the survey in South Yorkshire was carried out at three months after the flood compared to 6 months in Worcestershire, for logistical reasons. Further, we used different methods of data collection in conducting these surveys, which may have had different ascertainment biases. For this reason, we initially analysed the data separately for (i) South Yorkshire and Worcestershire, and (ii) data collected using face-to-face interviews and self-completed questionnaires in Worcestershire but found no differences in the associations detected and therefore we presented pooled analyses of these data in this paper, with adjustment for area and method of data collection.

Our study sample had a higher proportion of women, older people and unemployed people compared to the Office for National Statistics demographic profiles for these areas. Whilst these sample characteristics could lead to an overestimation of the prevalence of mental health symptoms in the survey areas, our main interest was in the relative rates of mental health symptoms between flooded and non-flooded groups, and we were able to adjust for these characteristics in multivariable analysis.

The response rates, although low, are typical of studies of this nature [22, 24, 28]. Though people with poorer mental health may have been more likely to respond than those who did not, it should be noted that the prevalence of psychological distress in individuals who were not affected by flooding was comparable to general population rates in the UK (10%-22%) [29]. In such cross-sectional studies we could not ascertain the extent to which pre-flood mental health affected post-flood mental health and individuals with pre-flood depressive symptoms may report greater increases of these symptoms post flood [30–32]. However we were able to adjust for self-reported history of health problems (respiratory, cardiac and mental health problems) prior to the floods although we were unable to validate these data against prescribing data or General Practitioner records so our results may be affected by under-reporting of mental health problems. Although we did not have data on individual measures of social class, which is a known risk factor for psychological distress, we were able to adjust for employment status (employed, unemployed, retired) in our analysis.