Abstract
Are high prevalence disorders (anxiety, depressive and dwellers? It is generally assumed the answer to this substance use) more common among urban than rural is yes [1, 2] and that urban living is a risk factor for the development of psychiatric disorders, particularly depression and anxiety which are strongly affected by stress. Without recourse to examining the data, this view has been supported by caricatured images of urban and rural environments. Urban areas have been portrayed as environments of change, interpersonal estrangement, anonymity and conflicting values, seen as a focus of technological development and social dislocation, viewed by some as crime-ridden, riot-torn havens for individuals without strong loyalties and consistent values. By contrast, rural areas have been depicted as environments of social stability, integration, supportive interpersonal networks, and consensus about moral and political issues; and, in addition, as aesthetically superior to large cities [2, 3].
There are several compelling reasons for re-examining whether urban or rural residence is a risk factor for psychiatric disorder, and if so, why. First, several authors’ vision of cities as evil and rural areas as peaceful havens is based more on romanticized views than realistic assessments of the two environments [4–6]. Furthermore, it has been suggested that certain values which are considered typical of rural areas, such as strong conservative, religious and puritanical views, individualism, traditionalism, familiarism, fatalism, and person-centred relationships, may not beneficially affect the mental health of rural residents [7, 8].
Second, it is well recognized that the physical health of rural residents is poor when compared to their metropolitan counterparts [9]; for example, people living in capital cities have greater longevity than those living in other parts of Australia, and deaths from coronary heart disease, asthma and diabetes are higher in rural than in metropolitan areas (Table 1). Many of the factors identified as having a potentially negative impact on physical health may have a similar effect on mental health. These include geographical isolation and problems of access to care, shortage of health care providers and health care services, socioeconomic disadvantage and poor health-related behaviours [9, 10]. In addition, physical illnesses, particularly chronic disorders, are often accompanied by significant psychiatric comorbidity; for example, approximately 15% of medical inpatients have an anxiety disorder [11] and one in three have a major depressive disorder [12].
Health of urban and rural residents (source: Australian Institute of Health and Welfare [9])
Third, rural life has changed substantially over the past few decades. The traditional characteristics of rural life have been affected by a variety of factors including: globalization, economic restructuring and economic rationalization [13], movement of the younger generation to cities, free-flowing access to information (particularly via the internet), more ready access to urban settings through improved transportation and inward migration of people who commute to urban jobs, and relatively high rates of poverty and unemployment [5, 6].
Fourth, many of the studies cited by those claiming urban-rural differences were done many years ago [1]. Methodological changes over the past 20 years have enabled the conduct of new, so-called ‘third-generation’ studies characterized by improved diagnostic reliability and validity. Studies investigating the prevalence of psychiatric disorders in communities are generally grouped into first-generation or pre-World War II studies, secondgeneration studies (conducted between World War II and 1980), and third-generation or post-1980, studies [14]. Third-generation studies, conducted following the introduction of DSM-III, are characterized by the use of standardized diagnostic interviews, improved survey research design, and computerized data processing.
Finally, and perhaps most importantly, most studies examining rural/urban differences have simply focused on the ‘quantity’ question – are disorders more or less prevalent in urban than in rural settings. This is a crude and relatively uninteresting question, in which rural or urban are simply treated as a locations, that is, positions identified by grid coordinates; but rural and urban are places, that is, locations, in which social relations are constituted [15]. As place may have broad socioenvironmental influences on health, the more interesting question is the ‘quality’ question – how does place (rural or urban) influence mental health?
Here we review relevant studies, with an emphasis on key third-generation cross-sectional community studies which have compared urban and rural populations, and the three national household surveys conducted over the past two decades. These studies are described and their weaknesses are highlighted. The limitations of these studies point to the need for more sophisticated approaches to the epidemiology of mental disorders; in particular, studies need to go beyond the ‘one size fits all’ terms rural and urban, which assume location is the key issue, and examine the mix of economic, physical, social, environmental and socio-cultural factors within both rural and urban settings which may be important determinants of mental illness.
Cross-sectional community studies
First-generation studies relied on key informants and agency records to provide information to identify cases. Thus, these studies did not account for untreated cases of disorder. Second-generation studies used direct interviews of community residents and so were concerned with measuring the true prevalence of psychological symptoms or psychopathology in communities. Two different types of interview were used: one approach involved a personal interview by a psychiatrist to generate clinical diagnoses, but with the interview procedures not necessarily made explicit; in the other, standard and explicit data collection procedures were used and the classification of ‘caseness’ was made by evaluation of protocols compiled from the interview responses and sometimes also ancillary data, such as informants and information from previous records. A variation of this involved dispensing with clinical judgement and using objectively scored measures of psychopathology. These studies lacked content validity as there was little consensus at the time about the signs and symptoms to be elicited, different nomenclatures were used, and in many studies, diagnosis of a named disorder was bypassed for the general concept of caseness [16] and impairment [17]. A number of these studies documented the relatively high rate of certain types of impairment in rural areas, and though confined to these areas, and thus not comparative, they nevertheless provide important data [16, 18–20].
Two key second-generation studies have been widely examined and discussed and are of direct relevance to this paper. The Mid-town Manhattan study sampled 110 000 adult residents of a section of mid-town Manhattan [17]; the Stirling county study sampled 20 000 residents of a Canadian rural county [16], populations at the extreme of a commonsense definition of an urban-rural continuum. Both studies relied on nonclinician interviewers recording information on structured interview protocols, which were subsequently rated by a psychiatrist. While the two studies used different screening instruments, both had as their core a portion of the items from the Psychosomatic Scale of the Neuropsychiatric Screening Adjunct [21]. Srole [6] roughly matched the two studies on major demographic characteristics (race, nativity, age, sex and socioeconomic class) and found the current prevalence of ‘psychiatric caseness’ was significantly higher in Stirling county than in mid-town Manhattan. This finding, which is contrary to many studies investigating urban/rural differences, while of interest, must be viewed with caution. Matching on these variables may have washed out any real differences between the two locales. Furthermore, a variety of problematic assumptions were made by Srole, particularly the assumption that the way in which ‘caseness’ was derived in the two studies was equivalent [22]. There are also difficulties with interpretation of the Manhattan data, given the exceptionally high rates of psychiatric morbidity identified. Nevertheless, it raises the possibility that disorders in rural areas may be more prevalent than is often assumed.
The National Institute of Mental Health (NIMH) epidemiologic catchment area survey (ECA) [23] was the first of the third-generation studies. The study was characterized by its sample size of at least 3500 subjects per site (about 20 000 total), the focus on diagnostic interview Schedule (DIS)-defined DSM-III [24] mental disorders, one-year interview-based longitudinal design to obtain incidence and service use data, the linkage of epidemiologic and health service use data, and the replication of design and method in multiple sites.
Two of the five ECA sites (Durham and St Louis) had sufficient populations to examine rural-urban differences. In the Durham site there were four rural counties contiguous to a small central city; most rural residents were farmers or in local retail or trade services. In the St Louis site, the rural sample was drawn from two rural counties contiguous to the (larger) metropolitan centre. Rural residents in these areas were less likely to be farmers and more likely to commute to the urban area for employment opportunities.
As shown in Table 2, perhaps not surprisingly given the within-rural differences, urban-rural comparisons at these two sites gave results that were not consistent across diagnostic categories [23, 25].
Urban/rural differences in prevalence rates: epidemiologic catchment area study (St. Louis and Durham sites) [23]
Subsequent to publication of the initial findings from the Durham site [23], further analysis using logistic regression was undertaken to examine the effects of urban/rural residence on major depression, controlling for demographic and social characteristics that previous literature indicated are risk factors for major depression (age, sex, race, marital status, education, socio-economic status, stressful life events, mother with small children, availability of confidante) [3]. Current major depression was nearly three times more common in the urban than in the rural communities (OR 2.99; p < 0.01). Rural residence decreased the risk of major depression for some but not all demographic subgroups: the risk of major depression was decreased for young rural residents (aged 44 years and younger) compared to their urban aged peers, and rural residence was more protective for young women than for young men.
It should be noted that this study treated depression as a stand-alone disorder, when in fact individuals may have carried other additional high prevalence disorders, for example, substance misuse disorders and anxiety disorders. Given this likely comorbidity, it is important to consider that the predictor set for depression may have been as powerful (or even more powerful) in predicting the additional disorders. A second point at issue is that the authors of the study did not analyse interaction effects but treated the independent variables as main effects. While the authors acknowledge this problem, the fact is that interaction effects may ‘trump’ the main effects, and be more illuminating. For example, consider the three–way interaction of age, sex, and rurality. Could it not be that the risk for depression might be synergistically increased for young urban males? Understanding of urban-rural differences would therefore be advanced by the investigation of interaction effects. Of course larger sample sizes (in the order of 10 000) would be needed to identify 2- and 3-way interactions involving high-prevalence disorders. Such future investigations would also be facilitated by using continuous (quasi-normally distributed) variables in place of categorical measures.
In considering these findings, in addition to the points made above, the particular features of this sample should be noted. The sample had age and sex distributions characteristic of both North Carolina and the USA generally, but a large proportion of non-white (41% rural, 32% urban) subjects, most of whom were black. Durham county is a major metropolitan centre in an area that contains more than 500 000 people and three major universities and a large industrial park that is the site of the research facilities for a number of major industrial firms. Approximately 24% of the population had at least a college education, and the mean per capita family income was over $20 000, reinforcing that this is not a typical urban setting. By contrast, the four rural counties are most representative of the rural south, and as such are characterized by a higher percentage of blacks and persons of lower educational attainment and socioeconomic status than comparable national figures.
A variety of other studies conducted over the past 20 years have investigated the prevalence of psychiatric disorders in rural as opposed to urban communities [26–35]. The findings of several of these studies should be viewed with caution. Few studies have used reliable diagnostic processes to define specific disorders. In addition, urban/rural comparisons have often been made across rather than within studies. This approach, which does not take into account regional, cultural, or methodological variations between studies, is likely to obscure any urban/rural differences which may be present. Very few studies have taken compositional differences between urban and rural areas into account; this is particularly important when investigating those disorders in which there are age/sex differences in prevalence. If urban/rural residence is a risk factor for psychiatric morbidity, that relationship should be observed once compositional differences (e.g. age, sex, socioeconomic status) between urban and rural settings are statistically controlled [3]. Furthermore, reliance on cross-sectional data, unless accompanied by data about migration history, means it is not possible to know if current urban or rural residence precedes or follows the onset of disorder. The impact of psychiatric disorder on location of residence has been inadequately researched. While greater accessibility of services in urban regions might tend to attract individuals suffering psychiatric disorders [36–38], it might equally be hypothesized that more vulnerable or less resourceful individuals gravitate to less competitive rural environments. In addition, it is well recognized that in Western countries there has been an outflow of lowincome households, a group at increased risk of mental illness, from the major capital cities to rural areas [39].
Studies which have used reliable diagnostic processes and made within-study urban/rural comparisons are shown in Table 3. No significant urban/rural difference in the rate of depression was found in any of these studies [30, 33, 34]; one study found alcohol dependence, agoraphobia and panic disorder were more common in rural than urban residents, while antisocial personality disorder was more common in urban dwellers [30].
Key third-generation studies assessing urban-rural differences in ‘high prevalence’ disorders
Lee and colleagues [30] provided no data regarding the composition of the sample, and no examination of possible differences due to age/sex/socioeconomic status and other differences between the rural and urban samples (see Lee et al. [40] for methodology). By contrast, the other two studies also investigated possible differences in rates of a variety of known risk factors for anxiety and depression [33, 34], which may have confounded any findings with respect to urban/rural differences.
Parikh and colleagues compared socio-demographic characteristics of their urban and rural samples (age, sex, education, household income, marital status, employment) and found the urban group had higher educational achievement, higher employment, higher family income, and a lower rate of being married [33]. When demographic information on those with affective disorders was examined, similar risk factors seemed to operate in both rural and urban settings. Both urban and rural mood disorder respondents were more likely to be poor, unemployed, female and unmarried compared to the rest of the population (all p < 0.01).
Romans-Clarkson and colleagues [34] conducted a two-phase study using questionnaires and face-to-face interviews to examine sociodemographic risk factors for psychiatric morbidity in urban and rural women. The researchers used weighted logistic regression to assess simultaneous effects of age, marital status, social class, employment status, and urban/rural residence on total present state examination (PSE) score. Age was associated with PSE caseness for urban but not rural women, but only for women aged 65 years and older; for both urban and rural women high PSE case rates were found in separated women and those who had never been married, and those of lower socio-economic class. Paid employment was associated with a lower case rate for urban women but not for rural women. Factor analysis was used to reduce the data to a manageable number of relatively independent variables. Identified factors were then entered into a regression analysis using the total PSE score as the dependent variable. Three factors (selfrelated alcohol problems, childhood sexual abuse, social networks) individually explained a significant amount of the variance in total PSE score; the same three factors applied to both the urban and rural data sets.
Subsequently, Romans-Clarkson and colleagues reported a more detailed examination of the association between psychiatric morbidity in women and social interaction in the two geographical locations [41]. Using the Interview Schedule for Social Interaction (ISSI) [42], the researchers examined availability and adequacy of attachment, and availability and adequacy of social integration. Women who described less available and less adequate attachments and social integration showed more psychiatric morbidity. Rural women had higher adequacy of attachment and social integration scores than did urban women, but as noted above they did not have less psychiatric disorders.
Each of the studies described above examined a particular rural and a particular urban setting, each with its own particular social, cultural, economic and demographic characteristics; however, each simply talks of urban/rural differences, assuming each is representative of whatever ‘urban’ and ‘rural’ are. None attempted to capture the specific features of the setting (e.g. community in growth/decline, community cohesiveness, community attractiveness, lay systems of beliefs and behaviours, services provided) which might be of particular relevance in determining risk for the development of psychiatric disorder.
In summary, although many studies have been reported, few fulfil basic methodological requirements such as the use of reliable diagnostic procedures to define disorders, designs which enable rural/urban comparisons within the same study, examination of interaction effects, and measures to control for compositional differences. Where the latter have been examined, it seems that more often than not, demographic and social characteristics previously found to be risk factors for depression are more powerful predictors of depression than place of residence.
National household surveys
Three genuine national household surveys have now been completed around the world, and the details of these are outlined in Table 4. These allow more definitive statements about nationwide psychiatric morbidity and also about regional and socio-demographic variation. However, the ability of such surveys to fulfil the latter function (examine regional and socio-demographic variation) is much more limited than is generally acknowledged. A ‘one size fits all’ construct of urban and rural focusing on location rather than place can only answer the quantity not quality question. Furthermore, the presentation of findings in aggregated de-identified form means any intra-area differences will be averaged out and disparate towns and cities grouped together, so providing little or no information on patterns of illness in either rural or urban areas [43].
National household surveys
The 1990 US national comorbidity survey (NCS) included 8098 subjects aged 15 to 54 years to form a probability sample of 48 contiguous states [44]. Urbanicity was examined at the county level by distinguishing major metropolitan counties (major metropolitan areas), urbanized counties that are not in the major metropolitan areas (other urban areas) and rural counties (rural). The effects of urbanicity at county level were generally not significant. The one exception noted was that residents of major metropolitan counties were more likely than residents of rural counties to have comorbidity in the 12 months before interview. This pattern was thought to reflect a low rate of comorbidity in rural America rather than a high rate in major metropolitan counties.
The 1993 UK Household Survey of the national morbidity survey of Great Britain used a postcode address file sampling frame to identify adults 16 to 64 years old living in all the areas of England, Scotland and Wales except the highlands and islands of Scotland [45]. In total, 9777 subjects were interviewed. High prevalence disorders were assessed using the Revised Clinical Interview Schedule (CIS-R) [46] leading to ICD-10 [47] diagnostic categories, and questions about alcohol and drug misuse and dependence using quantity/frequency questions from the regular national surveys of alcohol and tobacco consumption. Those scoring greater than or equal to 12 on the CIS-R were regarding as suffering from a neurotic disorder while alcohol and drug dependence was defined according to responses to questions on quantity/frequency of use. Higher rates of psychiatric morbidity (CIS-R score ≥ 12), and alcohol dependence and drug dependence, were found in urban compared to rural areas; semirural areas were generally intermediate (Table 5).
National morbidity survey of great britain: prevalence of disorder by urban–rural residence [45]
Despite including a large sample, the study treated each of the three diagnostic categories as independent of one another. Logistic regressions were run for each diagnostic category, ignoring likely comorbidity. Naturally, an individual may have been positive for one, two or three of the categories. Predictor sets may well have had different power according to the number as well as type of category endorsed.
The authors looked at main effects and also at simple two–way interactions with rurality in predicting diagnostic category, but they failed to look at three-way or more complex interaction terms. Such complex interactions could potentially be more predictive of diagnostic category. For example, a young, poorly educated and unemployed male living in a rural area is likely be at much more risk of a mental disorder than an older financially secure married male who has moved to the country to grow grapes.
The 1997 Australian national survey of mental health and wellbeing surveyed adults aged 18 to 99 years who were identified in a cluster sample of households selected so the result would be representative of the entire Australian population [48]. No urban/rural differences were found for affective disorders, anxiety disorders, alcohol or drug dependence [49]. There was, however, a gender–urbanicity interaction: for males, the rate of mental disorder (sum of anxiety, affective and substance use disorders) was slightly greater for those living in capital cities (17.5% vs 17.1%) while for females the rate was greater for those living outside capital cities (18.9% vs 17.5%). It is to be hoped that further fine-grained analyses of this comprehensive data set are forthcoming.
Each of these surveys aimed to answer three primary questions: how prevalent are mental disorders; how disabling are these disorders; and which health services are used? Thus it is perhaps not surprising, albeit disappointing, that only one of these studies has published a detailed analysis of possible urban/rural difference [45]. In the UK survey, urban subjects, when compared to rural subjects, were significantly younger, not currently married, of lower social class, non-white, less welleducated, living in flats or non-detached houses – a lower proportion of which were owned outright. Semirural subjects tended to be intermediate. The proportion of subjects employed full time was similar in rural and urban areas, but overall employment was higher in rural areas due to more part-time work. Urban residents were more likely to have experienced a stressful life event in the last year, to perceive themselves as lacking in support, and to have a small primary support group.
When these social differences were taken into account, the urban/rural differences in psychiatric morbidity, alcohol and drug dependence were reduced, and those with alcohol and drug dependence were no longer significant. The relationship between psychiatric morbidity (CIS-R ≥ 12) and urban/rural residence was examined by logistic regressions to determine the extent to which it was due to the social differences between urban and rural settings. In a logistic regression using urban/rural residence alone, the effect was highly significant overall with odds ratios (OR) for urban residents compared to rural of 1.63 (p < 0.001) and for semirural compared with rural of 1.22 (NS). Pairwise analyses conducted using two independent variables, area of residence and one other social or demographic factor in turn (age, sex, employment status, any life event in the last year, primary support group size, perceived level of social support, tenure of housing, social class, educational level, ethnicity, accommodation type, marital status) revealed the relationship with the area of residence was largely unaffected by any of these. A multiple logistic regression analysis incorporating all the independent variables entered simultaneously, reduced the effect of area of residence but it remained significant (p < 0.05). The strongest independent effect was due to occurrence of any life event in the last year (OR 2.50, p < 0.001), followed by size of primary support group, sex, marital status, perceived social support and employment status.
Similar analyses were conducted for alcohol and drug dependence. For alcohol there was a significant effect of area of residence analysed alone (OR 1.6, p < 0.02 for urban vs. rural, 1.06 NS for semirural vs. rural). In the multiple logistic regression analysis employing all the independent variables the effect of urban/rural residence was not significant. The highest relationships with alcohol dependence were for sex, age, marital status and life events. Findings for drug dependence were similar. There was a significant effect for area of residence (OR 2.31, p < 0.05 for urban versus rural, 1.31 NS for semirural vs rural). In the multiple logistic regression employing all independent variables, the effect of area of residence fell below significance. The highest relationships were with age, life events, marital status, unemployment and sex.
In summary, two of the three national household surveys found few differences in prevalence rates between urban and rural residents, while the third (the UK survey) found higher rates of disorder in urban areas. Importantly however, none of the studies were specifically constructed to investigate rural/urban differences. Of note, urban residents in the UK survey were younger, not currently married, of lower social class, less welleducated, more often had experienced a stressful life event and felt less supported. When these factors were taken into account statistically, urban-rural differences remained only for neurotic disorder. However, statistical models are unlikely to be able to unravel and account for the contagion of stressors which may coalesce to produce psychiatric disorder, leaving the question of whether place of residence or sociodemographic factors truly accounted for urban/rural differences unanswered.
Discussion
Contrary to popular belief that urban living is a risk factor for psychiatric disorder, there appears to be limited data to support the popular view that ‘high prevalence’ disorders are more common in urban residents. Studies examining possible urban/rural differences in prevalence suggest other factors are more powerful than location of residence. These include poverty, unemployment, being female, not being married [33], lower socioeconomic class, women who were separated or never married, self-related alcohol problems, history of childhood sexual abuse, poor social networks [34], life event in previous 12 months, size of primary support group, marital status, low perceived social support, employment status, and sex [45].
The failure to demonstrate a difference in prevalence of disorder between urban and rural settings should not be a surprise. The reliance on rural or urban as units of analysis can have the effect of averaging out differences between communities which are likely to be highly variable, and obscure the localization of factors possibly contributing to mental ill health such as poverty, deprivation [43] and a variety of health-promoting and healthdamaging behaviours.
Studying the prevalence of disorder in rural versus urban environments is of interest; urban or rural residence is an obvious social characteristic to consider as a risk factor for psychiatric morbidity and assessing the ‘quantity’ question is an important first step. However, the more important and informative question is what is the nature of any differences or what issues are specific to, or especially important in, the rural setting? One set of variables will most likely relate to place, and in evaluating these, heterogeneity within urban and rural settings must be recognized – one size does not fit all. In addition, individual differences between people need to be taken into account. This interaction will be further compounded by the presence or absence of more specific risk factors such as traumatic life events, substance misuse, stress and significant life changes. These interactions are likely to occur in both rural and urban settings.
Is there a need to further examine urban/rural differences in high prevalence disorders? We believe there is, but that the question asked should be – which factors operative in each environment are likely to be important in the development of psychiatric disorder and to be somewhat unique to that environment? Thus, further studies of prevalence need to go beyond the crude variable of ‘rural’ or ‘urban’ and to examine both settings at a more micro level. Rather than simply comparing urban and rural, studies should be directed to identifying particular groups in both settings whose risk of illness is increased. It is likely that the within-group differences will exceed the ‘averaged out’ differences between a ‘one size fits all’ view of rural versus urban location.
Studies conducted to date have essentially treated risk factors as main effects. In no case did any one study examine more than two-way interactions. Clinicians are very aware than there are a variety of risk factors that interact synergistically to contribute to the development of disorder(s) in any one individual. Future studies need to more closely mirror the real situation with respect to patients in need of care and take into account a variety of individual- and community-based risk factors. Also, given that comorbidity is the rule not the exception, future studies also need to take into account in their data analyses the fact that a person may meet criteria for more than one diagnosis.
What benefits will flow from such studies? Understanding how rural or urban place contributes to the development of psychiatric morbidity has obvious implications for efforts directed towards prevention and early intervention approaches. While the effect of ‘rurality’ on prevalence of mental illness is unclear, there is clear evidence that residence in a rural location significantly influences people's behaviour with respect to how they address their needs for health care services [50, 51]. Accurate data regarding prevalence and the factors contributing to this are required to inform service delivery, and to identify the groups of people with the greatest need for treatment. Obtaining such data should be a priority for researchers, policy makers and clinicians in terms of both risk factors for mental illness and access to care.