Abstract
Area-based funding under the first national mental health plan
The first national mental health plan in Australia encouraged redistribution of funds from institutional to community service settings [1]. At the point in the process of deinstitutionalization where closure of the major psychiatric stand-alone hospitals is possible, the decision arises as to where the institutional resources might best be reallocated in support of development of new community based services. The needs assessment perspective necessarily shifts from the dominant consideration of the needs of patients in institutions and the infrastructure needs of these facilities, towards a population health perspective, with consideration of where people with incident and prevalent disorders might live. Within the change management process to deal with this transition, the adoption of an area-based resource distribution formula is a rational step, and this indeed occurred in Victoria in the mid 1990s. One of us (PB) led development of the first such Victorian resource distribution formula (VRDF), proposed in 1994 [2] and the first two authors were members of a reference group regarding the formula through 1995–1996. The early conceptual history of the elaboration of the implemented formula has been reviewed in an earlier paper in this journal [3]. Since the time of these contributions, the VRDF has undergone further development [4, 5]. The VRDF through all its revisions has included a factor derived from the index of relative socioeconomic disadvantage (IRSED). The IRSED is a supplied variable by the Australian Bureau of Statistics (ABS) derived from the regular census [6]. The index measures the socioeconomic resources of areas based on the proportions of households or individuals with particular attributes (e.g. income levels, unemployment rates, etc.). Hence it is a variable reflecting the characteristics of an area of residence rather than an individual within an area. This variable influenced the final outputs from the formula by up to 20% away in one direction or another from mean per capita funding.
Beyond the use of the IRSED, additional elements of the VRDF included substantial adjustments for the demographic structure of the catchment area population, since service utilization rates were found to vary very substantially with age, sex and marital status. The formula included smaller adjustments for rurality, and the size of various catchment area subpopulations including indigenous people, homeless people, and people from non-English speaking backgrounds. The formula also provided for adjustment in consideration of private sector activity in each area, which might be performing functions substitutive for public sector inputs.
Area-based funding under the second national mental health plan
The second national mental health plan [7], while continuing the imperatives of the first plan in respect of serious and psychotic disorders, also explicitly mandates for a broadening of the range of disorders being treated. The VRDF is an example of a formula targeting the disparities to be expected in needs for care predominantly of psychotic disorders. In terms of implications for funding formulae, the second plan also prompts consideration of the perhaps different characteristics of a formula describing disorders with higher underlying prevalence such as depression and anxiety.
In contrast to the position in 1995–1997 when the VRDF was elaborated, there are now additional and significant sources of information on which to base responses to a number of the questions posed. The National Survey of Mental Health and Wellbeing (NSMHWB) was conducted in 1997 [8]. The data set from this survey includes some indication of IRSED scores of the areas of residence of survey participants.
Work from overseas has suggested that whereas indices like the IRSED may be valid for estimation of population needs generally [9–11], they may not function as well outside cities as they do in urban areas [12]. Epidemiological surveys in the USA, with similar designs to recent work carried out in Australia, have found that sociodemographic variables do not consistently influence disorder rates for all disorders. In this work from North America, it has generally been found that prevalence of anxiety disorders may be more strongly influenced by such variables than that of affective disorders [13]. The performance of the IRSED in predicting morbidity rates for high prevalence disorders, and its comparative performance in metropolitan and nonmetropolitan areas has not previously been investigated with epidemiological data regarding the general population in Australia.
Aims
This paper used the NSMHWB data sets to address some questions regarding resource distribution for mental health services in Australia at this time. Some of these questions arise from the development of the VRDF, and some of them are more related to the different imperatives presented to mental health services under the policy environment of the second national mental health plan. The specific questions we will address here are:
Do the data from the NSMHWB provide further useful information regarding age and sex specific rates for the major disorder groups, particularly those to be targeted under the second plan?
Do the data from the NSMHWB lend any empirical support to the use of IRSED across the whole of Australia, including urban, rural and remote areas?
Given the emphasis of the second national mental health plan on higher prevalence disorders including depression [7], what can the information from the survey tell us about the geographical distribution of these disorders, and of consumer assessed need relating to these disorders?
Methods
The methodology of this major epidemiological study has been described in detail elsewhere [8, 14–19] and so will not be repeated at any length here. Briefly, this survey was conducted in July to August 1997 with a sample of 10 641 people, being a 78.2% response on a clustered probability sample of Australian households. Trained interviewers used sections of the composite international diagnostic interview (CIDI) and other measures to collect the information on which diagnostic allocation and other assessments of need for care and service utilization were made. Data has been released by the ABS in the form of a series of progressively refined confidentialized unit record files (CURFs). Measures to ensure absolute confidentialization by the ABS include limiting the data available on area of residence, and stratification of continuous variables such as the IRSED.
Results
Overall results
Most readers will be familiar with the outputs of the NSMHWB in summary. The household survey [8, 14] found an overall one year prevalence of active disorders as assessed by the CIDI modules of 17.7%, within which are anxiety disorders 9.7%; affective disorders 5.8%; substance abuse disorders 7.7%; and comorbidity between at least 2 disorder groups being 4.8% of the population. The study of low prevalence disorders found the best estimate for the treated prevalence of all psychotic disorders to be 4.7 per 1000 [20].
Age, sex and rates of disorder
The Household survey component of the NSMHWB has provided a series of age and sex strata specific prevalence rates for each of the major groups of high prevalence disorders [15]. Hence age and sex stratified rates of disorder are now available for the Australian population which can be used to inform the targeting of services to these disorder types. We will not present these data in full here. Rather the point for the current argument is that these data are now relatively available, either in summary form in generally available publications [15] or through accessing the CURF [21]. Such strata specific rates for sections of the population can be used to refine the estimations of the effect of demographic variables on morbidity rates, and so support the development of variables for use in such formulae that can provide funding adjustment for different population structures. As regards low prevalence disorders there is also information available on the age and sex specific rates [20]. Although the low prevalence survey did not cover the Australian population in the manner of a population probability sample, it does provide information based on community prevalence rather than selected service utilization and so can also serve as an additional resource.
Socioeconomic disadvantage
The impact of the IRSED
Here we put the question: Does the NSMHWB data suggest that the IRSED is useful in a formula such as the VRDF, and is it useful in both metropolitan and non-metropolitan areas? Ideally this question could be addressed for both the high and low prevalence disorders. However, the low prevalence survey design was not such that this is possible, since the catchment areas selected were not representative of the dispersion of this variable across the country. We therefore address the question using the high prevalence disorder data set using simultaneous logistic regression conducted with the SUDAAN statistical package [22]. Design error is estimated through jackknife replication using replicate weights provided by ABS in the 2000 issue of the CURF [21]. In Table 1, the first column contains the results for metropolitan areas and in the second column are the results for those areas of Australia where the population is described as being resident in communities of below 100 000 inhabitants. The dependent variable here is the presence of any mental disorder.
IRSED and morbidity: metropolitan and non metropolitan areas Logistic regression with any mental disorder as dependent
The early reports based on the NSMHWB data (e.g. [15, 17]) concentrated on presentation of the results derived from the CIDI based examination of major morbidity categories. Later work [23] has come to draw on a wider range of diagnostic categories sampled in the survey, and here we have also taken a broader view of the classification of mental disorder. In addition to the CIDI diagnostic data (i.e. affective, anxiety, and substance abuse disorders), there are a number of other pointers to the presence of other possible diagnoses. The people over 65 years of age were engaged in the completion of the mini mental state examination (MMSE) and this yields a category of cognitive disorders. A screening instrument was used to ascertain possible histories of psychotic disorders. A series of inventory items were asked as provisional indicators of the presence of personality disorders. Thus ‘any disorder’ in this analysis included the presence of a positive indicator of any of the above listed diagnostic categories [23].
The independent variables measure the IRSED and demography of the population. The IRSED is here reflected in the quintiles of the distribution of the scores. The CURF does not contain raw scores for the IRSED, since the ABS confidentialization process precluded the release of such precise data. Instead the CURF contained a variable reflecting deciles of the IRSED, with the least disadvantaged two deciles grouped as a quintile. Our initial exploration of the data shows that consistently grouping all the data as quintiles retains the trends discernible in the decile level analyses, with the advantage of retaining greater cell sizes in source data for cross-tabulations.
Since the VRDF included adjustment for demographic variables including age and sex in a variable separate from that incorporating the IRSED, we control for age group and sex in this analysis, including these as additional independent variables.
The results in Table 1 confirm the progressive decline in overall prevalence with age, which has already been reported from the survey [14, 15, 24], and this is consistent across areas.
In metropolitan areas, there is less likelihood of the occurrence of any mental disorder for the least socioeconomically disadvantaged areas (odds ratio of 0.66), compared to the more disadvantaged areas. The intermediate IRSED quintiles have similar odds-ratios, but the results are not inconsistent with a progressive likelihood trend associated with the IRSED. For the non-metropolitan areas, the relationship is more complex. There is a substantial drop in prevalence of disorder in the 4th quintile (odds ratio of 0.59), but there is little difference between the other quintiles. This suggests that the use of the IRSED in linear combination as a predictor of psychiatric morbidity rates of any kind in non-metropolitan areas is probably invalid.
Place of residence and absolute disorder rates
While establishing associations through logistic regression analyses is an important strategy in investigation of possible causal associations, the preoccupation of planners is often with absolute prevalence rates in particular areas, and differentials between these. Hence in Table 2 we examine the effect of the IRSED on absolute disorder rates and on specific categories of disorders in both metropolitan and nonmetropolitan areas.
Area of residence, perceived need for mental health care and specific disorder groups
For metropolitan areas, in the case of any disorder the highest prevalence is in the most disadvantaged areas (25.7%), and the lowest is in the least disadvantaged areas (18.6%). A pairwise comparison between the most disadvantaged and the least disadvantaged quintiles, suggests a significant difference in overall prevalence of 7.1% (95% CI 4.0% to 10.2% p < 0.05). Examination of the specific disorder groups in metropolitan areas shows a reasonably progressive decline in prevalence with reducing disadvantage being accompanied in exact rank order by lower prevalence of affective and substance abuse disorders. For the anxiety disorders, rates are more stable in the metropolitan areas, except for the least disadvantaged areas where a substantially lower rate is observed.
Pairwise comparisons of prevalence between the most disadvantaged and the least disadvantaged quintiles are as follows: for affective disorders 1.8% (95% CI −0.2% to 3.8% ns); for anxiety disorders 2.6% (95% CI 0.2% to 5.0% p < 0.05); for substance abuse disorders 3.0% (95% CI 0.6% to 5.4% p < 0.05); for other disorders 4.6% (95% CI 2.2% to 7.0% p < 0.05). For anxiety disorders there is also a significant difference between the fourth and fifth quintile of 3.5% (95% CI 1.4% to 5.6% p < 0.05).
For the non-metropolitan areas the trend of prevalence of any disorder observed is apparently non linear, with the most disadvantaged quintile having the highest prevalence (24.2% and the fourth quintile having the lowest prevalence (16.3%). There was little difference in prevalence among the other three quintiles (ranging from 21.5% to 21.9%). A similar trend was observed for the prevalence of each of the disorder groups in non-metropolitan areas, with the lowest prevalence consistently occurring in the 4th quintile.
Place of residence and perceived need
The NSMHWB also provides for an additional analysis in terms of perceptions of the population to the meeting or otherwise of need [17, 19]. Hence, as another way of viewing the relationship between area and the experience of disorder in the population, we explore the percentages of perceived need in the final column of Table 2. Here perceived need is calculated when endorsed by people who used services and also by people who met subthreshold symptom criteria for anxiety, affective or substance misuse disorders. This definition is more inclusive than previous analyses, which have been restricted to those with service use or ICD-10 disorders only [17].
The pattern of perceived need is consistent with the prevalence of disorder in both metropolitan and non-metropolitan areas. In metropolitan areas, the highest perceived need (19.1%) is found in the most disadvantaged areas and the lowest perceived need (14.7%) in the least disadvantaged areas. In non-metropolitan areas, again a more complex and nonlinear trend is observed, with the least perceived need being in the fourth quintile. There is a significant difference between the percentage of the population with any perceived need in the lowest and highest quintiles in metropolitan areas, estimated as 4.4%. (95% CI 0.9% to 7.9% p < 0.05).
Overall disorder rates, and perceived need, findings for the IRSED
In summary, the NSMHWB suggests that overall rates of disorder vary in a reasonably progressive fashion with the IRSED in metropolitan areas but that this pattern is more complex in non-metropolitan areas. In metropolitan areas, there is some corroboration of the findings from the perceived need data, with a greater proportion of perceived need in the more disadvantaged metropolitan areas, but again, no clear trend in non-metropolitan areas.
Discussion
Limitations of the NSMHWB data sets
This paper has examined some of the information now available from the NSMHWB, which could inform further development of resource distribution formulae in Australia. The low prevalence studies [20] did not cover a sufficiently varied set of areas to provide for similar estimation exercises. Hence this examination is limited by the data set here examined, which only makes it feasible to estimate differentials in prevalence rates within the high prevalence disorder groups. An additional limitation is that the household survey CURF only included the IRSED of all the five socio-economic indicators for Australia (SEIFA) variables [25]. It is possible that other indices might perform better than the IRSED for this purpose. Furthermore, the stratified presentation of the IRSED in the CURF limits the analytical techniques we can apply to the current questions.
The NSMHWB was conducted within the confidentiality guidelines governing the work of the ABS, and was primarily designed to give accurate estimates of rates for the whole of Australia, rather than to answer specific questions around resource distribution. The most salient example of a survey designed with this specific aim in mind is the Colorado social health survey [9], and the sampling strategy and compilation of the data file were conducted very differently from that of the NSMHWB.
Findings for disorder rates and perceived need
In prediction of morbidity rates, the IRSED appears to be validated by the survey as a weighting of some use in metropolitan areas. However, the trends are different for different disorder groups. Rank ordering is most consistent and progressive for affective and substance misuse disorders. In contrast anxiety disorder rates are remarkably stable in the metropolitan areas, except for the least disadvantaged areas where a substantially lower rate is observed. This finding is at some variance with the findings of comparable USA studies [13]. In seeking to explain this we might consider the influences of social causation, social drift, and treatment as an agent in reduction of prevalence. The stepwise changes in prevalence for affective and substance misuse disorders might be compatible with social causes, however, we are unaware of any reason why social causes should differently affect prevalence of anxiety disorders over the other categories examined. In considering treatment, we know that there are effective treatments for anxiety disorders, but that the desirable first line interventions are predominantly psychological rather than pharmacological [26]. These are more dependent on specialist care for their delivery than are medications, and access to private specialist mental health services is not equitable across the range of classification of socioeconomic areas in this country [27, 28]. Considering public services, it is likely that the concentration of the first national mental health plan on ‘serious mental illness’ reduced the capacity of public mental health services to respond to the needs of those with nonpsychotic disorders [7].
Access to primary mental health care services shows less evident inequity than do specialist services [28]. Prescribing in primary care can treat depression in many cases reasonably effectively, and analyses of this survey data have shown accompanying high levels of meeting of perceived need for medication in this context [29]. In contrast, these analyses also show that perceived needs are less well met in primary care treatment for those with anxiety disorders. Perhaps the disparity of access to specialist services between socioeconomic areas impacts differently on the anxiety disorders.
In non-metropolitan areas the index appears to behave in a more complex, less linear and less easily interpretable way in predicting overall morbidity rates.
For perceived need, the rates within IRSED strata in metropolitan areas are compatible with substantial progressive proportional difference in the percentage of the population with perceived need in these differently characterized areas. As for disorder rates, the picture in nonmetropolitan areas does not suggest that the IRSED is particularly helpful as a proxy for rates of perceived need in these areas.
There is a limited ability to draw inference in respect of causation from these findings. The IRSED is a composite variable, and its components cannot be isolated in this analysis. The range of inclusions means that causation may in the individual case relate to any of a variety of factors. These may include the presence of stressors such as unemployment, poor housing, or being a single parent. Alternatively the effect might in part be ascribed to any of: social drift towards areas with cheaper housing by people with long-term disability, the absence of protective factors such as higher educational level, or perhaps relative disadvantage in access to services.
Conclusions
The findings of this investigation show that socioeconomic characteristics of area affect likelihood of psychiatric morbidity in the different major categories in different ways, and that these influences are more readily understandable in metropolitan than in non-metropolitan areas. This finding, along with other outcomes of the survey, can be important in providing national data and in contributing to the developing understanding of psychiatric epidemiology internationally.
Despite the substantial investment in the survey, it has limitations for some purposes and this investigation has exposed some of these. In respect of further refinement of understandings of geographical distributions of mental disorders in the population the survey is a relatively blunt instrument for investigation. The limitation of census variables it is possible to examine, and the stratification of those variables given the particular strategies adopted to ensure confidentiality, are two matters that influence the extent to which the CURF can be interrogated around resource distribution issues. Some of these could be overcome with further data extractions commissioned from the ABS, however, it would be necessary to seek additional funding for these given that the ABS contract for the NSMHWB has concluded following the delivery of a series of CURFs. Further data extraction might be worthwhile, however, there will remain limitations because the NSMHWB was not designed around this particular brief. For resource distribution decisions in Australia to be based on sound epidemiological survey data would probably require further large-scale survey work designed around this particular question, and with a different approach to securing confidentiality of the data.
While noting these limitations of currently available data sets, this investigation has revealed quite substantial and significant differences in experience of mental health between areas with different socioeconomic characteristics in Australian cities. While this country might pride itself as being a land of relative social equality and the ‘fair go’, in the case of mental health problems at least, being in identifiably socioeconomically disadvantaged urban areas of this country carries a significant risk of adverse mental health. The differences between areas are large enough to require systematic consideration in mental health services planning.
Footnotes
Acknowledgements
We acknowledge the Australian Commonwealth Department of Health and Aged Care, Mental Health Branch, who funded and otherwise supported the NSMHWB, and the Australian Bureau of Statistics who carried out data collection and preparation. The Commonwealth Department of Health and Aged Care provided funding to support some of the analyses described here through General Practice Evaluation Program grant 661.