Background and Significance

State mental health agencies (SMHAs) administer more than $27 billion dollars annually,1 approximately 32% of the $85.4 billion estimated total mental health care expenditures in 2001.2 Although there is a wide variety of organizational structures and funding mechanisms within SMHAs,1 these agencies are required by law to present sub-state estimates of severe mental illness in their applications for block grant funds.3 Although the number of people with severe mental illness may be a reasonable measure of “need,” Aoun et al.4 have shown that there is no one “best” type of population-needs assessment for mental health care, as assessments are used not only by healthcare professionals for planning services; but also by policy makers for monitoring and promoting equity. In addition to multiple definitions of need, there are also different methodologies for assessing the need for mental health care, including: use of routine data, consumer surveys, modeling need for services, and surveys of the general population.4 Spatial analysis is one method of examining need and utilization.5

Medicaid and indigents

As public mental health agencies consider need, finances are also important. Indeed, there is a complicated relationship between Medicaid funding and mental health care for low-income persons. Although Frank and Glied2 have suggested that Medicaid funding has allowed states to spend more on total mental health care, especially care outside the purview of SMHAs, Davoli6 has argued that federal Medicaid policies have worsened mental health treatment for indigent persons with severe mental illness, primarily due to funding restrictions on state mental hospitals.7 It does appear that, as many states and counties shift more of their mental health budgets onto Medicaid, there are fewer funds to provide services for low-income, uninsured people with serious mental illness who are not eligible for Medicaid.8 As an example, during the 1990s in California, the number of Medi-Cal (Medicaid) clients served by county mental health increased by 131%, while the number of indigents decreased by 8%.9

Surveys from the nation-wide Community Tracking Study (CTS) found that low-income persons with serious mental illness who are not eligible for Medicaid continue to be a concern among public mental health providers, with many believing that reductions in mental-health services have led to higher levels of homelessness, incarceration, and emergency department visits.8 Within California, a majority of emergency department directors believe that limited mental health treatment resources for uninsured persons is a significant factor in increased emergency room visits related to suicide attempts.10

Information needed within public mental health

There has been limited research into what types of information would be most helpful in best managing public mental health systems;11 but it is clear that the format for much of the currently available information is dictated by state and federal funding sources. Although SMHAs receive state and local funding and federal block grants, Medicaid currently provides more than half of the funding and is expected to account for two thirds of funding by 2017.7 Not surprisingly, Buck7 has shown numerous changes in data systems, performance measurement, planning and reporting requirements, as states move from a “community model” to a “health plan” model of state mental health services resulting from increased Medicaid funding.

Regardless of their motivation and autonomy, SMHAs are obligated to use a variety of data to best manage their system and perform activities such as rate setting (allocation of resources), provider profiling, and system-level monitoring, as discussed by Tompkins and Perloff11 in their review of programs in Arizona. State mental health agencies have generally not taken full advantage of geographic information systems (GIS), which is a combination of computer hardware, software, data, and expertise. Geographic information systems have been used for many years to “link diverse layers of population and environmental information to characterize the many dimensions of health care need for small areas.”5 Beyond analyzing need, GIS is often used to analyze access to care and geographic variation in utilization.5 An example of GIS being used within a planning process is given by the New York State Office of Mental Health,12 and geographic analysis of change in Medicaid-funded psychotropic prescription patterns has been demonstrated in Michigan and New York.13 Despite limited application within SMHAs, the application of geography to mental health services began in the late 1800s with the discovery of “Javis’s Law,” i.e., that distance from a mental health hospital predicts utilization, and there have been many studies since then looking at the location of delivery sites.14

Hospitalization and GIS

Notwithstanding the decades-long trend away from inpatient psychiatric treatment, there were an estimated 2.3 million psychiatric discharges from non-Federal short-stay hospitals in 2004, accounting for 6.6% of all discharges that year.15 Not surprisingly, there continues to be research into this expensive form of treatment.16–18 Although socio-economic status (SES) measures may account for nearly 50% of the variation in admission rates between locations,19 with poverty, social isolation and social disorganization likely being the constructs linking area-based low SES measures with increased psychiatric utilization,20 it appears that acute psychiatric hospitalizations also reflect on the mental health care system as a whole, including the network of outpatient care and support services, overall funding, and hospital management.18 Thus, there are public administrators in all 50 states with reason to be interested in small-area variation in hospitalization rates. Within California alone, there are 56 county-level mental health plans (and two plans for a handful of cities).

Given the financial challenges associated with lack of health insurance, there is a practical need to conduct small-area needs assessments regarding low-income, uninsured populations with mental illness, particularly if targeted efforts could reduce unnecessary hospitalizations. Such a practical need also dovetails with a challenge for mental health geographers to place greater emphasis on social justice, given the ‘displacement’ service model in many urban areas which churns very poor people between temporary housing, psychiatric wards, jails, prisons, and out-of-home placement facilities.21 However, just as there are multiple definitions of need, there are different definitions of “small-area” and there is not a consensus as to which level of geography is “best.”22,23 Some researchers suggest that health disparities are best examined at the census tract level.23,24 Almog and Curtis18 have successfully used Zip-Code-level data in New York City analysis, though it has been acknowledged that Zip Codes may be problematic as their boundaries are based on mail delivery routes and may not correspond well to the natural boundaries of demographic or socioeconomic characteristics of the underlying population.25

This study applied GIS techniques to existing data to present a method of needs assessment which may provide useful information for both public mental health managers and those interested in health disparities. The specific objectives were to (1) clearly present small-area variations in the proportion of psychiatric discharges in which patients had no other source of payment, i.e., were indigent, and (2) compare those areas of increased (or decreased) indigent psychiatric discharges to areas estimated to have increased (or decreased) poverty among those with severe mental illness. This exploratory study used publicly available data from California hospitals rather than data directly from the California Department of Mental Health (DMH). Although this study is primarily concerned about indigent psychiatric care in the context of public funding, it has been suggested that studies of access for low-income persons do not have to be restricted only to the perspective of the overall public mental health system; it may be useful to examine data focusing on psychiatric hospitals,26 as those hospitals may cost-shift to absorb treatment costs for indigents.

Methods/Analysis

Data

Public-use hospital discharge data for calendar years 1999–2003 were obtained from the California Office of Statewide Health Planning and Development (OSHPD; http://www.oshpd.ca.gov/HQAD/PatientLevel/index.htm). Information is included for all discharges from licensed acute general care and psychiatric care hospitals, excluding state mental health hospitals and Department of Veteran’s Affairs Medical Centers (which are not mandated to report to OSHPD). The statistical program SAS 8.1 (SAS Institute, Gary, NC, USA) was used to extract records having a primary diagnosis of schizophrenia, other psychosis, affective disorders, or anxiety disorders. The 24 secondary diagnosis fields were also examined to determine if there were co-occurring substance use/abuse related conditions. The resulting file, containing patient Zip Code (five digits), primary payer source, age category, gender, race, ethnicity and hospital license type (general acute or acute psychiatric) was then geocoded, or spatially linked, to a national Zip Code layer using ArcView 9.1 software (ESRI, Redlands, CA). Only records for Zip Codes within California were retained for further analysis. In the public-use data, OSHPD follows a protocol of masking age, gender, ethnicity, race and Zip Code values to ensure patient confidentiality.

Descriptive spatial mapping

A GIS-based, multi-method approach developed to explore health disparities27 was used to analyze the hospital discharge data. Smoothed or spatially filtered maps were first created to represent the continuous distribution of indigent psychiatric discharges as naturally occurring phenomena, without the artificial administrative boundary constraints of Zip Codes. Using the Spatial Analyst extension of ArcGIS 9.1, an adaptive kernel approach27,28 was used to display areas with higher or lower ratios of indigent discharges.

As opposed to conventional spatial smoothing, the adaptive kernel method implements spatial filters (or kernels) defined in terms of near constant population size rather than constant geographic size, i.e., greater emphasis is placed on population size rather than geographic size. A standardized risk ratio (SRR) was calculated for approximately 120,000 grid point evenly placed over a map of California. The GIS program searched through the adjacent ZIP Codes until at least 500 psychiatric discharges (set a priori)27 were associated with each grid point. The SRR at each point was calculated by dividing the number of indigent discharges within the disc centered at the corresponding grid point by the number of psychiatric discharges, divided by the California proportion of indigent to all psychiatric discharges. To take into account sociodemographic differences among locations, smoothed SRRs were adjusted for differences in age, gender, race, ethnicity, diagnosis, and license status of the hospital.

Inferential spatial analysis

A circular spatial scan statistic (SCS)29 was used to find contiguous areas, or clusters, having a statistically significantly higher or lower standardized mortality ratio. The spatial scan statistic is a robust test of spatial randomness that determines the location of any area of California representing a probable cluster of indigent psychiatric discharges, after adjusting for multiple testing inherent in many possible locations and sizes of the area.30 The growing application of SCS within public health for a variety of issues, such as analysis of organ transplantation patterns27 and detection of cancer clusters31, offers promise for its effective use in health services research. Calculations were carried out using free SaTScan 6.1 software (developed by Martin Kulldorff and available at www.satscan.org).

As with the descriptive mapping, a 2 × 2 km lattice of nearly 120,000 observational locations was overlayed on the state. For each grid point, the radius of the scan window was set to contain at most 30% of the total number of psychiatric discharges. By searching for clusters without specifying their size or location, the method circumvents pre-selection bias. At each grid point, the SCS provides a measure of whether the observed number of indigent discharges is unlikely for a window of that size, using reference values for the entire state. The null spatial model was based on a Poisson point process model and hypotheses were evaluated using a maximum likelihood ratio test, whose P value was obtained through Monte Carlo randomization.32 If there was a cluster, the spatial location and extent was extracted and mapped. Analyses were run unadjusted and adjusted for age, gender, race, ethnicity, diagnosis, and license status of the hospital.

Census-based synthetic estimation

The population-based estimates of persons with severe mental illness were developed using the Census Bureau’s Public Use Microdata Sample (PUMS) 5% state file,33 based on the 1-in-20 census responses. These files provide a cross-tabulation of the number of individuals within each PUMA (minimum population of 100,000), broken out by gender, age group, race/ethnicity, marital status, education, poverty status, and residence status. This detailed aggregated data is particularly useful for examining geographic patterns of household-level data.34 Note that PUMS refers to the data, while PUMA refers to the geographic area.

Holzer35 was one of the first to combine survey data with local socioeconomic data, initially providing estimates of the number of persons with mental illness in Texas. Starting with PUMS data, a process of synthetic estimation35 was used to calculate both the total number of persons with severe mental illness and the number of persons living in households less than 200% of federal poverty level (FPL) having severe mental illness. This process required the development of a model based upon the relationships between socio-demographic characteristics and mental health prevalence drawn from the National Comorbidity Study (NCS).35 Basically, the known prevalence of severe mental illness within each socio-economic subgroup, based on the NCS, was multiplied times the number of residents in the corresponding subgroup. After sufficient iterations, estimated numbers of individuals with severe mental illness were obtained within each age, race/ethnicity, education, marital status, and poverty status category. For this study, the number of householders at less than 200% FPL estimated to have a severe mental illness was divided by the total estimated number of persons with severe mental illness to obtain the estimated amount of poverty among those with mental illness in each PUMA. The results were geocoded to a Census TIGER shapefile for 5% PUMAs in California.36

Within California, there are 233 five percent PUMAs, which allows for greater spatial precision than the 1% PUMS (1-in-100 census responses) used for the initial county-level prevalence estimates of severe mental illness prepared for the California Department of Mental Health. It is noteworthy that these county-level California prevalence estimates were later updated for 2004 and 2006, continue to be posted on the DMH website,37 and are still used for planning and reporting purposes.

Results

The independent variables for both the indigent and non-indigent geocoded discharges are shown in Table 1. Approximately 91% of all identified psychiatric discharges were successfully geocoded to locations in California, with 3.2% of all psychiatric discharges formally identified as homeless and 5.7% of psychiatric discharges having no known address, partial information, or an out of state residence. Discharges which were not geocoded were significantly more likely to have missing or unknown values for age, gender, race and ethnicity.

Table 1 California psychiatric inpatient discharges, 1999–2003
Full size table

Of the 684,408 geocoded psychiatric discharges, 31,936 (4.7%) were indigent, compared to 7.3% among discharges that were not geocoded (data not shown). There was a 92% increase in the number of geocoded psychiatric discharges between 1999 and 2003, jumping from 82,144 to 158,123. The proportion of psychiatric indigents more than doubled, jumping from 1,781 discharges in 1999 (2.2% of geocoded discharges) to 8,613 discharges in 2003 (5.4%). Meanwhile, the percentage of all indigent discharges in California (data not shown) decreased slightly, from 2.04% of 3,416,716 acute general-care discharges (excluding psychiatric hospitals) in 1999 to 1.88% of 3,645,998 acute discharges in 2003.

Nearly all psychiatric discharges (95.8%) were from specialty psychiatric hospitals. Among the non-indigents, Medicare accounted for 31% of discharges, Medi-Cal for 29%, and private coverage for 27%. The public-use dataset had many unknown values for gender (36.6%), age (12.4%), race (45.6%), and ethnicity (48.4%). As the percent of unknown values was similar for both the indigents and non-indigents, it may be reasonable to assume that the unknown values did not bias the adjusting process in the cluster analysis. Indigents were more likely to be 18–34 years of age (44.2 vs 25.3% among those with known ages), male (54.3 vs 46.8%) and Hispanic (23.3 vs 13.8%). They were also more likely to have a diagnosis of “other psychosis” (22 vs 11%).

Table 2 summarizes data by level of geography, for both the 2,189 Zip Codes and the 233 PUMAs. There was great variability, particularly among the discharge data, with one Zip Code reporting only one psychiatric discharge in 5 years and another nearly 4,000. Furthermore, nearly 28% of Zip Codes had fewer than 11 psychiatric discharges and 37% had no indigent psychiatric discharges. A mean of 2,993 persons were estimated to have severe mental illness within each PUMA (range 583 to 9,241). On average, 45% of all persons with a severe mental illness lived in a household at less than 200% of the federal poverty level (range of 7 to 86%).

Table 2 Table 2 Summary of geographic data
Full size table

Spatial patterns

Table 3 presents the results of the cluster detection analysis. Approximately 56% of all indigent discharges were selected to be in a cluster. In the unadjusted analysis, simply looking at indigent versus non-indigent discharges, there were seven clusters (four with significantly increased rates compared to the state-wide average and three with lower rates). All of the clusters with significantly lower rates occurred along the Pacific coast—West Los Angeles, Orange County, and San Francisco. After adjusting for age, gender, race, ethnicity, diagnoses, and license status of the hospital, there were nine clusters (five with increased rates and four with lower rates). In addition to the three coastal clusters, a small cluster of lower rates was found in northeastern Los Angeles County.

Table 3 Clusters of higher or lower indigent psychiatric discharge rates
Full size table

In the adjusted analysis, significant clusters of increased rates of indigent discharges were found in Western Riverside County (26.5% of 26,866 psychiatric discharges were indigent), San Joaquin County (41.7% of 7,899 discharges), the Fresno/Tulare/King area (14% of 14,640 discharges), southeast Los Angeles County (30% of 1,519 discharges), and northern California (8.8% of 23,758 discharges). Of particular interest, note the relatively close spacing of increased and decreased clusters. In the Bay area, San Francisco’s low indigent cluster (0.01% of 40,448 discharges) is not far from San Joaquin’s high cluster. Orange County’s low indigent cluster (0.01% of 48,233 discharges) sits adjacent to Western Riverside County’s high indigent cluster. Likewise, the low indigent cluster of Ventura/Western Los Angeles County (0.01% of 214,125 discharges) is close to the high indigent cluster in Southeast Los Angeles.

As seen in Figure 1, which shows the clusters overlayed on a map with smoothed discharge rates, the rate of indigent psychiatric discharges was not uniform within each of the circular clusters. This underscores the importance of using more than one technique to examine spatial data. In general, there was concordance between statistical clusters and areas of higher or lower smoothed rates. The most obvious case of variation within a cluster was seen in cluster 8—northern California counties. Within that geographically large cluster which includes all or parts of 22 counties, but no major urban center, there were Zip Codes with both elevated and decreased rates. Cluster 6, in the agricultural basis of Fresno/Tulare/King Counties, whose major urban area is the city of Bakersfield, also included areas that were not uniformly high. There were two areas of California where Zip Codes having relatively high rates of indigent discharges were not selected to be in a cluster: the Lake Tahoe region of Northern California and central/eastern Riverside County in Southern California. The non-selection may have been due to the relatively low population density in those areas. Previous analysis by the authors looking at kidney transplants also found high smoothed rates around Lake Tahoe; but not a significant cluster.27

Figure 1
figure 1

Adjusted clusters of areas with higher or lower rates of indigent psychiatric discharges and smoothed discharge rates. Circles show SaTScan clusters within which ZIP Codes had risk ratios of indigent discharges significantly greater or lower than the State-wide average. The shading presents smoothed standardized mortality ratios for indigent discharges, with the SRR at each point taking into account the number of discharges within its surrounding area

Full size image

Figure 2 displays the indigent discharge clusters overlayed on the PUMA data. Within areas highlighted in black, more than 60% of the individuals estimated to have severe mental illness were in households living at less than 200% of the federal poverty level. Two obvious areas where these areas were not included in an indigent discharge cluster were the majority of the agricultural basin of Central California and the state’s desert southern tip of eastern Riverside and Imperial counties. These may be areas of decreased access to psychiatric hospitalization among those with severe mental illness living in poverty. The detailed maps of the Bay and Los Angeles areas clearly show that there are areas of high poverty among those with severe mental illness located within clusters of decreased indigent hospitalization rates. Given that each of the highlighted areas is a PUMA with at least 100,000 individuals (and an average of nearly 3,000 individuals with severe mental illness), this map suggests decreased psychiatric hospital access by those living in poverty within San Francisco, Los Angeles, and Orange counties. The Los Angeles Area map also shows several PUMAs with poverty rates greater than 60% which are outside of indigent hospitalization clusters: two in eastern Los Angeles County and one in San Bernardino County (directly north of the Riverside cluster). These are also possible areas with decreased access.

Figure 2
figure 2

Adjusted clusters of areas with higher or lower rates of indigent psychiatric discharges and estimates of poverty among those with severe mental illness. Shaded areas show within each PUMA the percent of estimated householders with severe mental illness living below 200% of the federal poverty level

Full size image

Discussion

The overall California rate of indigent psychiatric hospitalization (at least 4.7%) does appear to be high, as a study of psychiatric hospitalizations in Massachusetts found that 3.7% of hospitalizations in 2000 were for free care.17 In this study, robust GIS analytic techniques were applied to California statewide hospital discharge data to find regional clusters of increased (and decreased) rates of indigent psychiatric hospitalizations. As one uses spatial analysis techniques such as cluster detection and presentation methods such as smoothed maps, some areas can be clearly seen to have significantly greater or significantly lower rates of indigent acute psychiatric hospitalizations, although these two different techniques result in slightly different “pictures.”

The relatively small differences between the unadjusted and adjusted analyses suggest that patient-level characteristics do not account for much of the variation in geographic-based differences in the proportion of indigent psychiatric discharges. Of particular interest in the adjusted analysis was the finding of high clusters close to low clusters in the large urban centers of Los Angeles and San Francisco. Although this may be reflective of the “displacement model” referred to by Wolch and Philo,21 there is a possibility that such findings are in part due to the presence of many geographically small but densely populated Zip Codes or be reflective of referral patterns to state and county mental health hospitals. Regardless, the tabular data clearly show a temporal increase in both the number of psychiatric discharges and the proportion of indigent discharges. Further GIS efforts could examine spatial-temporal patterns, and this increase in indigent psychiatric hospitalizations provides impetus for additional data analysis and presentation.

Given the decentralized accountability of local mental health programs, there are policy and social justice imperatives to evaluate “area-specific inequalities in health need and service utilization,” especially as geographic variation in utilization raises questions about access and quality of care.14 Furthermore, simply identifying particular regions on which to focus efforts is a good example of psychiatric epidemiology contributing to more effective government policy.38 Indeed, as Fortney, et al.39 commented in their thorough spatial study including individual and community-level data, simply identifying areas of elevated hospitalization rates is of interest to payers, even without additional sophisticated analysis.

Presenting the spatial analysis of indigent psychiatric hospitalization data along with estimated need (in this case prevalence of poverty among those with severe mental illness) revealed that there were some areas where increased indigent psychiatric utilization coincided with estimated poverty among mental patients. However, there were also areas with elevated proportions of mental patients living in poverty, where there were not increased rates of indigent admissions. Others have noted the challenges of directly comparing utilization to population-based measures of mental health need. On the one hand, Kessler et al.40 have noted that it is difficult in population-based studies to accurately determine which percentage of individuals actually have a mental disorder. Even if the prevalence of mental illness is accurately measured, Mechanic has argued that prevalence is not a good measure of the need for mental health services, as there is not always a direct relationship between a mental diagnosis and disability which requires care.41 Within California, need for inpatient psychiatric care is clearly defined based on diagnosis, symptoms, and behaviors.9 Despite the conceptual and methodological challenges though, public mental health agencies must allocate limited resources, and needs assessment is one method that may contribute to better decision-making.

The results suggest that in areas such as West Los Angeles, San Francisco, and Orange County, psychiatric hospitals in general may be in a better financial position due to the relatively small number of indigent hospitalizations. On the other hand, there may be fewer financial resources in Riverside, San Joaquin, and Fresno counties, as hospitals and local mental health agencies appear to be bearing a higher rate of indigent psychiatric hospitalizations. It has been acknowledged within the California public system that county-level inequities in mental health funding have persisted since the 1960s.9 These maps may indicate one result of inequitable financing.

Although this GIS analysis is useful in producing needs-assessment maps that suggest areas with potential disparities, it cannot explain if those apparent disparities are due to inequitable financing, ineffective outpatient care, or a myriad of other factors. However, there are theoretical models one can use to examine potential disparities. One such approach is the Behavioral Model of Health Care Utilization,42 which is used to examine variation in utilization, considering personal characteristics and the context in which those services are provided. The Behavioral Model has been extended to more formally account for the effects of “community” on health care access and outcomes.43 Using this approach, researchers found differential effects of community measures in relation to low-income adults as they considered different types of access (potential and realized) and different types of patients (insured versus uninsured).44

A recent review of the literature45 concluded that there are three major groups of conceptual models for explaining how locations influence mental health status (ultimately, need for mental health services). These three major approaches are: structural characteristics models (sociodemographics), neighborhood disorder models (incivilities and feelings of safety), and environmental stress models (resources and stresses inherent in the built environment).45 Future research guided by a theoretical model could be used to study areas (identified by GIS techniques) in which there appears to be increased or decreased access to psychiatric hospital care by those who are uninsured. However, this would require sub-county data, such as general sociodemographics, details about Medicaid beneficiaries, public mental resources and expenditures, etc.

Limitations

On a theoretical level, ecological fallacy is a common limitation in spatial analyses. This is actually not a problem in the present study as person-level data for each discharge is available in both the numerator and denominator, allowing for direct control of individual differences in age, gender, race, and ethnicity. However, reductionism must be considered, as Zip Codes are not defined just by the sum of resident psychiatric discharges. The whole nature of an area is defined by many features, including residents, buildings, jobs, and social arrangements. This GIS analysis must be viewed as one step in the process of understanding and improving access issues.

On a technical level, standard limitations that have been identified with spatial analysis include the modifiable areal unit problem, i.e., one may have different results with different levels of geography and edge effects,39 particularly in calculating rates near the state boundary. As previously mentioned, there are problems with using Zip Codes.25 If one looks at multiple years worth of data, there may also be an issue with boundaries changing over time.46 Unfortunately, unless one has street-address-level data, there often is little other choice in analyzing hospitalization data.18

In addition to standard limitations inherent when analyzing administrative service data, there may be problems of population denominators, particularly among mobile sub-populations.14 Also, discharges among the severely mentally ill are not complete in this dataset due to the exclusion of state mental health hospitals. Given the lower likelihood of geocoding among indigents, discharge rates in some areas were no doubt shown as lower than actual practice. Another potential numerator confounder would result if some counties were more successful at getting low-income persons enrolled in Medi-Cal; in the past, some counties were more aggressive in obtaining state funding.9 Finally, although discharges are appropriate if one is concerned about costs, perhaps on a clinical level, a more useful denominator might be individuals, as roughly 40% of psychiatric inpatients are readmitted within a year of discharge.47 However, the use of multiple-robust GIS methods and a large sample size over several years may partially address these technical concerns.

Regarding estimated need, analysis of adults with severe mental illness in Virginia suggested a tendency to migrate away from more affluent urban and poor rural areas into less affluent urban areas, which is also where mental health services tend to be clustered.3 Furthermore, these need measures are based on household-level surveys, which excludes the homeless and incarcerated. Thus, these synthetic estimates based on static census data might over-represent need in some areas and under-represent need in others.

Regarding greater use of GIS analyses within state mental health agencies, often cited limitations of GIS are the costs associated with running and maintaining systems and the difficulty of obtaining current and complete data.25 Although this study was conducted in a university with existing geographic information systems capacity, utilization analysis was done using readily available public-use data. Many SMHAs no doubt do have sufficient information technology infrastructure to support GIS applications. Furthermore, as previously noted, GIS has been explicitly used in public mental health planning in New York.12 As a former analyst within a California county mental health department, the lead author has given formal GIS presentations48 and witnessed the benefits of limited application of spatial analysis within a public mental health agency.

Implications for Behavioral Health

The federal government has been encouraging the use of geographic information systems in the public health community, with the US Department of Health and Human Services recommending that all major local, state, and national health data systems that geocode data use GIS.25 Furthermore, many resources are freely available through the National Center for Health Statistics.49 To date, there has been limited application of GIS to mental health services within government agencies, with the exception of the Department of Veterans Affairs.50 However, GIS can be an important tool for combining disparate data sources to conduct needs assessments, including examination of potential small-area disparities of utilization relative to need. Many state mental health agencies already possess data which includes geographic identifiers. Such data could be transformed into maps that clearly present areas with increased or decreased utilization rates, and these utilization maps could be spatially linked with other data. Such maps could enhance internal decision-making and also be posted on public websites, being sensitive to legitimate privacy and confidentiality concerns. Such distribution among stakeholders would contribute to a shared understanding of the “meaning” of these maps. Appropriate spatial analysis techniques should be applied to obtain robust results which are not as dependent on the underlying geographic scale of the data. Existing theoretical models should be used in conjunction with GIS techniques to better understand reasons for inequitable use of services.