Validation of a Model for Estimating State and Local Prevalence of Serious Mental Illness

            Background / Purpose: Considerable progress has been made in recent years in the development of national estimates of mental illness rates through large-scale epidemiological surveys, notably the 2001-2002 replication of the National Comorbidity Study (NCS-R) (Kessler, et al., 2005). Unfortunately such sophisticated methods are prohibitively expensive for local communities, or even many states, which are faced with problems in equitably allocating scarce resources to local services. Yet, with the continuing development of a variety of regression-based synthetic estimation methodologies (Bajekal, et al., 2004), it has become feasible to use a mixed methods design for applying results of national studies, when in conjunction with other sources of data, to produce synthetic estimates of the prevalence of serious mental illness for local areas.             Methods: This study produces state and local-level estimates of serious mental illness in the U.S. through the application of small area estimation methodologies using data collected as part of a replication of an analysis of the NCS-R (Kessler & Merikanigas, 2004). Its does this through an adaptation of a recently developed methodology known as “regression synthetic estimation fitted using area-level covariates” (Heady et al., 2003). This involves estimation of a predictive model of variations in SMI rates on the individual level. This first phase of the project, which uses logistic regression with the NCS-R data, is in some respects a replication of work down by Kessler (2005). In the second phase, the coefficients derived from the foregoing model are used with parallel predictors on the area level, using data obtained through the 2000 census, and coded using the same categories used to estimate the individual-level model, to compute area-level estimates for 48 U.S. states and sample localities. Estimates were then regressed on independent indicators of serious mental illness to assess their validity.              Results: Estimates generated in this study show not only face validity and internal consistency, but also predictive validity. The logistic model has an overall predictive accuracy of 91.1%, based on predictors involving socioeconomic disparities. Pearson r validity coefficients for the area estimates range from 0.43 to 0.75. The model generates a national estimate of SMI adults of 5.5%; for the 48 states, rates ranging from 4.7% to 7.0%; and for a Northeastern state, local rates from 1.1% to 7.5%.             Conclusions and Implications: The results of this project have applications in several areas. Rates estimated on the basis of studies such as the NCS-R, with its demonstrated reliability and generalizability, provides an inexpensive and credible data source for the assessment of many types of need. Policy researchers and program evaluators often struggle to make meaningful comparisons between systems and agencies as to effectiveness and impact, but are hamstrung by problems in making adjustments for differential case and problem mix in targeted communities, and for this purpose, prevalence rates provide an invaluable basis for such adjustments. SMI rates also have implications for the deployment of specific service modalities, such as assertive case management, clubhouses, and psychiatric rehabilitation, as well as the development of service access strategies.