[Federal Register Volume 64, Number 121 (Thursday, June 24, 1999)]
[Notices]
[Pages 33890-33897]
From the Federal Register Online via the Government Publishing Office [www.gpo.gov]
[FR Doc No: 99-15377]
-----------------------------------------------------------------------
DEPARTMENT OF HEALTH AND HUMAN SERVICES
Substance Abuse and Mental Health Services Administration
Estimation Methodology for Adults With Serious Mental Illness
(SMI)
AGENCY: Center for Mental Health Services, Substance Abuse and Mental
Health Services Administration, HHS.
ACTION: Final notice.
-----------------------------------------------------------------------
SUMMARY: This notice establishes a final methodology for identifying
and estimating the number of adults with serious mental illness (SMI)
within each State. This notice is being served as part of the
requirement of Public Law 102-321, the ADAMHA Reorganization Act of
1992.
EFFECTIVE DATE: October 1, 1999.
FOR FURTHER INFORMATION CONTACT: Ronald W. Manderscheid, Ph.D., Chief,
Survey and Analysis Branch, Center for Mental Health Services, Parklawn
Building, Rm 15C-04, 5600 Fishers Lane, Rockville, MD 20857, (301) 443-
3343 (voice), (301) 443-7926 (fax), [email protected] (e-mail).
Scope of Application
All individuals whose services are funded through the Federal
Community Mental Health Services Block Grant must fall within the
definition announced on May 20, 1993, in the Federal Register, Volume
58, No. 96, p. 29422. Inclusion or exclusion from the estimates is not
intended to confer or deny eligibility for any other service or benefit
at the Federal, State, or local level. Additionally, the estimates are
not intended to restrict the flexibility or responsibility of State or
local governments to tailor publicly-funded systems to meet local needs
and priorities. Any ancillary use of these estimates for purposes other
than those
[[Page 33891]]
identified in the legislation is outside the purview and control of
CMHS.
Background
Pub. L. 102-321, the ADAMHA Reorganization Act of 1992, amended the
Public Health Service Act and created the Substance Abuse and Mental
Health Services Administration (SAMHSA). The Center for Mental Health
Services (CMHS) was established within SAMHSA to coordinate Federal
efforts in the prevention, treatment, and the promotion of mental
health. Title II of Pub. L. 102-321 establishes a Block Grant for
Community Mental Health Services administered by CMHS, which permits
the allocation of funds to States for the provision of community mental
health services to children with a serious emotional disturbance (SED)
and adults with a serious mental illness (SMI). Pub. L. 102-321
stipulates that States will estimate the incidence (number of new cases
in a year) and prevalence (total number of cases in a year) in their
applications for Block Grant funds. As part of the process of
implementing this new Block Grant, definitions of the terms ``children
with a serious emotional disturbance and ``adults with a serious mental
illness'' were announced on May 20, 1993, in the Federal Register,
Volume 58, No. 96, p. 29422. Subsequent to this notice, a group of
technical experts was convened by CMHS to develop an estimation
methodology to ``operationalize the key concepts'' in the definition of
adults with SMI. A similar group has prepared an estimation methodology
for children and adolescents with SED. The final SED estimation
methodology was published on July 17, 1998, in the Federal Register,
Volume 63, No. 137, p. 38661.
Summary of Comments
This final notice reflects a thorough review and analysis of
comments received in response to an earlier draft notice published in
the Federal Register, on March 28, 1997, Volume 62, No. 60, p. 14928.
CMHS received only nine comments expressing opinions about the
proposed methodology. Several questions were raised. These questions
are summarized in four broad areas: Operational definition of SMI,
complexity of the methodology, differences among States, and other
related comments.
Operational Definition of SMI
Some comments suggested that the SMI definition was too broad.
The final definition of SMI was published on May 20, 1993, in the
Federal Register, Volume 58, No 96, p. 29422. This definition cannot be
changed by the methodology outlined below.
SMI was defined as the conjunction of a DSM mental disorder and
serious role impairment. The Diagnostic Interview Schedule (DIS)
estimates were not enhanced. A respondent had to have a DIS/Composite
International Diagnostic Interview (CIDI) diagnosis and an impairment
to qualify for the operational definition of SMI. This means that the
estimated annual prevalence of SMI is always equal to or less than the
DIS/CIDI estimates of disorder prevalence. The charge to the technical
committee was to make what it considered to be the best decisions based
on available data about impairment to operationalize the definition of
SMI. The report of the committee describes in great detail how and why
the technical experts chose specific indicators.
It is important to note that Pub. L. 102-321 explicitly states that
SMI includes impairments in functioning. As a result, the technical
experts were required to include one component of the operational
definition that assesses functioning in social networks. Strict
criteria were used, such as reports of extreme deficits in social
functioning to qualify for this type of impairment. A respondent must
either have one of the following two profiles: (i) Complete social
isolation, defined as having absolutely no social contact of any type--
telephone, mail, or in-person--with any family member or friend and
having no one in his or her personal life with whom he/she has a
confiding personal relationship; or (ii) extreme dysfunction in
personal relationships, defined as high conflict and no positive
interactions and no possibility of intimacy or confiding with any
family member or friend. These persons comprise about 10% of those
classified as having SMI. The remaining 90% either have a severe
disorder like schizophrenia or bipolar disorder, or a disorder and work
impairment, or a disorder and report being suicidal.
The rationale for the 57% prevalence estimate of SMI among prison
inmates is well documented in the committee's report. A review of
epidemiological studies in inmate populations found that the average
estimated prevalence of any DIS disorder is 57%. The technical experts
concluded that all inmates with one of these disorders, by definition,
were functioning inadequately in social roles by virtue of the fact
that they were incarcerated.
This definition was adopted for very practical reasons. It is
important to remember that the inmate population represents less than
one percent of the adult population, and the prevalence estimate of 57%
is based on published work.
Some comments urged that the definition of SMI did not constitute
the service population for public mental health services.
This final notice includes a statement about the scope of
application of the estimates. That statement defines what is and is not
intended by the definition and the methodology.
Complexity of the Methodology
Some comments noted that the use of the Baltimore sample as a basis
for estimating national SMI rates among elderly persons may have
introduced errors into the estimates for persons 55 years and older.
The technical experts were mandated to arrive at the best estimate
based on currently available data. The Baltimore ECA data were the best
currently available for persons 55 years and older. Nationally
representative data would have been used if such existed. It will be
important in the future to improve the data available to produce
estimates for all age groups.
Some comments were made about distortions in State estimates and
lack of theory.
The technical experts used all available data on State-level
variables that could be obtained readily from the Federal government on
an annual basis and explored the effects of these variables in
predicting SMI. Such variables were deliberately selected to increase
the ease of application of the estimation methodology by the States in
the future. The experts believed and continue to believe that they
could do no less than exhaustively consider the full range of
potentially important predictors of SMI, irrespective of available
theory. The analytical iterations are explained in the committee's
report. These explanations provide all the detail a specialist in
applied statistics or demography would need to evaluate the procedures
adopted. These procedures are consistent with currently accepted
methods for making small area estimates. Government agencies currently
use similar methodologies to make estimates of other State-level social
policy variables.
Some comments suggested that confidence intervals were not provided
for State prevalence estimates.
Confidence intervals have been provided in this final notice, since
estimates are based upon samples rather than a complete enumeration.
[[Page 33892]]
Some comments suggested that the estimation methodology paper was
difficult to understand and that complex statistical procedures were
inadequately explained, with insufficient rationale.
In writing the paper, the authors were sensitive to the importance
of being clear about major decisions. The authors have had a great deal
of experience writing reports of empirical studies for critical
scientific and peer review. By the standards of this scientific review
process, the level of documentation presented in the estimation
methodology report is quite high.
Some comments indicated that no adjustment was made in the
methodology to address the phenomenon of different levels of reporting
of psychiatric symptoms by ethnic groups.
The technical experts included information to discriminate
nonhispanic whites from all other racial groups in the model. No fine-
grained distinctions were made about race/ethnicity because of the
small numbers of people in specific race/ethnicity subsamples in the
surveys that were analyzed. As part of the analysis, the technical
experts obtained all the information that was readily available from
the Census Bureau on Census Tract-level, County-level, and State-level
demographic variables. All these variables were included in efforts to
predict and estimate the prevalence of SMI.
Some comments suggested that the factor analysis was inadequate and
that important issues not described (e.g., the number of variables in
the analysis or how missing data were handled) could have affected the
results.
The factor analysis was carried out on a Census data file
containing County-level data from the 1990 Census. The sample size was
the number of Counties in the U.S., while the number of variables was
over 100 Census characteristics. Some of the characteristics were quite
highly correlated across Counties, like median household income and
mean household income, or the number of men in a County and the number
of women in a County. Factor analysis was used as a way of reducing
redundancy prior to performing further analyses. The factor analytic
procedures employed represent the state-of-the-art for similar data
reduction procedures.
Some comments were made about the use of varimax rather than
oblique rotation, the decision to examine only the first ten factors in
the solution, and the use of factor-weighted scores.
The group of technical experts explored both oblique and rigid
rotations and also looked at the unique factors after the first ten.
``Unique factors'' refer to factors in which there is only a single
variable with a high loading. Variance was noted to be trivial after
the first ten factors. No factors after the first ten had more than one
variable with high loading. Factor-weighted and factor-based scales are
very highly correlated, therefore the choice of one over the other did
not affect the results of the analyses.
Some comments noted that Census data are stronly influenced by
population size and suggested that this effect could be removed to find
a more meaningful structure.
A similar procedure was actually used. All count variables were
transformed (e.g., number of vacant houses, number of people on
welfare) into population proportions. This procedure removes the
effects of population size.
Some comments suggested that users of the public mental health
system have low levels of income. However, the key significant income
predictor was an interaction term for high income and urbanicity
associated with reduced prevalence of SMI.
The technical experts were surprised to find the absence of high
income people was a stronger predictor of SMI than the presence of low
income people. This was investigated in considerable detail, trying a
number of different specifications in search of a low income effect.
These included a specification involving the assessment of
neighborhoods with a bimodal distribution of high income and low income
people, as well as a specification that examined the effect of degree
of variation in income in the community (e.g., differentiation between
a community with an average income of $30,000 due to all families
having this income versus another with an average of $30,000 due to 10%
of families making $210,000 and another 90% making $10,000. After a
careful review, the technical experts concluded that the data did not
support a low income effect or any effect of income variance for SMI.
It is important to note that there is a strong low income effect for
estimates of persons with severe and persistent mental illness (SPMI),
even though such an effect could not be found for SMI.
It is noteworthy that the analysis of income effects was confined
to neighborhoods (Census Tracts) due to the fact that the Census Bureau
would not release individual-level family income data cross-classified
by other Census variables at either the Tract, County, or State levels.
The Census Bureau decision was based on the concern to maintain
confidentiality of Census records.
Some comments requested future consideration of SMI incidence.
Currently, no nationally representative data are available on
incidence of SMI. The group of technical experts has made
recommendations to CMHS regarding the need for future data collection
to obtain incidence data.
State Differences
Some comments suggested that SMI prevalence was higher in the West
and the Southwest, compared with other regions of the US.
The magnitude of the SMI estimates, averaging approximately 5-6% of
the adult population in a year, is very plausible. It is generally
agreed that 2-3% of the adult population suffer from severe and
persistent disorders such as schizophrenia, other nonaffective
psychoses, and bipolar disorder. Based upon the estimation methodology,
an additional 2-3% of the adult population suffer from serious anxiety,
nonbipolar mood disorders, and other disorders, for a total of 5-6%. It
would be highly suspicious if the estimates were any less.
In the draft notice of the estimation methodology, point estimates
were provided for State SMI prevalence figures. In this final notice, a
95% confidence interval is used to calculate the SMI prevalence rate as
a range. State prevalence of SMI is estimated to be between the lower
and upper percent limits for each State. Based on these analysis, one
cannot conclude that rates differ among States. Hence, the same
prevalence rate and percentage standard error are applied to all States
to produce the numerical estimates provided in table 1. See the
footnote to table 1 for further information on this estimation
procedure.
Some comments noted that the inclusion of Alzheimer's disease
contributes appreciably to the counts and that, since the definition
cannot be changed at this point, the report should clearly note that
this is the case.
This is a good suggestion.
Some comments suggested that only 10 States are at or below the
national average, and that the majority of these States are quite
small, therefore a mathematical explanation of this phenomenon would be
appropriate.
This comment does not reflect the nature of the estimation
methodology. As stated in the draft Federal Register notice of March
23, 1997, Volume 62, No 60, page 14931, the national total estimated
number of persons with SMI is derived from direct, weighted counts
[[Page 33893]]
from the surveys used. However, the State totals were computed from
synthetic modeling at the County level, and county estimates were
summed to arrive at State totals. These two approaches are not the
same. Therefore, they are subject to different types of sampling and
non-sampling errors. As a result, the sum of State totals will not
necessarily equal the U.S. total, and State estimates cannot be
compared directly with the national average.
Some comments suggested that use of national probability estimates
did not permit consideration of regional and state differences, which
could affect the relationship between key analytical variables.
Because of the difficulty of obtaining data, the technical experts
made the assumption that the effects of all the predictor variables
were the same across all States. More precise estimates could have been
made if representative samples from each State were available.
Other Related Comments
Some comments noted that the exclusion of homeless and
institutionalized persons, those living in group quarters, and those
without telephones excludes the segments of the population with the
highest risk of SMI.
The Epidemiologic Cachement Area (ECA) and the National Commobidity
Survey (NCS) studies were both household surveys, so there is no
exclusion of non-telephone households. Although national data were used
to estimate the overall U.S. prevalence of the omitted population
groups, due to lack of data, no attempt was made to estimate how many
homeless people or persons in the other excluded segments reside in
each State.
Some comments suggested the need to have prevalence estimates for
Puerto Rico.
The prevalence estimates for Puerto Rico are included in this
notice.
Some comments suggested validity studies that could form the basis
for modifications and refinements to the estimation methodology.
Validation studies could help refine the estimation methodology.
However, the mandate to the technical experts was to develop the best
estimates with currently available data rather than only propose new
data collections. As noted earlier, the technical experts have
recommended that CMHS carry out a nationally representative survey once
each decade in the Census year explicitly designed to assess the
prevalence of SMI and SPMI, with oversampling to allow estimation by
State. Execution of validation studies as part of this survey would
permit the evaluation of and increased precision in State-level
estimates.
Some comments urged SAMHSA to increase Block Grant Funds for States
to offer services to the number of persons who have SMI.
The first step in such a process is the one currently being
undertaken, i.e., using the estimation methodology to produce estimates
showing that the number of adults with SMI exceeds the number who can
be served with currently available funds.
SMI Estimation Methodology
Data Sources
Data from two major national studies, the NCS and the ECA, were
used to estimate the prevalence of adults with SMI. The NCS, a
nationally representative sample household survey conducted in 1990-91
assessed the prevalence of DSM-III-R disorders in persons aged 15-54
years old. This sample included over 1,000 census tracts in 174
counties in 34 States. The ECA, a general population survey of five
local areas in the U.S., was conducted in 1980-85 to determine the
prevalence of DSM III disorders in persons age 18 and older. The ECA
data utilized for the present analysis were limited to the Baltimore
site because that was the only site that had disability data needed to
operationalize the criteria for SMI. Although the Baltimore sample is
not nationally representative, it is used in this analysis because the
ECA provides a rough replication and check on the NCS data. Also, the
NCS does not have data on persons age 55 and older, so the ECA data are
used to estimate the prevalence of serious mental illness among persons
55 years and older.
The group of technical experts determined that it is not possible
to develop estimates of incidence using currently available data.
However, it is important to note that incidence is always a subset of
prevalence. In the future, information on both incidence and prevalence
data will need to be collected.
Serious Mental Illness (SMI)
As previously defined by CMHS, adults with a serious mental illness
are persons 18 years and older who, at any time during a given year,
had a diagnosable mental, behavioral, or emotional disorder that met
the criteria of DSM-III-R and ``* * * that has resulted in functional
impairment which substantially interferes with or limits one or more
major life activities.* * *.'' The definition states that ``* * *
adults who would have met functional impairment criteria during the
referenced year without the benefit of treatment or other support
services are considered to have serious mental illnesses. * * *'' DSM-
III-R ``V'' codes, substance use disorders, and developmental disorders
are excluded from this definition.
The following criteria were used to operationalize the definition
of serious mental illness in the NCS and ECA data:
(1) Persons who met criteria for disorders defined as severe and
persistent mental illnesses (SPMI) by the National Institute of Mental
Health (NIMH) National Advisory Mental Health Council (National
Advisory Mental Health Council, 1993).
To this group were added:
(2) Persons who had another 12-month DSM-III-R mental disorder
(with the exclusions noted above), and
--Either planned or attempted suicide at some time during the past 12
months, or
--Lacked any legitimate productive role, or
--Had a serious role impairment in their main productive roles, for
example, consistently missing at least one full day of work per month
as a direct result of their mental health, or
-Had serious interpersonal impairment as a result of being totally
socially isolated, lacking intimacy in social relationships, showing
inability to confide in others, and lacking social support.
Estimation Procedures
Two logistic regression models were developed to calculate
prevalence estimates for adults with SMI.
(a) A Census Tract Model for years in which the decennial U.S.
census is conducted.
(b) A County-Level Model to be used in intercensal years.
In non-censal years, the county-level model will be used to
estimate SMI prevalence, after adjusting for its known relationship
with the census tract model.
Formula
Census-Tract Model
Using 1990 census data, a logistic regression model was developed
to calculate predicted rates of SMI for each cell of an age by sex by
race table for each of the 61,253 Census Tracts in the country. Next,
the rates were multiplied by cell frequencies and subtotaled to derive
tract-level estimates. Finally, the tract-level estimates were
aggregated to arrive at county-level and state-level prevalence
estimates of adults with SMI. This regression methodology is often used
in small area estimation (Ericksen,
[[Page 33894]]
1974; Purcell & Kish, 1979). The actual Census Tract Model equation is
specified immediately below:
Parameter Estimates for Census Tract Model
------------------------------------------------------------------------
95% Confidence
Predictor Odds ratio interval
------------------------------------------------------------------------
Intercept..................... *0.02 (0.01-0.04)
------------------------------------------------------------------------
Individual-Level Variables
------------------------------------------------------------------------
Age:
18-24......................... *1.94 (1.18-3.17)
25-34......................... 1.32 (0.86-2.03)
35-44......................... 1.46 (0.96-2.21)
45-54......................... 1.00
Sex:
Female........................ *2.23 (1.57-3.19)
Male.......................... 1.00
Race:
Nonhispanic white............. 1.00
Black/Hispanic/other.......... *0.49 (0.28-0.87)
Marital Status:
Married/Cohabiting............ 1.00
Never Married................. *3.90 (1.15-3.08)
Separated/Divorced/Widowed.... *1.88 (2.41-6.31)
------------------------------------------------------------------------
Census Tract Level Variables
------------------------------------------------------------------------
F2 (High socio-economic 1.16 (0.90-1.49)
status)......................
F4 (Immigrants)............... 0.99 (0.85-1.14)
------------------------------------------------------------------------
County-Level Variables
------------------------------------------------------------------------
County Urbanicity:
Metropolitan.................. 1.12 (0.85-1.49)
Other......................... 1.00
------------------------------------------------------------------------
Interactions Among Variables
------------------------------------------------------------------------
FemaleXSeparated/Divorced/Widowed. *0.47 (0.24-0.91)
FemaleXNever Married.............. *0.47 (0.28-0.78)
Non WhiteXSeparated/Divorced/ *2.62 (1.29-5.33)
Widowed..........................
Non WhiteXNever Married........... 1.81 (0.95-3.44)
FemaleXF2......................... *0.70 (0.51-0.96)
UrbanicityXF2..................... *0.75 (0.52-0.95)
F2XF4............................. *0.78 (0.64-0.94)
------------------------------------------------------------------------
*Significant at the .05 level, two tailed test; F2=Census Tract factor
score for high socioeconomic status (SES); F4=Census Tract factor
score for immigrants.
The estimate for persons 55 years and older is derived from
analysis of ECA data in conjunction with NCS data. The prevalence
ratios among ECA respondents ages 55-64 and 65 years and above, were
found to be 84 and 31 percent as large, respectively, as the prevalence
estimate for NCS respondents 18-54 years old, after controlling for
differences in gender and race. NCS State-level estimates were
extrapolated using these ratios. These ratios did not differ
significantly by sex or race. A factor of .81 was applied to State-
level SMI estimates for the age range 18-54 to derive the rate for the
age range 55-64, and .31 was used to arrive at the estimate for person
65 and older. A weighted sum (by age distribution of each State) was
calculated to determine the final State-level SMI prevalence estimate.
County Model
U.S. Census Bureau tract-level data are available only for years in
which the decennial U.S. Census is conducted. To obtain prevalence
estimates for adults with SMI during intercensal years, the group of
technical experts used biennial individual- and county-level data from
the Census Bureau's small area estimation program. Predicted values
from the logistic regression equation were used to calculate county-
level estimates. In contrast to the Census Tract Model, the initial
estimates using this approach were generated at the county level. These
county-level estimates are then summed to provide State-level
prevalence estimates. The actual county-level model equation is
specified immediately below:
Parameter Estimates for County-Level Model
------------------------------------------------------------------------
95% Confidence
Predictor Odds ratio interval
------------------------------------------------------------------------
Intercept..................... * 0.04 (0.02-0.07)
[[Page 33895]]
Individual-Level Variables
------------------------------------------------------------------------
Age:
18-24......................... 1.69 (1.00-2.85)
25-34......................... 1.10 (0.65-1.88)
35-44......................... 1.24 (0.71-2.15)
45-54......................... 1.00 .................
Sex:
Female........................ 1.58 (1.17-2.13)
Male.......................... 1.00 .................
------------------------------------------------------------------------
County-Level Variables
------------------------------------------------------------------------
Urbanicity:
Metropolitan.................. 1.35 (0.99-1.85)
Other......................... 1.00 .................
------------------------------------------------------------------------
*Significant at the 0.05 level, two-tailed test.
Adjustment for persons age 55 years and older is carried out as in
the Census Tract Model. An adjustment factor (Census Bureau, Fay, 1987;
Fay & Herriot, 1979) based on the ratio of County-Level Model estimates
for 1990 and Census Tract Model estimates for 1990 can be used to
adjust estimates for subsequent years from the County-Level Model. This
procedure assumes that the Census Tract Model is more accurate than the
County-Level Model.
County and State Estimates
As stated earlier, Census Tract Model prevalence estimates were
summed to derive county estimates, and county estimates were summed to
arrive at State estimates. The 12-month prevalence of SMI is estimated
nationally to be 5.4 percent (with a standard error of 0.9 percent) or
10.2 million people in the adult household population (95 percent
confidence interval ranging from 7.0 million to 13.4 million), of which
2.6 percent or 4.8 million adults have SPMI (figure 1). When the
standard error is considered, State estimates do not vary. Hence, State
estimates are defined as 5.4 percent of the adult population, with a 95
percent confidence interval of plus or minus 1.96 times 0.9 percent.
The above estimates are based on noninstitutionalized persons
residing in the community. Limited information currently exists on SMI
estimates for persons institutionalized (i.e., persons in correctional
institutions, nursing homes, the homeless, persons in military
barracks, hospitals/schools/homes for persons who are mentally ill or
mentally retarded). Fischer and Breakey (1991) indicate that, on
average, the SMI prevalence rate for these groups (including about 5
million people or 2.7 percent of the U.S. adult population) is about 50
percent. The following assumptions were made in deriving rough
estimates of SMI prevalence for persons who are institutionalized: (a)
For 1.1 million residents of correctional institutions, 100 percent of
whom are adults, prevalence of SMI is estimated to be 57 percent; (b)
For 1.8 million residents of nursing homes, 100 percent of whom are
adults, prevalence of SMI is estimated to be 46 percent; (c) For 0.5
million persons who are homeless, 80 percent of whom are adults,
prevalence of SMI is estimated to be 50 percent; (d) For 0.6 million
persons in military barracks, all of whom are adults, the SMI
prevalence rate is equivalent to that of the adult household
population; (e) For 0.4 million persons in hospitals, homes, and
schools for persons who are mentally ill, 80 percent of whom are
adults, prevalence of SMI is estimated to be 100 percent. (f) For 0.6
million persons in other institutional settings such as chronic disease
hospitals, homes and schools for persons with physical disability, and
rooming houses, 50 percent of whom are adults, prevalence of SMI is
estimated to be 50 percent.
State estimates of each of these populations can be added to the
State SMI populations identified below.
Only a portion of adults with SMI seek treatment in any given year.
Due to the episodic nature of SMI, some persons may not require mental
health service at any particular time.
Provision of Estimates to States
CMHS will provide each State mental health agency with estimates in
order to initiate the first cycle of use. Subsequently, CMHS will
provide technical assistance to States to implement the methodology
using State demographic information.
The intial set of State estimates is provided in table 1 below.
Further background information on these estimates can be found in
Kessler, et al. (1998).
Table 1.--Estimated 12-Month Number of Persons With Serious Mental Illness, Age 18 and Older
[By State, 1990 *]
----------------------------------------------------------------------------------------------------------------
95% confidence interval
State Point estimate -------------------------------
Lower limit Upper limit
----------------------------------------------------------------------------------------------------------------
Alabama......................................................... 161,017 110,327 211,708
Alaska.......................................................... 20,396 14,730 26,817
Arizona......................................................... 144,942 104,680 190,572
Arkansas........................................................ 93,398 63,995 122,801
California...................................................... 1,188,502 814,344 1,562,660
Colorado........................................................ 131,389 90,026 172,752
[[Page 33896]]
Connecticut..................................................... 137,027 93,889 180,165
Delaware........................................................ 27,153 18,605 35,701
District Columbia............................................... 26,450 18,123 34,776
Florida......................................................... 543,871 372,652 715,090
Georgia......................................................... 256,549 175,784 337,315
Hawaii.......................................................... 44,718 30,640 58,795
Idaho........................................................... 37,711 27,235 49,582
Illinois........................................................ 458,149 313,917 602,381
Indiana......................................................... 220,763 151,263 290,262
Iowa............................................................ 111,125 76,141 146,109
Kansas.......................................................... 98,062 67,190 128,933
Kentucky........................................................ 147,485 101,054 193,915
Louisiana....................................................... 161,606 110,730 212,482
Maine........................................................... 49,622 34,000 65,244
Maryland........................................................ 195,438 133,911 256,965
Massachusetts................................................... 251,821 172,544 331,098
Michigan........................................................ 369,173 252,952 485,394
Minnesota....................................................... 173,249 118,708 227,790
Mississippi..................................................... 98,629 67,579 129,678
Missouri........................................................ 205,321 140,683 269,959
Montana......................................................... 31,156 21,348 40,964
Nebraska........................................................ 62,066 42,527 81,605
Nevada.......................................................... 48,864 33,481 64,247
New Hampshire................................................... 44,847 30,728 58,965
New Jersey...................................................... 320,259 219,437 421,082
New Mexico...................................................... 57,690 39,528 75,851
New York........................................................ 741,469 535,505 974,894
North Carolina.................................................. 271,214 185,832 356,597
North Dakota.................................................... 25,024 17,146 32,902
Ohio............................................................ 434,558 297,753 571,363
Oklahoma........................................................ 124,663 85,417 163,909
Oregon.......................................................... 114,382 78,373 150,392
Pennsylvania.................................................... 490,689 336,213 645,165
Puerto Rico..................................................... 195,719 159,550 231,817
Rhode Island.................................................... 42,000 28,778 55,222
South Carolina.................................................. 138,591 94,960 182,221
South Dakota.................................................... 26,867 18,409 35,325
Texas........................................................... 656,136 449,575 862,698
Tennessee....................................................... 197,671 135,441 259,901
Utah............................................................ 59,152 40,530 77,774
Vermont......................................................... 22,662 15,528 29,797
Virginia........................................................ 252,861 173,257 332,466
Washington...................................................... 194,686 133,396 255,977
West Virginia................................................... 72,895 49,946 95,843
Wisconsin....................................................... 194,550 133,303 255,798
Wyoming......................................................... 17,175 11,768 22,582
-----------------------------------------------
Total....................................................... 10,191,412 7,043,431 13,374,301
----------------------------------------------------------------------------------------------------------------
Does not include persons who are homeless or are institutionalized.
* Because there are no differences among States, the estimate for each State is calculated as 5.4 percent of the
total State adult population. The size of the 95 percent confidence interval for each State is equal to the
percentage estimate plus or minus 1.96x0.9 percent. The percentage estimate and the percentage standard error
are identical across States. However, the numeric estimate and numeric standard error vary depending on the
State adult population. The percentage standard error (0.9 percent) used to compute the upper and lower 95-
percent confidence limits is estimated using jackknife repeated replication (JRR) variance analysis (Kish and
Frankel 1974). The JRR calculations assume that the imputation ratios and the population proportions in the
different age groups based on the census data are correct. The confidence limits simulate the error introduced
into the estimates by imprecision in the prevalence estimates for NCS respondents in the age range 18-54.
Limitations
The ECA and NCS were designed to study lifetime prevalence of
mental disorders rather than 12-month prevalence. As a result, the
emphasis in diagnostic assessment was on lifetime disorders. In
addition, functional impairment was not a primary focus in either the
ECA or the NCS.
Current data cannot provide estimates of incidence. Additional
information needs to be collected in the future.
It is anticipated that additional work will be done in future years
to refine and update the estimation methodology. CMHS will apprise
States as this work develops.
References
Ericksen, E.P. (1974). A regression method for estimating population
changes of local areas. Journal of American Statistical Association,
69, 867-875.
[[Page 33897]]
Fay, R.E. (1987). Application of multivariate regression to small
domain estimation. In R. Platek, J.N.K. Rao, C.E. Sarndal & M.P.
Singh (Eds.), Small Area Statistics: An International Symposium, pp.
91-102. New York: John Wiley and Sons.
Fay, R.E., & Herriot, R. A. (1979). Estimates of income for small
places: An application of James-Stein procedures to Census data.
Journal of the American Statistical Association, 74, 269-277.
Fischer, P.J., Breakey, W.R. (1991). The Epidemiology of alcohol,
drug, and mental disorders among homeless persons. American
Psychologist. 46, 1115-1125.
Kessler, R.C., et al. Estimation of the 12-month Prevalence of
Serious Mental Illness (SMI). (1996). Unpublished reports to CMHS.
Kessler, R.C., et al. Population-Based Analyses: A Methodology for
Estimating the 12-Month Prevalence of Serious Mental Illness. R.W.
Manderscheid and M.J. Henderson, (eds.) Mental Health, United
States, 1998. DHHS Pub. Washington, D.C.: Supt. of Docs. U.S. Govt.
Print Off., pp. 99-112, 1998.
Kish, L., and Frankel, M.R. Inferences from complex samples, Journal
of the Royal Statistical Society, Series B 36:1-37, 1974.
National Advisory Mental Health Council. (1993). Health care reform
for Americans with severe mental illness. American Journal of
Psychiatry, 150, 1447-1465.
Purcell, N.J., & Kish, L. (1979). Estimation for small domains.
Biometrics, 35, 365-384.
Regier, D.A., Narrow, W.E., Rae, D.S., Manderscheid, R.W., Locke,
D.Z., Goodwin, F.K. (1993). The de Facto US Mental and Addictive
Disorders Service System. Archives of General Psychiatry, 50, 85-94.
Dated: June 7, 1999.
Richard Kopanda,
Executive Officer, Substance Abuse and Mental Health Services
Administration.
BILLING CODE 4162-20-P
[GRAPHIC] [TIFF OMITTED] TN24JN99.048
[FR Doc. 99-15377 Filed 6-23-99; 8:45 am]
BILLING CODE 4162-20-C