 |
This Web site is a component of the SAMHSA Health Information Network. |
|
Section III: Population Dynamics
PDF
version
You will need Adobe
Acrobat Reader to view this file.
Chapter 12. Estimating the Prevalence and Correlates of Serious Mental Illness in Community Epidemiological Surveys
Ronald C. Kessler, Ph.D.
Department of Health Care Policy, Harvard Medical School
Patricia A. Berglund, M.B.A.
Institute for Social Research, University of Michigan
Meyer D. Glantz, Ph.D.
Division of Epidemiology, Services, and Prevention Research, National Institute
of Drug Abuse
Doreen S. Koretz, Ph.D.
Division of Mental Disorders, Behavioral Research and AIDS, National Institute
of Mental Health
Kathleen R. Merikangas, Ph.D.
Mood and Anxiety Disorders Program, National Institute of Mental Health
Ellen E. Walters, M.S.; Alan M. Zaslavsky, Ph.D.
Department of Health Care Policy, Harvard Medical School
Introduction
Public Law (PL) 102 321, the Alcohol, Drug Abuse, and Mental Health Administration
(ADAMHA) Reorganization Act, established a block grant for States to fund community
mental health services for adults with serious mental illness (SMI) and children
with severe emotional disturbances (SED). The law required States to include
SMI incidence and prevalence estimates in their annual applications for block
grant funds. The law also required the Substance Abuse and Mental Health Services
Administration (SAMHSA) to develop an operational definition of SMI and to create
an estimation methodology based on this definition for use by States. The definition
of SMI in PL 102 321 requires an individual to have at least one 12-month Diagnostic
and Statistical Manual (DSM) disorder other than a substance use disorder
(SUD), and to have "serious impairment." SAMHSA subsequently decided that "serious
impairment" should be defined as a Global Assessment of Functioning (GAF) score
below 60 (Endicott, et al., 1976; SAMHSA, 1993).
SAMHSA established a task force to develop a methodology to estimate the incidence
and prevalence of SMI. The first step was to reanalyze data from the two recent
major psychiatric epidemiological surveys of DSM disorders in the United States,
the Epidemiological Catchment Area (ECA) study (Robins and Regier, 1991), and
the National Comorbidity Survey (NCS) (Kessler et al., 1994). The goals of the
task force were to estimate the prevalence of SMI in the country as a whole
and to examine sociodemographic correlates of SMI. This work was carried out
with the recognition that neither the ECA nor the NCS was designed with the
goal of estimating SMI. Imprecise post hoc indicators of impairment consequently
had to be used to approximate the SAMSHA definition. On the basis of this work,
the task force estimated that 5.4 percent of the U.S. noninstitutionalized civilian
population meet criteria for SMI at some time each year (Kessler et al., 1996).
The task force found that SMI is significantly more common among women than
men, more common among previously married and never married people than among
married people, and inversely related to both income and education (Kessler
et al., 1999). SMI was found to be insignificantly related to race/ethnicity,
living in an urban setting, and region of the country.
The second step for the task force was to explore whether precise estimates
of the prevalence of SMI could be generated for counties and States by applying
standard, small-area estimation methods to the ECA and NCS data (Schaible, 1996).
This exploration showed clearly that precise estimates of this sort cannot be
made currently because of the weak associations between available small-area
demographic variables and SMI (Kessler et al., 1999). The task force began its
exploration by developing prediction equations for SMI in the NCS data that
exclusively used predictors available for each county and State through the
Area Resources File (ARF). The ARF is a compendium of data on a wide range of
topics (e.g., demography, topography, weather conditions, road and traffic conditions,
health care resources, criminal justice) that is assembled from a variety of
government sources and continually updated. ARF information was applied to these
equations to generate a predicted prevalence of SMI for each county and State
in the country. Appropriate standard errors of these estimates were then generated
to adjust for the imprecision of the prediction equations. The resulting estimates
were shown not to differ significantly across counties and States. Although
this result could have occurred because the true prevalence of SMI is the same
in all counties and States in the United States, a more plausible interpretation,
and the one adopted by the SMI task force, is that the prediction equations
were too weak and the geographic variation in these predictors was too small
to detect the true differences in the prevalence of SMI across counties and
States.
A Screening Scale for SMI
On the basis of these results, the task force recommended that an SMI screening
scale be developed for use in ongoing government surveys, such as the National
Health Interview Survey (NHIS) and the SAMHSA National Household Survey on Drug
Abuse (NHSDA), and that analysis of these results be used to generate small-area
estimates of SMI for counties and States. SAMHSA funded a methodological study
to implement this recommendation. This study considered three measures as possible
screens for SMI. The first was a truncated version of the World Health Organization
(WHO) Composite International Diagnostic Interview Short-Form (CIDI-SF) scales
(Kessler, et al., 1998). These disorder-specific scales are designed to assign
predicted probabilities of meeting 12-month criteria for a number of DSM-IV
anxiety and mood disorders on the basis of a short series of questions for each
disorder.
The second measure was a modified version of the K10/K6 scales of nonspecific
psychological distress (Kessler et al., in press). These short (6 and 10 questions)
scales were developed for use in the core of the redesigned U.S. NHIS to measure
the frequency of commonly occurring symptoms of psychological distress (e.g.,
worry, restlessness, sadness) over a 30-day recall period. The K10/K6 scales
were modified for use in the SAMHSA methodological study to ask about symptoms
in the month in the past year when the respondent's emotional problems were
worst.
The third measure was a truncated version of the WHO Disability Assessment
Schedule (WHO DAS) (Rehm et al., 1999). The WHO-DAS was developed to operationalize
the core dimensions in the WHO International Classification of Functioning,
Disability, and Health (World Health Organization, 2001) by asking about the
severity of illness-related impairments in a variety of role domains over a
30-day recall period. The WHO-DAS was modified for use in the SAMHSA methodological
study to ask about impairments caused by emotional problems during the month
in the past year when the respondent's emotional problems were worst.
All three scales were administered to the second-stage sample of a two-stage
general population convenience sample. The first stage of this sample consisted
of 1,000 people who were screened by telephone for serious mental health problems.
The second stage consisted of 155 respondents selected from the first-stage
sample to oversample people with suspected serious mental health problems. Second-stage
respondents were interviewed face to face in their homes by trained clinical
interviewers. The interviews began with respondents self-administering the CIDI-SF,
K10/K6, and WHO-DAS scales in a way that kept responses hidden from the interviewer.
The interviewer then administered the Structured Clinical Interview for DSM-IV
(SCID) (First, et al., 1997) and rated respondents on the GAF while remaining
blind to the responses on the screening scales. Respondents were classified
as having SMI or not on the basis of their SCID and GAF ratings. Logistic regression
analysis was then used to estimate associations between the screening scales
and SMI. As reported previously (Kessler et al., 2003), the K6 was found to
be the most powerful predictor of SMI, with an area under the Receiver Operator
Characteristic Curve of .86. None of the other screening scales significantly
predicted SMI after scores on the K6 were controlled. The optimal cut-point
on the K6 in terms of equalizing false positives and false negatives in the
weighted second- stage sample was 0 to 12 versus 13+, coding item responses
0 to 4 and summing items to yield a scale with a 0 to 24 range. Sensitivity
(standard error in parentheses) was .36 (.08), specificity was .96 (.02), and
total classification accuracy was .92 (.02) at this cut-point.
Scoring the Screening Scale
On the basis of these results, the K6 was added to the NHSDA in 2001. Results
are not yet available. However, as noted, the K6 was originally developed for
use in the NHIS, and it has been included in the NHIS since 1997. With a random
adult sample of more than 40,000 respondents each year, K6 reports have now
been obtained on approximately a quarter of a million people in the NHIS. Given
the strength of the association between the K6 and SMI in the SAMHSA methodological
study, this large number of nationally representative cases should be enough
to generate fairly precise estimates of the prevalence of SMI in most States.
However, doing so requires calibration rules to be available. The SAMHSA methodological
study was too small and too unsystematic to generate such rules. This statement
might be confusing to readers in light of the fact, noted above, that the optimal
cut- point in terms of balancing false positives with false negatives in the
methodological study was between the range 0 to 12 (predicted noncases) and
the range 13 to 24 (predicted cases) on the 0 to 24 range of the K6. A superficial
way to code the K6 in the NHIS and NHSDA surveys would be to use this rule to
define likely cases. However, it is important to recognize that this cut-point
is only an approximation and is only "optimal," even in the narrowly defined
sense proposed here (i.e., balancing false positives and false negatives), when
the probability of SMI at a given score on the K6 (positive predictive value)
is constant across samples. Positive predictive value will not be constant,
though, even if the separate conditional distributions of K6 scores among people
with SMI (sensitivity) and without SMI (specificity) are constant across samples,
unless the true prevalence of SMI remains constant. The reason is that any deviation
in the proportion of people with actual SMI in the new samples will lead to
changes in positive predictive value at a given level of the K6 (Goldberg, Oldehinkel,
and Ormel, 1998). As a result, specifying a single threshold for SMI on the
K6 that is applied to all samples is not appropriate.
This problem can be solved, but doing so requires access to a much larger calibration
sample than the 155 cases included in the SAMHSA methodological study. When
such a sample is available, scoring rules can be developed that allow predicted
probabilities of SMI to be estimated for respondents in other samples, such
as the NHIS or NHSDA, on the basis of the assumption of consistent sensitivity
and specificity of the K6 across samples. This assumption is much more plausible
than the assumption of consistent positive and negative predictive values. The
assumption of consistent sensitivity and specificity can be made either for
an entire sample or for important subsamples (e.g., gender, age, educational
attainment). In the ideal case, subsample differences in the calibration sample
should be analyzed to evaluate the plausibility of assuming that sensitivity
and specificity are constant across subsamples and, when this assumption is
rejected, to estimate separate sensitivity and specificity values in informative
subsamples (Furukawa, et al., 2003).
The assumption of sensitivity and specificity being constant across samples
is equivalent to the assumption that stratum-specific likelihood ratios (SSLRs)
are constant, but the parameterization of the assumption in terms of SSLRs has
computational advantages over the parameterization in terms of sensitivities
and specificities. An SSLR is an odds ratio (OR) that compares respondents who
have a specific score on a screening scale (in this case, the K6) with those
who have all other scores on the scale in terms of their odds of having a dichotomous
outcome (in this case, SMI) (Guyatt and Rennie, 2001). Once the SSLRs are calculated,
Bayes' theorem can be used to show that
where POO is the population odds of the dichotomous outcome (which can be calculated
from the population prevalence) and ROO is the individual respondent's odds
of the outcome. The individual's probability of the dichotomous outcome, p,
can easily be derived from ROO by using the transformation
The results in equations (1) and (2) can be used to assign individual-level
predicted probabilities of SMI based on individual K6 scores, but only after
knowing the prevalence of SMI in the population. In other words, equations (1)
and (2) make it clear that the probability of SMI for any one individual with
a given score on the K6 scale varies with the prevalence of SMI in the population
from which that individual was drawn. The real problem, then, is to estimate
the aggregate prevalence. Fortunately, this can be done using a method that
avoids the problem of having to assume constant positive predictive value. This
method uses maximum likelihood to compare the empirical K6 distribution in the
sample under consideration with the theoretical distributions generated by the
sensitivities and specificities in some calibration sample applied to all possible
hypothetical prevalences of SMI. The maximum- likelihood estimate of SMI is
the one associated with the theoretical distribution of K6 scores most similar
to the empirical distribution in the sample. Once this prevalence estimate is
obtained, it can be used in equations (1) and (2) to generate individual- level
probabilities from individual-level K6 scores for purposes of more detailed
analyses of the correlates of SMI. The latter can include the estimation of
prevalences in counties and States.
We are unable to implement this optimal scoring approach because the SAMHSA
methodological study sample is too small to allow the K6 prevalence to be estimated
accurately in the NHIS. However, this problem is being solved by including the
K6 in the National Comorbidity Survey Replication (NCS- R) (Kessler and Walters,
2003), a nationally representative face-to-face general population survey of
10,000 respondents that was carried out in 2001- 2002. The full NCS-R sample
will be used as a calibration sample to generate accurate estimates of the prevalence
of SMI from K6 distributions in the NHIS and NHSDA. Scoring rules, as soon as
they are available, will be posted on the NCS Web site (http://www.hcp.med.harvard.edu/ncs/).
The diagnostic interview in the NCS-R is a modification of the WHO CIDI (Robins
et al., 1988), a fully structured diagnostic interview that assesses a number
of commonly occurring mental disorders according to the definitions and criteria
of DSM-IV. Respondents are classified as having SMI if they met criteria for
any qualifying DSM-IV mental disorder during the year before the interview and
if they either scored less than 60 on a structured version of the GAF scale
or some equivalent indicator of serious impairment associated with their mental
illness during the past year. Because the NCS-R sample is fairly large, it will
also be used to explore whether the SSLRs for the relationship between the K6
and SMI are stable across important sociodemographic subsamples and, if not,
to generate separate sets of SSLRs in informative subsamples, as well as to
collapse K6 scores in ways that optimize the stability of imputations (Peirce
and Cornell, 1993). A computer program that generates both an aggregate estimate
of the prevalence of SMI using maximum likelihood and individual estimates of
the probability of having SMI on the basis of equations (1) and (2) will be
posted on the NCS Web site as soon as the NCS-R data are ready for analysis
and the K6 calibration has been completed.
Preliminary SMI
Prevalence Estimates
Even before the optimal scoring approach is worked out from the NCS-R, it is
possible to present very preliminary aggregate estimates of the prevalence of
SMI from the NHIS using the simple classification rule that K6 scores in the
range 13 to 24 represent likely SMI. On the basis of this rule, the estimated
30-day prevalence of likely SMI in the NHIS is 3.3 percent in 1997, 3.0 percent
in 1998, 2.4 percent in 1999, and 2.7 percent in 2000. With a sample of more
than 40,000 respondents each year, the standard errors of these annual estimates
are less than one-tenth of one percent. This means that the 0.9 percent range
of year-to-year variation in the prevalence of SMI over these years is reliable.
On a population base of approximately 209 million adults ages 18 and older in
the United States, these prevalence estimates are equivalent to an average of
between 5.0 million and 6.9 million Americans who met criteria for SMI in any
given month of these years.
It would be possible to generate more finely disaggregated NHIS time-trend
prevalence estimates to see if there is seasonal variation in SMI, because the
NHIS is made up of separate nationally representative samples of about 800 interviews
with adult respondents each week of the year. It would be useful to link these
fine-grained time trends to information about current events (e.g., stock market
trends, extreme weather patterns, widely publicized world events that might
lead to anxiety or depression) in an effort to get insight into reasons for
the statistically significant changes in the prevalence of likely SMI over the
years 1997 to 2000. Given that the NCS-R calibration rules for transforming
K6 scores into estimated probabilities of SMI will soon be available, this analysis
of detailed time trends should probably be based on these calibrations rather
than on the 0 to 12 versus 13 to 24 dichotomous scoring of the K6.
As noted above, the K6 was modified in the SAMHSA methodological study to ask
about symptoms in the month in the past year when the respondent's emotions
were worst rather than in the past 30 days. This was done because PL 1021321
defines SMI as a 12-month disorder, and the goal of the methodological study
was to screen for the presence of SMI at any time in the year before the interview.
The NHSDA adopted this K6 modification. We should consequently expect that the
prevalence estimates of SMI in the NHSDA, when they become available, will be
higher than in the NHIS because the NHSDA will be estimating 12-month prevalence,
while the NHIS estimates 30-day prevalence. A glimpse into this prediction is
provided by preliminary data from the first half sample of 5,000 respondents
in the NCS-R, 3,015 of whom were administered the K6 questions in exactly the
same way as in the NHSDA. The estimated 12-month prevalence of likely SMI in
this preliminary sample is 7.2 percent, with a standard error of 0.5 percent.
As expected, this is considerably higher than the 2.4 to 3.3 percent 30-day
prevalence estimates in the NHIS. On a base of 209 million Americans ages 18
and older, the 12-month SMI prevalence estimate in the preliminary NCS-R data
is equivalent to approximately 15 million Americans who met criteria for SMI
at some time in the year prior to their participation in the survey.
A decomposition of the preliminary NCS-R SMI data by age and sex is shown in
the first column of table 1. Twelve-month
SMI is estimated to be more prevalent among women than men (8.6 percent vs.
5.6 percent; z = 3.25, p = 0.001). Among men, 12- month SMI is
estimated to be most prevalent in the age range of 30 to 44 (7.2 percent) and
least prevalent in the age range of 60 and older (3.0 percent). Among women,
12-month SMI is estimated to be more prevalent in the age range of 18 to 44
(10.2 to 10.6 percent) and, like men, least prevalent in the age range of 60
and older (4.8 percent). Because of the great deal of interest in the comorbidity
between SMI and SUDs, table 1 also
shows preliminary estimates for the 12-month prevalence of DSM-IV alcohol or
drug abuse or dependence. These SUDs have an estimated 12-month prevalence of
5.0 percent in the total sample, with a standard error of 0.4 percent. It is
noteworthy that the DSM-IV criteria for SUD are more restrictive than the criteria
in earlier versions of the DSM. This fact partly explains why the estimated
prevalence of SUD is lower in the NCS-R than in either the ECA (Robins and Regier,
1991), which was based on DSM-III criteria, or the NCS (Kessler et al., 1994),
which was based on DSM-III-R criteria. As in these earlier surveys, though,
the 12-month prevalence of SUD is estimated to be much higher among men than
women (7.7 percent vs. 2.5 percent; z = 6.4, p < 0.001). Also
consistent with earlier surveys is the strong inverse relationship between age
and SUD in the NCS-R. The prevalence range of SUD among NCS-R men is from 16.2
percent in the youngest age group (ages 18 to 29) to 0.2 percent in the oldest
age group (60 and older). The comparable range among women is from 6.3 percent
in the youngest age group to 0.2 percent in the oldest age group.
If SMI and SUDs were totally unrelated to each other, the prevalence of comorbid
SMI and SUD would be the product of the prevalences of the two individual disorders.
For example, with total-sample prevalences of 7.2 and 5.0 percent for SMI and
SUD, respectively, we would expect that 0.4 percent (.072 x .05) of the population
would meet criteria for both SMI and SUD. The actual estimated prevalence of
comorbid SMI-SUD is 0.8 percent-twice as high as the prediction based on independence.
This lack of independence is consistent with the findings of previous research
(Kessler et al., 1996). The last column of table
1 shows ORs between 12-month SMI and 12-month SUD. All the ORs are greater
than 1.0, indicating that the prevalence of comorbid SMI-SUD is consistently
larger than the product of the component prevalences. Overall, 10.6 percent
of respondents with likely SMI also have SUD, compared with 4.6 percent of respondents
without SMI. Among respondents with SUD, 18.1 percent of those with substance
dependence also have SMI, compared with 12.4 percent of those with substance
abuse. Among those without any 12-month SUD, 6.8 percent have SMI.
Sociodemographic Correlates of SMI, Substance Disorders, and Their Comorbidity
Table 2 shows the results of logistic
regression analysis, in which we examined basic sociodemographic predictors
of the outcomes in table 1. The
first four rows show the age associations in a somewhat different way than in
table 1. Significant relationships
with age are found both for SMI and SUD. In the case of SMI, the highest prevalence
is in middle age, whereas for SUD, the highest prevalence is in young age. The
age gradient is also much stronger for SUD than for SMI, leading to an inverse
relationship between age and comorbid SMI-SUD. The next entries in the table
are for sex. Women are significantly more likely than men to meet criteria for
likely SMI, but significantly less likely than men to meet criteria for SUD.
The negative association with SUD is stronger than the positive association
with SMI, leading to a negative, although nonsignificant, association with comorbid
SMI-SUD.
The next entries in table 2 are
for education, which is inversely related to SMI, to SUD, and to comorbid SMI-SUD.
The associations are essentially multiplicative in predicting comorbid SMI-SUD.
That is, the product of the ORs in predicting SMI and SUD is roughly equivalent
to the ORs in predicting comorbid SMI-SUD. A rather different pattern is found
in the next entries in the table, which deal with marital status. Married people
consistently have the lowest rates of disorder, with separated- divorced and
never married people having elevated rates of SMI and these same two groups
plus cohabiters having elevated rates of SUD. The ORs in predicting comorbid
SMI-SUD are roughly multiplicative for the never married, but for cohabiters
and the separated-divorced they deviate substantially from this pattern. In
the case of cohabiters, there is no elevated risk of comorbid SMI-SUD (OR =
1.3; 95 percent confidence interval = 0.1-17.5), despite the fact that cohabiters
have significantly elevated rates of both SMI and SUD. In the case of the separated-divorced,
the risk of comorbid SMI-SUD (OR = 17.9; 95 percent confidence interval = 4.0-80.0)
is substantially higher than the product of the ORs for SMI and SUD.
The remaining entries in table 2
show that SMI is not significantly related either to race-ethnicity, region,
or urbanicity net of the effects of the other predictors. The failure to find
an effect of race-ethnicity is striking in light of concerns among health care
policy researchers about health disparities. This finding might be taken to
suggest that no such disparities exist with regard to serious emotional or substance
problems among racial and ethnic minorities after adjusting for the other variables
in this model. It is important to note, however, that previous research has
sometimes found interactions between socioeconomic status (e.g., education,
income) and race-ethnicity in predicting mental health problems (House, 2002).
It may be that the simple additive model estimated here, in which the effects
of education are assumed to be the same for people in all race-ethnicity subsamples,
masks important effects of minority status that are confined to the sector of
the population with low education.
A similar caution can be raised about the failure to find an effect of urbanicity
in predicting any of the outcomes in table
2. Considerable concern exists about rural mental health (Hartley et al.,
2002), and especially about the possibility that unmet need for treatment is
higher in rural areas than urban areas because of low access to specialty care
(Bull, Krout, Rathbone-McCuan, and Shreffler, 2001). Our failure to find evidence
of significantly elevated SMI or SUD in rural areas is based on a simple additive
prediction equation. Perhaps the rural poor or rural minorities have elevated
rates of these disorders that could be detected in more detailed analyses. Such
analyses were not carried out with the NCS-R data because of low statistical
power in the preliminary half-sample data. These analyses are needed, however,
when the full NCS-R data set is available. In addition, much more precise data
for carrying out these subsample analyses will soon be available in the NHIS
and NHSDA once K6 calibration rules are developed on the basis of the full NCS-R
data set.
Needs Assessment for Counties
K6 calibration rules based on the full NCS-R data set will be available late
in 2003. It will become possible at that time to apply these rules to the data
from the NHIS and NHSDA, as well as to other surveys in which the assumption
of consistency of SSLRs is plausible. Given what we know of the strength of
sociodemographic correlations of SMI and intercounty variation in these correlates,
the precision of these estimates will almost certainly come largely from the
direct aspect of the estimates; that is, from the number of cases with K6 scores
that were assessed in the separate counties and States included in the samples.
The number of respondents in even the smallest States will be large enough to
generate stable State-level estimates of the prevalence of SMI. For example,
with a sample of approximately 250,000 respondents (which could be obtained
by pooling the NHIS over 6 years, pooling the NHSDA over 4 years, or pooling
and blending the samples from these two surveys over 2 to 3 years), a subsample
of approximately 500 respondents will be included from each of the three least
populated States (Alaska, Vermont, Wyoming). This is a large enough sample to
generate a fairly stable estimate of the prevalence of SMI from K6 scores. However,
the sample will be too small to estimate the prevalence of SMI in counties.
For example, there will be fewer than 50 respondents in the pooled sample for
a county of 50,000 inhabitants. Close to one-fifth of all Americans live in
counties with fewer than 50,000 inhabitants.
Mental health needs assessment data are needed at the county level to allow
States and counties to make rational resource allocation decisions. One way
to obtain such data would be for the Behavioral Risk Factor Surveillance Survey
(BRFSS), which is carried out with coordination from the Centers for Disease
Control (CDC) every year in each State, to include the K6. BRFSS samples typically
include at least 1,500 respondents per State per year. With a sample of this
size, it would be possible to generate useful SMI prevalence estimates for groups
of counties based on pooled BRFSS data collected over several years. However,
the BRFSS currently does not include the K6. The BRFSS includes a small number
of mental health screening questions. These questions were evaluated in an early
phase of the SAMHSA methodological study and found to be of little value. It
would be very useful for the K6 to be substituted for these questions in the
BRFSS.
Even the inclusion of the K6 in the BRFSS, however, would not give individual
counties the detailed data they need to make informed resource allocation decisions.
An intriguing possibility for collecting such county-level data on SMI is to
carry out a low-tech county-level telephone needs assessment survey based on
the design and methods used in volunteer political polls for local and State
elections. In the latter, pollsters who specialize in carrying out local and
State election surveys throughout the country develop a preset telephone interview
schedule that is known to cover the important issues for such elections, with
the possible addition of modest modifications for a particular survey. This
interview is programmed onto laptop computers that are shipped to local areas.
Local volunteers are then trained to carry out the interview on these computers.
The interviews are usually conducted at a temporary phone survey facility (e.g.,
a church basement, a community center) over a period of 1 or 2 weeks. A team
of 20 to 40 volunteer telephone interviewers, each working 10 to 20 hours, can
usually complete 500 to 800 telephone interviews over this period. The laptop
computers, still containing the completed survey data, are then shipped back
to the pollster and a report is prepared.
There is no reason why exactly the same approach could not be used to carry
out inexpensive county-level mental health needs assessment surveys. These surveys
could include the K6 to screen for SMI along with brief screens for other issues
of interest to community mental health planners (e.g., substance problems, suicidality,
relationship violence) and questions about barriers to care. A local civic or
church group could be recruited to carry out the survey as a voluntary activity.
The local telephone company could be asked to donate the cost of installing
a temporary telephone bank for the time it would take to carry out the survey.
The survey could be repeated every 5 years or so to monitor trends in the prevalence
of SMI and service needs for people with SMI.
Results of such surveys could be of great value to county mental health service
planners. It would not be difficult to create an interview schedule for these
surveys, a laptop computer program that could be used to record interview data,
or a standard electronic report format to synthesize and communicate survey
results to local mental health policymakers. The only real constraint is in
obtaining a large enough stockpile of laptop computers to carry out the surveys
on a production basis in a number of counties at a time. The ideal situation
would be for SAMHSA to underwrite the purchase of these computers and to allow
them to be used on a rotating basis by counties throughout the country to carry
out local needs assessment surveys.
Summary
This chapter reviewed the research done on the prevalence and correlates of
SMI since that research was first advanced as a concept in PL 102 321. Preliminary
prevalence estimates based either on post hoc secondary analysis of previously
collected psychiatric epidemiological surveys or on indirect estimates from
screening scales suggest that 2.4 to 3.3 percent of the U.S. adult population
meet criteria for SMI in any given month and that 5.4 to 7.2 percent do so at
some time during the year. The demographic risk profile for SMI includes being
female, young or middle-aged, unmarried, and of low socioeconomic status. SMI
is significantly related to SUDs, although most people with SMI do not have
a co-occurring SUD. A brief screening scale for SMI has been developed and is
now in use in two of the three large government health surveys that are carried
out on an annual basis in the United States: the NHIS and the NHSDA. The third
large annual government health survey, the BRFSS, does not include this screening
scale, but we recommend that the screening scale be included in this survey
as well. Sophisticated estimation and calibration procedures to convert these
screening scale scores into aggregate prevalence estimates as well as individual-level
estimates of the predicted probability of SMI have been developed and will soon
be applied to a large calibration survey that is currently under way. These
data and methods will make it possible to track trends and subgroup differences
in the prevalence of SMI with excellent precision as well as to make reliable
estimates for each State. County-level estimates, in comparison, cannot be made
with precision using these methods. A procedure that addresses this problem
was described for conducting periodic inexpensive county-level needs assessment
surveys. The technology exists for carrying out these small-area surveys efficiently,
but coordination and access to shared computer hardware and infrastructure are
needed to make such surveys feasible.
REFERENCES
Bull, C. N., Krout, J. A., Rathbone-McCuan, E., & Shreffler, M. J. (2001).
Access and issues of equity in remote/ rural areas. Journal of Rural Health,
17, 356-359.
Endicott, J., Spitzer, R. L., Fleiss, J. L., & Cohen, J. (1976). The Global
Assessment Scale: A procedure for measuring overall severity of psychiatric
disturbance. Archives of General Psychiatry, 33(6), 766-771.
First, M., Spitzer, R., Gibbon, M., & Williams, J. (1997). Structured clinical
interview for DSM-IV axis I disorders, research version, non-patient edition
(SCID-I/ NP). New York: Biometrics Research, New York State Psychiatric Institute.
Furukawa, T. A., Andrews, G., Slade, T., & Kessler, R. C. (2003). The performance
of the K6 and K10 screening scales for psychological distress in the Australian
National Survey of Mental Health and Well-Being. Psychological Medicine,
33(2), 357-362.
Goldberg, D. P., Oldehinkel, T., & Ormel, J. (1998) Why GHQ threshold varies
from one place to another. Psychological Medicine, 28(4), 915-921.
Guyatt, G., & Rennie, D. (2001). User's guide to the medical literature:
A manual for evidence-based clinical practice. Chicago: AMA Press.
Hartley, D., Britain, C., & Sulzbacher, S. (2002). Behavioral health: Setting
the rural health research agenda. Journal of Rural Health, 18, 242-255.
House, J. S. (2002). Understanding social factors and inequalities in health:
20th century progress and 21st century prospects. Journal of Health and Social
Behavior, 43(2), 125-142.
Kessler, R. C., Andrews, G., Colpe, L. J., Hiripi, E., Mroczek, D. K., Normand,
S.-L. T., et al. (in press). Short screening scales to monitor population prevalences
and trends in nonspecific psychological distress. Psychological Medicine.
Kessler, R. C., Andrews, G., Mroczek, D., Ustun, T. B., & Wittchen, H. U. (1998).
The World Health Organization composite international Diagnostic Interview Short
Form (CIDI-SF). International Journal of Methods in Psychiatric Research,
7, 171-185.
Kessler, R. C., Barker, P. R., Colpe, L. J., Epstein, J. F., Gfroerer, J. C.,
Hiripi, E., et al. (2003). Screening for serious mental illness in the general
population. Archives of General Psychiatry, 60(2), 184-189.
Kessler, R. C., Berglund, P. A., Walters, E. E., Leaf, P. J., Kouzis, A. C.,
Bruce, M. L., et al. (1999). Population- based analyses: A methodology for estimating
the 12- month prevalence of serious mental illness. In Manderscheid, R. W. &
M. J. Henderson (Eds.), Mental Health, United States, 1998 (pp. 99-109).
Washington, D.C.: U.S. Government Printing Office.
Kessler, R. C., Berglund, P. A., Zhao, S., Leaf, P. J., Kouzis, A. C., Bruce,
M. L. et al. (1996). The 12-month prevalence and correlates of serious mental
illness (SMI). In Sonnenschein, M. A. (Ed.), Mental Health, United States,
1996 (pp. 59-70). Washington, D.C.: U.S. Dept. of Health and Human Services.
Kessler, R. C., McGonagle, K. A., Zhao, S., Nelson, C. B., Hughes, M., Eshleman,
S., et al. (1994). Lifetime and 12-month prevalence of DSM-III-R psychiatric
disorders in the United States: Results from the National Comorbidity Survey.
Archives of General Psychiatry, 51(1), 8-19.
Kessler, R. C., Nelson, C. B., McGonagle, K. A., Edlund, M. J., Frank, R. G.,
& Leaf, P. J. (1996). The epidemiology of co-occurring addictive and mental
disorders: Implications for prevention and service utilization. American
Journal of Orthopsychiatry, 66, 17-31.
Kessler, R. C., & Walters, E. E. (2003). The National Comorbidity Survey. In
Tsaung, M. T., M. Tohen, & G. E. P. Zahner (Eds.), Textbook in psychiatric
epidemiology, second edition. New York: John Wiley & Sons.
Peirce, J. C., & Cornell, R. G. (1993). Integrating stratum- specific likelihood
ratios with the analysis of ROC curves. Medical Decision Making, 13(2),
141-151.
Rehm, J., Ustun, T. B., Saxena, S., Nelson, C. B., Chatterji, S., Ivis, F.,
et al. (1999). On the development and psychometric testing of the WHO screening
instrument to assess disablement in the general population. International
Journal of Methods in Psychiatric Research, 8, 110-123.
Robins, L. N., & Regier, D. A. (Eds.). (1991). Psychiatric disorders in
America: The Epidemiologic Catchment Area study. New York: The Free Press.
Robins, L. N., Wing, J., Wittchen, H.-U., Helzer, J. E., Babor, T. F., Burke,
J. D., et al. (1988). The Composite International Diagnostic Interview: An epidemiologic
instrument suitable for use in conjunction with different diagnostic systems
and in different cultures. Archives of General Psychiatry, 45, 1069-1077.
Schaible, W. (1996). Indirect estimators in U.S. Federal Programs. New
York: Springer.
Substance Abuse and Mental Health Services Administration. (1993). Final notice
establishing definitions for (1) children with a serious emotional disturbance,
and (2) adults with a serious mental illness. Federal Register, 58(96),
29,422-29,425.
World Health Organization. (2001). International classification of functioning,
disability, and health. Geneva, Switzerland.
FOOTNOTE
The developmental research for the CIDI-SF scales was supported by National
Institute of Mental Health (NIMH) grants MH46376, MH49098, and MH52861, with
supplemental support from the W.T. Grant Foundation (grant 90135190). The developmental
research for the K10/K6 scales was supported by the John D. and Catherine T.
MacArthur Foundation Network on Successful Midlife Development (Gilbert Brim,
director). The SMI screening scale project was supported by Contract 283-98-9008
from SAMHSA. The implementation of K10/K6 calibration in the new NCS-R is supported
by NIMH Collaborative Agreement U01-MH60220 with supplemental support from the
National Institute of Drug Abuse and the Robert Wood Johnson Foundation (grant
044780). In addition to the above grants and contracts, preparation of this
report was supported by NIMH grant MH066627.
Index | Previous
| Next
|
|
