Alzheimer's Disease: Estimates of Prevalence in the United States (Letter
Report, 01/28/98, GAO/HEHS-98-16).

Pursuant to a congressional request, GAO provided estimates of the
prevalence of Alzheimer's Disease (AD) in the United States, including
projections for prevalence in the near future.

GAO noted that: (1) on the basis of existing studies, it is estimated
that at least 1.9 million Americans 65 years of age or older suffered
from any level of AD--mild, moderate, or severe--in 1995; (2) this
number would be closer to 2.1 million if adjusted for the omissions, in
many of these studies, of cases with mixed dementia and cases missed by
the screening instruments used; (3) when only people likely to need at
least some active assistance with personal care are considered, the
results of GAO's meta-analysis show slightly more than 1 million people
with AD over the age of 64; (4) the result would be higher--closer to
1.4 million people with AD--if all mixed cases and missed cases were
included in all studies; (5) consistent with all earlier research, the
results for both any AD and moderate or severe AD demonstrate that the
prevalence rates increase sharply with age, doubling about every 5
years, at least until the age of 85, when the increase begins to slow;
(6) also consistent with some earlier research, the estimated rates for
women are higher than for men; (7) projecting the number of people with
AD into the future gives some indication of the long-term care and
research challenges that will face the nation as people grow older; (8)
GAO's meta-analysis, when combined with Bureau of the Census
projections, shows that more than 2.9 million people would have at least
a mild case of AD in 2015; of these, over 1.7 million would need active
assistance in personal care; (9) these figures jump to 3.2 million and
2.1 million, respectively, when mixed cases and missed cases are
included; (10) given the uncertainty surrounding existing estimates of
AD, a number of studies are now under way, supported by the National
Institute on Aging, that should yield better prevalence estimates of
African-Americans, Hispanics, and other nonwhite subpopulations; and
(11) the results of these studies, expected to be published over the
next several years, should improve GAO's picture of AD prevalence for
the United States as a whole, as well as for the specific population
segments studied.

--------------------------- Indexing Terms -----------------------------

 REPORTNUM:  HEHS-98-16
     TITLE:  Alzheimer's Disease: Estimates of Prevalence in the United 
             States
      DATE:  01/28/98
   SUBJECT:  Neurological diseases
             Disease detection or diagnosis
             Projections
             Elderly persons
             Health statistics
             Medical research
             Statistical data
             Statistical methods
             Health care planning
IDENTIFIER:  Alzheimer's Disease
             Medicare Program
             Medicaid Program
             
******************************************************************
** This file contains an ASCII representation of the text of a  **
** GAO report.  Delineations within the text indicating chapter **
** titles, headings, and bullets are preserved.  Major          **
** divisions and subdivisions of the text, such as Chapters,    **
** Sections, and Appendixes, are identified by double and       **
** single lines.  The numbers on the right end of these lines   **
** indicate the position of each of the subsections in the      **
** document outline.  These numbers do NOT correspond with the  **
** page numbers of the printed product.                         **
**                                                              **
** No attempt has been made to display graphic images, although **
** figure captions are reproduced.  Tables are included, but    **
** may not resemble those in the printed version.               **
**                                                              **
** Please see the PDF (Portable Document Format) file, when     **
** available, for a complete electronic file of the printed     **
** document's contents.                                         **
**                                                              **
** A printed copy of this report may be obtained from the GAO   **
** Document Distribution Center.  For further details, please   **
** send an e-mail message to:                                   **
**                                                              **
**                                            **
**                                                              **
** with the message 'info' in the body.                         **
******************************************************************


Cover
================================================================ COVER


Report to the Secretary of Health and Human Services

January 1998

ALZHEIMER'S DISEASE - ESTIMATES OF
PREVALENCE IN THE UNITED STATES

GAO/HEHS-98-16

Alzheimer's Disease Prevalence

(108317)


Abbreviations
=============================================================== ABBREV

  AD - Alzheimer's disease
  ADRDA - Alzheimer's Disease and Related Disorders Association
  CDR - Clinical Dementia Rating
  CT - computerized tomography
  NIA - National Institute on Aging
  NIH - National Institutes of Health
  NINCDS - National Institute of Neurological and Communicative
     Disorders and Stroke

Letter
=============================================================== LETTER


B-277607

January 28, 1998

The Honorable Donna Shalala
Secretary of Health and Human Services

Dear Madame Secretary: 

Alzheimer's disease (AD) is a neurological disease, devastating to
patients and their families and friends.  It is expected to cost this
country billions of dollars annually, with severe consequences to
patients and informal caregivers as well as to the health care
system.\1 Valid estimates of disease prevalence--that is, how many
people have a disease at a given time--as well as projections of
prevalence in the future, can play an important role in setting
health research priorities and in identifying the need for services. 
A number of published studies have cited AD prevalence rates that
range from 2 percent to 12 percent of the population older than 64. 

Meta-analysis is an analytic method that can be used, as we do in
this report, to (1) estimate the number of people with AD, (2)
project the numbers of people with AD in the near future, and (3)
determine the prevalence rates of AD for both men and women of
specific ages.  Meta-analysis uses existing studies to quantitatively
integrate prevalence data.  This approach uses data from all relevant
studies to derive prevalence estimates that are not likely to be
affected by the biases of any individual studies.  Because of
limitations associated with the estimates presented in this report,
we also identify ongoing efforts by the National Institute on Aging
(NIA) to develop more accurate prevalence estimates, including ones
applicable to racial and ethnic minorities. 

Our analysis was developed from 18 studies (see app.  I) that (1)
estimate prevalence rates for populations that are considered
relevant to the United States and (2) use widely accepted diagnostic
criteria for finding cases of AD (see app.  II for a list of the
criteria).  Many of the studies we relied on are of European
populations.  The major limitation of the studies is that they
include mainly whites and few members of the other racial groups that
are part of the U.S.  population, such as African-Americans and
Asian-Americans.  However, since whites currently represent a
majority (about 83 percent of the population in 1997), we would not
expect our results to change substantially if others were represented
in proportion to their numbers. 

Using meta-analysis, we integrated the prevalence estimates in order
to provide age- and gender-specific estimates of AD prevalence rates. 
We used the prevalence rates, in combination with population
estimates and projections from the Bureau of the Census, to estimate
(1) the number of Americans in 1995 with AD at any level of severity
and those with moderate or severe AD and (2) the number of Americans
likely to have the disease in the near future. 

We also adjusted the overall AD prevalence estimates to account for
two limitations of the age- and gender-specific rates presented in
the literature.  Specifically, we estimated the AD prevalence rates,
over all ages and both genders, that would be obtained if all studies
(1) counted cases of mixed dementia--persons with both AD and another
kind of dementia--as cases of AD and (2) corrected the estimates,
when necessary, for the expected number of cases missed by
insufficiently sensitive screens. 

At all stages of our work, we consulted with experts in the fields of
AD, disease measurement, and the quantitative integration of research
results, including experts at the National Institutes of Health
(NIH).  Details of our methods for (1) defining and locating the most
relevant studies, (2) extracting prevalence rates from them, and (3)
integrating these rates quantitatively are provided in appendixes
III, IV, and V, respectively.  Our evaluation of some of the
strengths and limitations of the studies integrated is presented in
appendix VI.  Our adjustments to the data are described in appendix
VII. 

We conducted our study from May 1995 to December 1997 in accordance
with generally accepted government auditing standards. 


--------------------
\1 J.  W.  Hay and R.  L.  Ernst, "The Economic Costs of Alzheimer's
Disease," American Journal of Public Health, Vol.  77 (1987), pp. 
1169-75. 


   RESULTS IN BRIEF
------------------------------------------------------------ Letter :1

Recognizing the limitations associated with such estimates, our
meta-analysis shows that at least 1.9 million Americans 65 years of
age or older suffered from any level of AD--mild, moderate, or
severe--in 1995.  This number would be closer to 2.1 million if we
adjusted for the omissions, in many of these studies, of cases with
mixed dementia and cases missed by the screening instruments used. 
Most of the estimated 1.9 million cases--58 percent, or 1.1
million--are among those in the 75-89 age group.  When we calculated
comparable projections of the number of cases in this age group based
on the data from individual studies, the values ranged from about
700,000 to 3.2 million, with most studies yielding values of 1.4
million or less. 

When only people likely to need at least some active assistance with
personal care are considered (those with moderate or severe AD), the
results of our meta-analysis show slightly more than 1 million people
with AD over the age of 64.  Here, too, the result would be
higher--closer to 1.4 million people with AD--if all mixed cases and
missed cases were included in all studies.  Consistent with all
earlier research, the results for both any AD and moderate or severe
AD demonstrate that the prevalence rates increase sharply with age,
doubling about every 5 years, at least until the age of 85, when the
increase begins to slow.  Also consistent with some earlier research,
the estimated rates for women are higher than for men. 

Projecting the number of people with AD into the future gives some
indication of the long-term care and research challenges that will
face this nation as people grow older.  Our meta-analysis, when
combined with Bureau of the Census projections, shows that more than
2.9 million people would have at least a mild case of AD in 2015; of
these, more than 1.7 million would need active assistance in personal
care.  These figures jump to 3.2 million and 2.1 million,
respectively, when mixed cases and missed cases are included. 

Given the uncertainty surrounding existing estimates of AD, a number
of studies are now underway, supported by NIA, that should yield
better prevalence estimates of African-Americans, Hispanics, and
other nonwhite subpopulations.  The results of these studies,
expected to be published over the next several years, should improve
our picture of AD prevalence for the United States as a whole, as
well as for the specific population segments studied. 


   BACKGROUND
------------------------------------------------------------ Letter :2

AD is a kind of dementia, with the essential feature the development
of multiple cognitive deficits, including memory impairment, and at
least one other deficit, such as impaired language functioning
(aphasia).  The definition of dementia also requires that the
condition be severe enough to cause a significant impairment in
social or occupational functioning that represents a decline from a
previous level of functioning.  Common clinical signs of dementia
include emotional and behavioral disturbances. 

AD is a dementia of gradual onset and progressive decline.  It may be
difficult to distinguish clinically between mild--that is, early--AD
and normal aging, but severe AD is characterized by a need for much
help with personal care as the result of incontinence and almost
total lack of comprehension of the environment.  AD is said to be
differentiated from other dementias on the basis of its cause, but
that cause is not, in fact, well understood.  That AD is accepted as
a distinct disease entity is because AD patients manifest specific
kinds of abnormalities in the brain--observable only in those who are
autopsied or undergo a brain biopsy, a rare procedure--differing from
the abnormalities found in other dementias with better understood
causes. 


      MEASURING THE EXTENT OF AD
---------------------------------------------------------- Letter :2.1

The prevalence of AD is defined as the number of people in a specific
population who suffer from the disease at some specified time.  It is
often expressed as a rate, the number of cases of that disease
existing at a given point in time divided by the total population at
that same time.  When the number of cases in the target population is
too costly to count, prevalence may be estimated by testing, in a
prevalence survey, a representative sample of the population.  An
alternative is to develop a register of people seeking services, but
this does not work well for AD because many cases are not treated. 

The restriction of AD prevalence surveys to the elderly population,
65 years of age or older, makes sense because the majority of cases
of AD are elderly.  In addition, because AD prevalence tends to
increase sharply with age, doubling about every 5 years, at least
over the age range of 65 to 85 years, it is common to estimate
age-specific prevalence rates of AD.\2 Thus, the population studied
is divided into age groups--for example, 65 to 69 years, 70 to 74
years, 75 to 79 years--and a prevalence rate is estimated for each
group.  This is especially important when comparing AD rates across
groups so that differences in prevalence stemming from differences in
age distribution can be separated from those stemming from real
health differences. 

Although the dependence of AD rates on gender is not as well
established in the scientific literature as dependence on age, there
is some tendency for prevalence to be greater for women than for men;
therefore, rates that are specific to both age and gender are of
interest.  The reasons for this tendency are not known.  It may be
that (1) women are more likely than men to contract the disease or
(2) women live longer once they get AD and are therefore more likely
to be counted when a prevalence survey is conducted or (3) both of
these factors operate. 


--------------------
\2 R.  Katzman and C.  Kawas, "The Epidemiology of Dementia and
Alzheimer Disease," in Alzheimer Disease, edited by R.  D.  Terry, R. 
Katzman, and K.  L.  Bick (New York:  Raven, 1994). 


      MEASURING THE SEVERITY OF AD
---------------------------------------------------------- Letter :2.2

When severity is measured, people with AD are categorized into
degrees or levels of illness; thus, for every specific prevalence
rate, several categories are used to describe the different levels of
severity.  For the dementias (including AD), descriptive categories
like mild, moderate, and severe are commonly used to indicate how
severe a person's AD is.  Sometimes a borderline category,
"questionable," is used for those whose impairment is not great
enough to qualify as even mild AD.  Today, there are standardized
systems for rating severity, often using such descriptive categories. 
According to one system, the Clinical Dementia Rating (CDR), a person
with mild AD needs only prompting in personal care activities, a
person with moderate AD needs some assistance, and a person with
severe AD needs much assistance and is frequently incontinent.\3


--------------------
\3 See, for example, L.  Berg, "Mild Senile Dementia of the Alzheimer
Type:  Diagnostic Criteria and Natural History," The Mount Sinai
Journal of Medicine, Vol.  55 (Jan.  1988), pp.  87-96. 


   ESTIMATED NUMBER OF PEOPLE WITH
   AD
------------------------------------------------------------ Letter :3

The results of our meta-analysis, based on the pooling of the data of
individuals from each of the studies, were used to project the
numbers of Americans, in 1995, with (1) AD of any level of severity
(mild, moderate, or severe) and (2) moderate or severe AD (see table
1).  The AD prevalence estimates are generated by multiplying
prevalence rates (discussed in the section below called "Estimated
Prevalence Rates") by the corresponding age- and gender-specific 1995
estimates from the Bureau of the Census.\4 When these results are
summed over the several age intervals and both genders, the overall
estimate for Americans 65 years of age or older with any AD is 1.9
million.  Of these 1.9 million cases, an estimated 1.1 million have
moderate or severe AD. 



                                Table 1
                
                  Estimates of Any AD and Moderate or
                Severe AD for Americans 65 Years of Age
                            or Older in 1995

                                                         Moderate or
                                            Any AD        severe AD
                                        --------------  --------------
                                                Percen          Percen
Age                                     Number       t  Number       t
--------------------------------------  ------  ------  ------  ------
65-69                                   104,78     1.1  61,815     0.6
                                             5
70-74                                   194,71     2.2  111,11     1.3
                                             6               1
75-79                                   304,39     4.6  169,54     2.5
                                             9               9
80-84                                   411,36     9.2  227,75     5.1
                                             3               7
85-89                                   412,76    17.8  232,72    10.0
                                             4               6
90-94                                   312,50    31.5  185,51    18.7
                                             9               6
95+                                     166,28    52.5  110,59    34.9
                                             7               5
======================================================================
Total                                   1,906,     5.7  1,099,     3.3
                                           822             069
----------------------------------------------------------------------
Source:  Our integration of prevalence rates from 18 studies in the
literature and the U.S.  Bureau of the Census population estimates in
Statistical Abstract of the United States:  1996 (Washington, D.C.: 
1996). 

To see how this overall prevalence estimate compares with the
projections that would be derived from individual studies, we
calculated comparable estimates when it was possible to do so.  Most
of the people with AD--58 percent, or 1.1 million--fall between the
ages of 75 and 89.  Within this age group, it was possible to project
from 9 of the 15 studies dealing with any AD and provide age- and
gender-specific estimates for the population based on each study. 
Two of the studies yield estimates below 1 million.  Six fall between
1.1 million and 1.4 million; one (the East Boston study) provides the
basis for an estimate of 3.2 million.  (For further detail, see app. 
VIII.)

These numbers correspond to overall percentages for Americans 65
years of age or older of 5.7 and 3.3 with any AD and with moderate or
severe AD, respectively.  When adjusted for the cases of mixed
dementia and missed cases not included in some of the studies, these
percentages become 6.3 and 4.1, respectively.  As noted, individual
studies of the AD prevalence rate have produced varied estimates of
the percentage of elderly with any AD, ranging from less than 2
percent to 12 percent. 


--------------------
\4 U.S.  Bureau of the Census, Statistical Abstract of the United
States:  1996, 116th ed.  (Washington, D.C.:  1996). 


   PROJECTIONS FOR NUMBERS OF
   PEOPLE WITH AD
------------------------------------------------------------ Letter :4

We also developed projections into the next century of the number of
AD cases and the number of AD cases requiring assistance.  We derived
these results from the prevalence rates, which we used in conjunction
with age- and gender-specific population projections from the Bureau
of the Census.  These projections, which take the aging of the U.S. 
population into account, are presented in 5-year intervals until
2015.\5 As shown in table 2, based on the Bureau's middle series of
population projections, the numbers of cases of AD are expected to
increase approximately 12 percent every 5 years.\6 When adjusted to
include all mixed cases and missed cases, the numbers in this table
increase by 10 percent and 24 percent for any AD and moderate or
severe AD, respectively.  These adjustments yield, for example, in
2015, 3.2 million cases for any AD and 2.1 million cases for moderate
or severe AD. 



                                Table 2
                
                   Projected Estimates of Any AD and
                 Moderate or Severe AD for Americans 65
                    Years of Age or Older, 1995-2015

                                                         Moderate or
                                            Any AD        severe AD
                                        --------------  --------------
                                                     %               %
                                                change          change
Year                                    Number      \a  Number      \a
--------------------------------------  ------  ------  ------  ------
1995                                    1,906,      \b  1,099,      \b
                                           822             069
2000                                    2,141,     +12  1,233,     +12
                                           772             932
2005                                    2,370,     +24  1,365,     +24
                                           615             085
2010                                    2,605,     +37  1,500,     +37
                                           231             727
2015                                    2,872,     +51  1,656,     +51
                                           420             046
----------------------------------------------------------------------
\a All percentage changes are relative to the baseline number for
1995. 

\b Zero by definition. 

Source:  The figures for 1995 are estimates based on the integration
of the literature (taken from table 1).  The figures for the other
years are projections based on the estimates and the Bureau of the
Census middle series of population projections (P-25, No.  1130). 

Because prevalence is partially determined by the length of time
people with AD survive, improvements in AD care will tend to increase
future prevalence, just as any general improvements in human
longevity will.\7 Thus, significant unanticipated improvements in the
longevity of either the elderly in general or people with AD in
particular may lead to even greater numbers of people with AD in the
future. 


--------------------
\5 See, for example, D.  A.  Evans and others, "The Impact of
Alzheimer's Disease in the United States Population," in The Oldest
Old, edited by R.  M.  Suzman, D.  P.  Willis, and K.  G.  Manton
(New York:  Oxford University Press, 1992), pp.  283-99. 

\6 This series is defined by the assumptions of (1) a continuation of
present fertility trends, (2) a gradual increase in life expectancy
to 82.0 years in 2050, and (3) a constant net immigration of 820,000
persons per year. 

\7 A.  B.  Graves and W.  A.  Kukull, "The Epidemiology of Dementia,"
in Handbook of Dementing Illnesses, edited by J.  C.  Morris (New
York:  Dekker, 1994), pp.  23-69. 


   ESTIMATED PREVALENCE RATES
------------------------------------------------------------ Letter :5

We developed age- and gender-specific prevalence rates based on our
meta-analysis, stopping at the age of 95, when the data become sparse
for both any AD and moderate or severe AD (see tables 3 and 4).\8 Our
first step in generating these estimates was to take severity of
disease into account by excluding any data presented in the 18
published studies that included persons with questionable AD.  Even
with this exclusion, the estimates of AD prevalence for a given
combination of age and gender that we obtained from the 18 studies
varied greatly.  (See app.  II for a list of the studies.) For
example, the estimated prevalence for men in the age interval from 85
to 95 ranged from 12 percent to 54 percent (see app.  IV).  Before
attempting to integrate these data, we took severity of disease
further into account by dealing separately with the three
studies--numbers 1, 5, and 18 (see table IV.1)--that excluded mild
cases. 



                          Table 3
          
           AD Prevalence Rates for Men and Women
              Ages 65-95, All Severity Levels

                     Men                    Women
            ----------------------  ----------------------
                               95-                     95-
                           percent                 percent
                        confidence              confidence
Age             Rate\a  interval\b        Rate  interval\b
----------  ----------  ----------  ----------  ----------
65                0.6%    0.6, 0.7        0.8%    0.7, 0.9
70                 1.3    1.2, 1.5         1.7    1.5, 1.9
75                 2.7    2.5, 3.0         3.5    3.2, 3.8
80                 5.6    5.2, 6.0         7.1    6.7, 7.5
85                11.1  10.3, 11.9        13.8  13.2, 14.5
90                20.8  19.2, 22.4        25.2  24.0, 26.5
95                35.6  32.9, 38.3        41.5  39.3, 43.8
----------------------------------------------------------
\a The rates were estimated by logistic regression model. 

\b The 95-percent confidence interval is a pair of values between
which the true rate is likely to fall 95 percent of the time. 



                          Table 4
          
           AD Prevalence Rates for Men and Women
            Ages 65-95, Moderate or Severe Cases

                     Men                    Women
            ----------------------  ----------------------
                               95-                     95-
                           percent                 percent
                        confidence              confidence
Age             Rate\a  interval\b      Rate\a  interval\b
----------  ----------  ----------  ----------  ----------
65                0.3%    0.2, 0.4        0.6%    0.4, 0.8
70                 0.6    0.4, 0.8         1.1    0.9, 1.5
75                 1.1    0.8, 1.5         2.3    1.8, 2.8
80                 2.3    1.7, 3.0         4.4    3.8, 5.3
85                 4.4    3.3, 5.9         8.6   7.2, 10.2
90                 8.5   6.3, 11.5        15.8  12.8, 19.5
95                15.8  11.2, 21.8        27.4  21.4, 34.5
----------------------------------------------------------
\a The rates were estimated by logistic regression model. 

\b The 95-percent confidence interval is a pair of values between
which the true rate is likely to fall 95 percent of the time. 

In our first integration of these data, the results of our
meta-analysis show prevalence rates for any AD in the 15 studies that
do not exclude mild cases.  (See table 3.) These results tend to
differ only slightly from estimates presented in previous articles
reviewing AD prevalence.  These rates are each increased by 10
percent when all mixed cases and missed cases are included. 

These results demonstrate that the AD prevalence rate increases
sharply with age, doubling about every 5 years at least until about
the age of 85, as expected from previous reports of this
relationship.  In addition, the rate is greater for women than for
men. 

In our second integration of these data, we included only studies
that provided number of cases with moderate or severe AD.  These
studies either excluded mild cases (the three studies mentioned
earlier) or enabled us to exclude mild cases by presenting the data
for these cases separately (numbers 6 and 13).  When only cases with
moderate or severe AD are counted, the rates are lower.  But the
increase in prevalence with age and the higher rates for women are
observed at the moderate and severe levels too.  These rates are each
increased by 24 percent when all cases of mixed dementia and missed
cases are included. 

Two of the studies allow for the counting of only the cases with
severe AD.  When this is done, the resulting rates are still lower,
as logic would dictate, but the margins of error are so large that we
do not present these results for severe AD. 


--------------------
\8 These estimates were needed to generate the estimated numbers and
projections discussed in the two previous sections. 


   NIA'S ONGOING STUDIES
------------------------------------------------------------ Letter :6

NIA is currently supporting a number of studies of AD prevalence in
the United States.  Many of these studies include minority groups in
addition to whites.  For example, one such study looks at prevalence
among African-Americans, Hispanics, and whites in a neighborhood of
New York City.  The results of these studies are expected to be
published within the next several years.  When they are, our
knowledge about the extent of AD in the United States will be
enhanced.  Not only will our ability to estimate AD prevalence for
the whole country improve but so will our ability to make such
estimates for specific racial and ethnic populations.  Projections
for future numbers with AD will then be able to take into account the
changing demographics of the country. 

The implications of these findings lie in the specific results and
projections presented.  The number of people with AD is at least 1.9
million now and can, with relatively conservative assumptions about
population growth, be expected to grow to at least 2.9 million by
2015.  Depending on severity, these cases will need some kind of
long-term care.  Such care will also be required by people with other
disabling diseases, both dementias and nondementias.  However, the
kinds of care needed by people with AD and other dementias differ,
for both patients and their caregivers, from the kinds needed by the
disabled without any dementia. 


   AGENCY COMMENTS AND OUR
   EVALUATION
------------------------------------------------------------ Letter :7

Noting that the results of our meta-analysis of AD prevalence (about
2 million) are lower than those NIA uses (about 4 million), NIA found
three methodological limitations in our study that it believes call
into question its validity.  First, NIA noted that only 3 of the 18
studies are of U.S.  populations and questions whether our combining
of U.S.  and non-U.S.  populations is warranted, given that the U.S. 
data tend to yield higher prevalence rates than do the non-U.S. 
data. 

Second, NIA was concerned about variation in how some of the studies
we reviewed applied diagnostic criteria.  The agency was concerned
that most of the questionable cases in the studies reviewed, which we
did not include in our estimates, were actually mild cases of
dementia.  Further, NIA believes that many of the studies we relied
on had insufficiently sensitive initial screens that led to their
missing many mild cases of dementia.  NIA was also concerned that in
many of the studies reviewed, only persons with pure AD were coded as
cases of AD but not those persons with a mixed dementia, including AD
as a component.  Third, NIA commented that the meta-analytic method
we used cannot compensate for the large differences in rates observed
across studies. 

Although we have made some adjustments to account for NIA's
criticisms, we believe our methodology remains useful for estimating
AD prevalence.  First, with regard to the use of non-U.S.  studies,
we note that when severity is taken into account, the results of U.S. 
and non-U.S.  studies are comparable when one of the U.S.  studies is
excluded.  This study, the East Boston study, yields prevalence
estimates that are far higher than any of the other studies,
suggesting a disparity in methodology rather than in population
characteristics. 

As for NIA's criticisms related to diagnostic criteria, we recognize
the utility of including cases of mixed dementia and of adjusting for
insensitive screens; we have, therefore, included in this report
estimates to reflect these adjustments.  We disagree, however, with
the idea that questionable cases should be included.  Although such
people may become demented, they do not at the time of the prevalence
survey satisfy the accepted diagnostic criteria for dementia or AD. 

Finally, we disagree with NIA's conclusion that meta-analysis is an
inappropriate method because of the heterogeneity of the prevalence
rates in the studies we reviewed.  With severity accounted for, 17 of
the 18 studies we reviewed reported relatively homogeneous rates,
with one outlying study. 

We also received a letter from the Alzheimer's Association expressing
similar concerns about our methodology.  The Alzheimer's Association
is especially concerned about how forthcoming data from studies
currently underway may change the picture of AD prevalence we
present.  Again, we acknowledge that NIA is supporting new and
hopefully better studies of the extent of AD in the U.S.  and that
these should improve our understanding of how the disease is
distributed in all the major subpopulations. 

The full text of NIA's comments, along with our response, is included
in appendix IX. 


---------------------------------------------------------- Letter :7.1

We will send copies of this report to the directors of the National
Institutes of Health, the National Institute on Aging, and the
Administration on Aging.  In addition, we will make copies available
upon request to others who are interested. 

If you or your staff have any questions about this report, please
call me at (202) 512-7119 or Donald M.  Keller, Evaluator-in-Charge,
at (202) 512-2932.  GAO staff acknowledgments are listed in appendix
X. 

Sincerely yours,

Bernice Steinhardt
Director of Health Services
 Quality and Public Health


18 STUDIES WE REVIEWED
=========================================================== Appendix I

The following 18 studies are the AD prevalence studies we reviewed
for this report; 3 other studies are supplementary sources, providing
data and other information about one or more of the studies reviewed,
as indicated.  The studies reviewed are numbered for reference in
table IV.1. 


   SOURCES FOR STUDIES REVIEWED
--------------------------------------------------------- Appendix I:1

1.  Bachman, D.L., and others.  "Prevalence of Dementia and Probable
Senile Dementia of the Alzheimer Type in the Framingham Study."
Neurology, Vol.  42 (1992), pp.  115-19. 

2.  Brayne, C., and P.  Calloway.  "An Epidemiological Study of
Dementia in a Rural Population of Elderly Women." British Journal of
Psychiatry, Vol.  155 (1989), pp.  214-19. 

3.  Canadian Study of Health and Aging Working Group.  "Canadian
Study of Health and Aging:  Study Methods and Prevalence of
Dementia." Canadian Medical Association Journal, Vol.  150 (1994),
pp.  899-913. 

4.  Coria, F., and others.  "Prevalence of Age-Associated Memory
Impairment and Dementia in a Rural Community." Journal of Neurology,
Neurosurgery, and Psychiatry, Vol.  56 (1993), pp.  973-76. 

5.  Corso, E.A., and others.  "Prevalence of Moderate and Severe
Alzheimer Dementia and Multi-Infarct Dementia in the Population of
Southeastern Sicily." Italian Journal of Neurological Sciences, Vol. 
13 (1992), pp.  215-19. 

6.  D'Alessandro, R., and others.  "Dementia in Subjects Over 65
Years of Age in the Republic of San Marino." British Journal of
Psychiatry, Vol.  153 (1988), pp.  182-86. 

7.  Evans, D.A., and others.  "Prevalence of Alzheimer's Disease in a
Community Population of Older Persons:  Higher Than Previously
Reported." Journal of the American Medical Association, Vol.  262
(1989), pp.  2551-56. 

8.  Fratiglioni, L., and others.  "Prevalence of Alzheimer's Disease
and Other Dementias in an Elderly Urban Population:  Relationship
with Age, Sex, and Education." Neurology, Vol.  41 (1991), pp. 
1886-92. 

9.  Lobo, A., and others.  "The Epidemiological Study of Dementia in
Zaragoza, Spain." In Psychiatry:  A World Perspective.  Proceedings
of the VIII World Congress of Psychiatry, edited by C.N.  Stefaniss,
C.R.  Soldators, and A.D.  Rabavilas.  Amsterdam:  Elsevier, 1990,
pp.  133-37. 

10.  Manubens, J.M., and others.  "Prevalence of Alzheimer's Disease
and Other Dementing Disorders in Pamplona, Spain." Neuroepidemiology,
Vol.  14 (1995), pp.  155-64. 

11.  O'Connor, D.W., and others.  "The Prevalence of Dementia as
Measured by the Cambridge Mental Disorders of the Elderly
Examination." Acta Psychiatrica Scandinavica, Vol.  79 (1989), pp. 
190-98. 

12.  Ott, A., and others.  "Prevalence of Alzheimer's Disease and
Vascular Dementia:  Association with Education.  The Rotterdam
Study." British Medical Journal, Vol.  310 (1995), pp.  970-73. 

13.  Pfeffer, R.I., A.A.  Afifi, and J.M.  Chance.  "Prevalence of
Alzheimer's Disease in a Retirement Community." American Journal of
Epidemiology, Vol.  125 (1987), pp.  420-36. 

14.  Rocca, W.A., and others.  "Prevalence of Clinically Diagnosed
Alzheimer's Disease and Other Dementing Disorders:  A Door-to-Door
Survey in Appignano, Macerata Province, Italy." Neurology, Vol.  40
(1990), pp.  626-31. 

15.  Roelands, M., and others.  "The Prevalence of Dementia in
Belgium:  A Population-Based Door-to-Door Survey in a Rural
Community." Neuroepidemiology, Vol.  13 (1994), pp.  155-61. 

16.  Rorsman, B., O.  Hagnell, and J.  Lanke.  "Prevalence and
Incidence of Senile and Multi-Infarct Dementia in the Lundby Study: 
A Comparison Between the Time Periods 1947-1957 and 1957-1972."
Neuropsychobiology, Vol.  15 (1986), pp.  122-29. 

17.  Skoog, I., and others.  "A Population-Based Study of Dementia in
85-Year-Olds." New England Journal of Medicine, Vol.  328 (1993), pp. 
153-58. 

18.  Sulkava, R., and others.  "Prevalence of Severe Dementia in
Finland." Neurology, Vol.  35 (1985), pp.  1025-29. 


   SUPPLEMENTARY SOURCES
--------------------------------------------------------- Appendix I:2

Beckett, L.A., P.A.  Scherr, and D.A.  Evans.  "Population Prevalence
Estimates From Complex Samples." Journal of Clinical Epidemiology,
Vol.  45 (1992), pp.  393-402.  (Relevant to study 7.)

Ebly, E.M., and others.  "Prevalence and Types of Dementia in the
Very Old:  Results from the Canadian Study of Health and Aging."
Neurology, Vol.  44 (1994), pp.  1593-1600.  (Relevant to study 3.)

Rocca, W.A., and others.  "Frequency and Distribution of Alzheimer's
Disease in Europe:  A Collaborative Study of 1980-1990 Prevalence
Findings." Annals of Neurology, Vol.  30 (1991), pp.  381-90. 
(Relevant to studies 2, 9, 11, 14, 16, and 18.)


DIAGNOSTIC CRITERIA FOR PROBABLE
AND POSSIBLE ALZHEIMER'S DISEASE
========================================================== Appendix II

The source of the diagnostic criteria is G.  McKhann and others,
"Clinical Diagnosis of Alzheimer's Disease:  Report of the NINCDS
[National Institute of Neurological and Communicative Disorders and
Stroke]--ADRDA [Alzheimer's Disease and Related Disorders
Association] Work Group Under the Auspices of the Department of
Health and Human Services Task Force on Alzheimer's Disease."\9


--------------------
\9 See Neurology, Vol.  34 (1984), pp.  939-44. 


   CRITERIA FOR DIAGNOSIS OF
   PROBABLE AD
-------------------------------------------------------- Appendix II:1

The criteria for the clinical diagnosis of probable Alzheimer's
disease (AD) include

  -- dementia established by clinical examination and documented by
     the Mini-Mental Test, Blessed Dementia Scale, or some similar
     examination and confirmed by neuropsychological tests;

  -- deficits in two or more areas of cognition;

  -- progressive worsening of memory and other cognitive functions;

  -- no disturbance of consciousness;

  -- onset between the ages of 40 and 90, most often after the age of
     65; and

  -- absence of systemic disorders or other brain diseases that in
     and of themselves could account for the progressive deficits in
     memory and cognition. 

The diagnosis of probable AD is supported by

  -- progressive deterioration of specific cognitive functions such
     as language (aphasia), motor skills (apraxia), and perception
     (agnosia);

  -- impaired activities of daily living and altered patterns of
     behavior;

  -- family history of similar disorders, particularly if confirmed
     neuropathologically; and

  -- laboratory results of normal lumbar puncture as evaluated by
     standard techniques; normal pattern of nonspecific changes in
     the electroencephalogram, such as increased slow-wave activity;
     and evidence of cerebral atrophy on computerized tomography
     (CT), with progression documented by serial observation. 

Other clinical features consistent with the diagnosis of probable
Alzheimer's disease, after exclusion of causes of dementia other than
Alzheimer's disease, include

  -- plateaus in the course of progression of the illness;

  -- associated symptoms of depression; insomnia; incontinence;
     delusion; illusions; hallucinations; catastrophic verbal,
     emotional, or physical outbursts; sexual disorders; and weight
     loss;

  -- other neurological abnormalities in some patients, especially
     with more advanced disease, and including motor signs such as
     increased muscle tone, myoclonus, or gait disorder;

  -- seizures in advanced disease; and

  -- CT normal for age. 

Criteria that make the diagnosis of probable Alzheimer's disease
uncertain or unlikely include

  -- sudden, apoplectic onset;

  -- focal neurological findings such as hemiparesis, sensory loss,
     visual field deficits, and lack of coordination early in the
     course of the illness; and

  -- seizures or walking disturbances at the onset or early in the
     course of the illness. 


   CRITERIA FOR DIAGNOSIS OF
   POSSIBLE AD
-------------------------------------------------------- Appendix II:2

Clinical diagnosis of possible Alzheimer's disease

  -- may be made on the basis of the dementia syndrome, in the
     absence of other neurologic, psychiatric, or systemic disorders
     sufficient to cause dementia, and in the presence of variations
     in the onset, in the presentation, or in the clinical course;

  -- may be made in the presence of a second systemic or brain
     disorder sufficient to produce dementia, which is not considered
     to be the cause of the dementia; and

  -- should be used in research studies when a single, gradually
     progressive severe cognitive deficit is identified in the
     absence of other identifiable cause. 


DEFINING AND LOCATING THE MOST
RELEVANT STUDIES
========================================================= Appendix III


   DEFINING RELEVANT STUDIES
------------------------------------------------------- Appendix III:1

We defined relevant studies as published studies of original research
satisfying each of three inclusion criteria.  The studies had to (1)
include age- and gender-specific prevalence rates of AD (2) diagnosed
by NINCDS-ADRDA (or equivalent) criteria (see app.  II), along with
the corresponding sample sizes, (3) from white (that is,
European-American or European) populations. 

Because the AD prevalence rate is known to vary by age and may vary
by gender, overall rates for elderly people are likely to be
sensitive to differences among populations in age and gender.\10 One
way in which the AD prevalence rates from different populations can
be validly compared is if the rates are specific to a particular
combination of age and gender (for example, the rate for women
between the ages of 70 and 74).  Thus, we include only studies that
present age- and gender-specific AD prevalence rates, along with the
sample sizes needed to weight them in a quantitative integration. 

The published studies presenting these rates include populations from
North America, Europe, and Asia.  These populations are typically
small, often a neighborhood within a city or a small town, and none
of them individually or in any combination can be assumed to be
representative of the U.S.  population.  The white (that is,
European-American and European) populations studied contain few
participants not of European background.  The best that can be done
until a sufficiently large population representative of the United
States is studied is to integrate the results from available studies,
excluding those with AD prevalence rates that are likely to differ
systematically from those of the majority white population of the
United States. 

Prevalence rates for AD from Asian countries tend to be lower than
those observed in Europe and North America, although Asian-American
rates are closer to those of the white population.\11 The reason for
this difference is not known, but we decided that to be cautious in
extracting prevalence rates, we would exclude the numerous studies of
Asian and Asian-American populations.  This leaves us with only
studies of populations not known to differ systematically from
European-Americans with respect to AD prevalence. 

If different diagnostic criteria are used to ascertain cases in
various studies, then observed differences in AD prevalence may
reflect the different criteria rather than true population
differences in AD prevalence.  Integrating only prevalence estimates
with the same diagnostic criteria can reduce the effects of criteria
as a source of differences among estimates.  In order to minimize the
possible role of differences in diagnostic criteria, we include only
studies using the NINCDS-ADRDA criteria for probable AD (or for
probable and possible AD--see app.  II) or equivalent diagnostic
criteria.\12


--------------------
\10 A.  F.  Jorm, The Epidemiology of Alzheimer's Disease and Related
Disorders (London:  Chapman and Hall, 1990), pp.  69-72. 

\11 A.  B.  Graves and others, "Prevalence of Dementia and Its
Subtypes in the Japanese American Population of King County,
Washington State:  The Kame Project," American Journal of
Epidemiology, Vol.  144 (1996), pp.  760-71.  Also see L.  White and
others, "Prevalence of Dementia in Older Japanese-American Men in
Hawaii," Journal of the American Medical Association, Vol.  276
(1996), pp.  955-60. 

\12 Given the difficulty of obtaining direct pathological evidence of
AD in living patients, a clinical diagnosis can be made only when
other potential causes of dementia have been ruled out.  Even then it
is called "probable"--not "definite"--AD.  (A diagnosis of definite
AD requires both evidence of AD pathology and satisfaction of the
clinical criteria for probable AD.) When other potential causes have
not been ruled out but appear unlikely to be the main cause of the
dementia, it is called "possible" AD. 


   LOCATING STUDIES
------------------------------------------------------- Appendix III:2

We used a systematic computer-assisted search of the medical and
social science literature, supplemented by expert advice and
references found in the literature, in order to locate published
studies on AD prevalence that meet the inclusion criteria listed
above.  We found 18 studies meeting these criteria (see app.  I). 


EXTRACTING PREVALENCE RATES FROM
REVIEWED STUDIES
========================================================== Appendix IV

Using published results (in tables or graphs) from each of the
studies, we recorded the age- and gender-specific AD prevalence rates
for all reported age intervals with lower limits of 60 years or
older.  In most of the studies, the AD rates excluded all other kinds
of dementia, but in four of them a number of cases of mixed dementia
(cases diagnosed with AD and another dementia) were included.  For
each age interval reported, we recorded the midpoint.  If the
open-ended age interval "85 and older" was used, we considered it as
extending to 95 and recorded the midpoint (90.5).  We considered "90
and older" and "95 and older" as extending to 99.  When prevalence
rates were not given explicitly, we computed them from available data
or read them from graphs.  When differing estimates of the same rate
were presented in different articles about the same study, we
consulted an expert to determine the correct values. 

The rates for analysis are listed in table IV.1, with each of the 18
studies numbered, as identified in appendix I.  We refer to cases of
mild, moderate, or severe AD as cases of "any AD." Some rating
systems include other categories of severity.  For example,
"questionable dementia"--a category intermediate between "normal" and
mild dementia--is used in the Clinical Dementia Rating (CDR) for
people who are only slightly impaired and do not satisfy the
NINCDS-ADRDA criteria for dementia.  People in this intermediate
category may or may not be counted as cases of dementia in different
studies.  We do not consider them to be cases of dementia, however. 



                                        Table IV.1
                         
                         AD Prevalence Rates According to Age and
                             Gender in 18 Studies We Reviewed

                                                    Men                    Women
                                           ----------------------  ----------------------
Study number,
place, ethnicity
of participants      Severity         Age      Sample        Rate      Sample        Rate
-----------------  ----------  ----------  ----------  ----------  ----------  ----------
1. Framingham       Moderate+       61-64         129       0.000         156       0.000
 (U.S.A.),
 Italian-
 American
                                    65-69         284      0.352%         384      0.781%
                                    70-74         190       1.053         314       0.000
                                    75-79         128       0.781         207       2.899
                                    80-84          81       3.704         147       8.844
                                    85-93          41       7.317         119      15.126
2. East                 Mild+       70-74           0          \a         185       1.622
 Cambridgeshire
 (U.K.), English
                                    75-79           0          \a         180       6.667
3. Canada,              Mild+       65-74       3,800       0.500       2,857       1.400
 Canadian
                                    75-84       1,691       5.500       2,654       7.800
                                    85-89         387      11.886         941      18.810
                                    90-94          91      27.473         280      33.929
                                    95-99          16      56.250          88      39.773
4. Turegano             Mild+       55-64          72       0.000          69       1.449
 (Spain), Spanish
                                    65-74          51       0.000          55       3.636
                                    75-84          32       0.000          46       8.696
                                    85-94          10      10.000          14       7.143
                                    95-99           1       0.000           1       0.000
5. Ragusa           Moderate+       60-64         197       0.000         241       0.415
 (Italy), Italian
                                    65-69         219       0.457         259       1.158
                                    70-74         181       1.105         209       2.392
                                    75-95         225       2.222         269       6.320
6. San Marino,          Mild,          67          82       2.439          81       0.000
 Italian            moderate,
                   and severe
                                       72          72       0.000          64       3.125
                                       77          48       4.167          63       6.349
                                       82          21       0.000          29      10.345
                                       87          14       7.143          14      35.714
7. East Boston          Mild+       65-69         506       1.433         747       3.232
 (U.S.A.),
 Italian-
 American
                                    70-74         399       1.146         653       3.938
                                    75-79         243      22.210         417       7.533
                                    80-84         130      32.839         246      27.448
                                    85-95         104      40.502         178      46.545
8. Stockholm            Mild+       75-79         193       2.073         522       3.831
 (Sweden),
 Swedish
                                    80-84         153       4.575         463       4.536
                                    85-89          59      10.169         264      12.121
                                    90-99          27       3.704         129      19.380
9. Zaragoza             Mild+       65-69          36       0.000          49       0.000
 (Spain), Spanish
                                    70-79          79       2.532          99       3.030
                                    80-89          29      17.241          37       8.108
                                    90-99           1       0.000           3      33.333
10. Pamplona            Mild+       72-74          71       1.408          75       0.000
 (Spain), Spanish
                                    75-79         152       3.289         159      11.321
                                    80-84         152       5.921         150      13.333
                                    85-89         142      11.972         137      20.438
                                    90-91          45      20.000          44      27.273
11. Cambridge           Mild+       75-79         382       1.309         655       2.901
 (U.K.), English
                                    80-84         290       8.276         496       8.065
                                    85-89         115      15.652         241      15.768
                                    90-94          22      40.909          85      24.706
12. Rotterdam           Mild+       55-64       1,118       0.179       1,495       0.134
 (Holland), Dutch
                                    65-74       1,116       0.806       1,447       1.037
                                    75-84         569       5.448       1,074       8.380
                                    85-95         136      25.000         573      27.225
13. Southern       Questionab       65-69          64       3.125          58       0.000
 California         le, mild,
 (U.S.A.),          moderate,
 European-         and severe
 American
                                    70-74          83       1.205          91       4.396
                                    75-79         102      15.686          88       9.091
                                    80-84          93      35.484          90      22.222
                                    85-95          80      53.750          68      51.471
14. Appignano           Mild+       60-69         173       0.578         186       0.538
 (Italy), Italian
                                    70-79         126       0.794         178       2.809
                                    80-89          43       6.977          65      12.308
                                    90-94           1       0.000           6      16.667
15. Heist-op-           Mild+       65-69         166       0.000         170       0.000
 den-Berg,
 (Belgium),
 Belgian
                                    70-74         192       1.563         164       2.439
                                    75-79         154       2.597         148       6.081
                                    80-84         118       9.322         123       9.756
                                    85-95          76      11.842          76      19.737
16. Lund                Mild+       60-69         191       0.000         177       0.565
 (Sweden),
 Swedish
                                    70-79          87       3.448         115       1.739
                                    80-89          21       9.524          43      11.628
17. Gothenburg          Mild+       85-85         143      11.888         351      13.390
 (Sweden),
 Swedish
18. Finland,        Moderate+       60-69         583       0.343         764       0.262
 Finnish
                                    70-74         208       2.404         346       2.601
                                    75-79         108       3.704         246       5.691
                                    80-89          83       7.229          \b          \b
                                    80-84          \b          \b         136      11.765
                                    85-89          \b          \b          41      21.951
-----------------------------------------------------------------------------------------
Note:  The study numbers are keyed to the list in appendix I. 

\a Rate undefined. 

\b In the presentation of these data, different age categories were
used for men and women. 

By our definition of AD, one of the 18 sets of prevalence rates--the
set from the Southern California study (study 13)--does not qualify
because it includes cases of questionable dementia.  Therefore, we
omitted these rates from further analysis.  However, using the
published reports of this study, it is possible to isolate the cases
of AD at each of the severity levels--mild, moderate, or severe--for
further analysis.  We did this, and the resulting data are included
in the analyses that follow. 

We extracted not just the overall AD prevalence rates as indicated
above but also, wherever possible, the rates according to each of
three cumulative severity levels:  (1) cases of mild or greater
severity (any AD), (2) cases of moderate or severe AD, and (3) cases
of severe AD.  The mild or greater severity level would include all
cases.  The moderate or severe level would include only cases of
moderate or severe dementia, and the severe level would include only
the cases of severe dementia.  For many of the studies, the data were
presented in such a way that a breakdown of this kind was not
possible. 

For 13 of the 18 studies (2-4, 7-12, and 14-17), the only information
available about severity for the age- and gender-specific AD
prevalence rates is that all cases of mild or greater severity are
included.  Their prevalence rates correspond exactly to their overall
rates, as listed in table IV.1.  For three more of the studies (1, 5,
and 18), the cases include only moderate or severe AD.  Their
prevalence rates also correspond exactly to their overall rates in
table IV.1, but the prevalence rates represent the level of moderate
or severe AD. 

In the remaining two studies (6 and 13), age- and gender-specific AD
prevalence rates are presented for different levels of severity.  It
is therefore possible to compute from these two studies, by the
process of summation, the rates of each cumulative severity level: 
any AD, moderate or severe AD, and severe AD. 


INTEGRATING PREVALENCE RATES
QUANTITATIVELY
=========================================================== Appendix V

To obtain relatively precise estimates, based on all the data for
each level of severity, we quantitatively integrated the estimates. 
To integrate the data for a given level to arrive at estimates of
age- and gender-specific prevalence, we used a method previously
employed by Maria Corrada and her associates at Johns Hopkins
University.\13 This method, which pools the data of individuals from
each of the studies, involves fitting a logistic regression model to
the data--age interval midpoints, gender, numbers of participants,
numbers of cases, and levels of severity--from a series of relevant
prevalence studies so as to estimate the age- and gender-specific
prevalence rates for each level of severity.\14 Such a model implies
that (1) AD prevalence at a given level is determined by age and
gender and (2) the quantitative nature of the relationship is of the
kind known to statisticians as logistic. 

Logistic regression models are similar to the more commonly
encountered linear regression models, but they are especially
designed to analyze variables that take on only two values (variables
that may be called binary or dichotomous).  An example of such a
variable is the presence or absence of disease in a person.  The
status of all people in a population with respect to this binary
variable determines the prevalence rate for that population.  Thus,
logistic regression is an appropriate method for analyzing data for
prevalence rate. 

We applied the approach of Corrada and her colleagues.  Our work
differs from theirs both in that we were able to include some more
recent studies than they were and in the way in which we took
severity into account.  The approach was applied to three sets of
data.  One set was composed of the age- and gender-specific AD
prevalence rates from the 15 studies that include such rates for
cases with any AD.  The second set was composed of the rates from the
five studies that include these rates for cases with moderate or
severe AD.  The third set was composed of the rates from the two
studies that include these rates for cases with severe AD.  The
results of this application are presented in tables 3 and 4, which
correspond to the first two sets of data.  The results of the
application to the third set are not presented because these results
included relatively imprecise prevalence estimates.  We did not
extend our estimates beyond the age of 95 because of the few people
in the study older than 95. 

These results can be compared with other reviews of the AD
literature.  Most published reviews of the literature on AD
prevalence rates are qualitative.  These reviews have not been
designed so as to obtain prevalence estimates through a systematic
quantitative integration of the study data, such as that provided by
meta-analysis.  One representative qualitative review notes that
rates are typically estimated at about 0.5 percent, 3 percent, and 10
percent for the ages of 65, 75, and 85, respectively, both genders
combined.\15 The percentages from one of the few quantitative
reviews--based on combining data from individual studies--were
similar to those from the qualitative reviews.\16


--------------------
\13 M.  Corrada, R.  Brookmeyer, and C.  Kawas, "Sources of
Variability in Prevalence Rates of Alzheimer's Disease,"
International Journal of Epidemiology, Vol.  24 (1995), pp.  1000-5. 

\14 For four of the studies, the number of cases had to be computed
from the prevalence rates and the number of people in the sample, and
for one study the number of people in the study had to be computed
from the prevalence rates and the number of people with AD. 

\15 M.  M.  B.  Breteler and others, "Epidemiology of Alzheimer's
Disease," Epidemiologic Reviews, Vol.  14 (1992), p.  76. 

\16 K.  Ritchie and others, "The Relationship Between Age and the
Prevalence of Senile Dementia:  A Meta-Analysis of Recent Data,"
International Journal of Epidemiology, Vol.  21 (1992), pp.  763-69. 


SOME STRENGTHS AND LIMITATIONS OF
THE STUDIES WE REVIEWED
========================================================== Appendix VI

The strengths of the studies reviewed are that they include
representative samples of well-defined populations we believed to be
similar to the population of the United States with respect to AD
prevalence and diagnosed by accepted diagnostic criteria for AD. 

There are some limitations concerning how well the study populations
can represent the population of interest, the residents of the United
States.  Although each sample represents a well-defined population
and each population provides estimates of AD, the samples, taken
individually or combined, are not representative of the U.S. 
population with respect to all likely determinants of the AD
prevalence rate.  Two of the possibly significant ways the study
populations differ from the U.S.  population are discussed here.  In
addition, the geographic difference must be acknowledged:  most of
the studies included in our analysis are based on European
populations.  We know of no argument, however, that the prevalence of
AD for white Americans differs from that of Europeans. 

None of the studies include significant amounts of data from major
U.S.  subpopulations, such as blacks (that is, African-Americans). 
It is not known whether blacks or other minorities (for example,
Native Americans) have different prevalence rates than do whites and
Europeans.\17 As indicated in appendix III, there is some evidence
that Asian-American rates differ little from the rates of white
populations, in spite of the racial similarity between
Asian-Americans and Asians; Asian rates tend to be systematically
lower than those for whites.  If minorities do have different rates,
it is desirable to know their rates for at least two reasons:  (1)
these rates affect the overall U.S.  estimates and (2) the kinds of
care these minorities require may differ, for cultural reasons, from
the kinds required by other Americans.  NIH supports research
designed to compare the AD prevalence rates of different racial and
ethnic groups. 

Most of the studies include institutionalized people in the
populations they survey, but two of the three U.S.  studies do not. 
Logically, one might expect that since AD rates for the
institutionalized are most likely higher than those for the
noninstitutionalized (that is, community dwellers), omitting the
institutionalized would lower prevalence estimates.  There is little
evidence from a previous analysis, however, that such omission has
any effect.\18 Further, the two U.S.  studies that omit the
institutionalized present prevalence estimates that are high relative
to those from most of the other studies.  It may be that too small a
proportion of the elderly population is institutionalized for the
assumed higher AD prevalence rate to have mattered much in these
studies.  Nevertheless, prevalence studies of AD should ideally
include all elderly people, whether institutionalized or
noninstitutionalized.  We have no reason to conclude, however, that
variation across studies in the handling of institutional status
compromises the validity of the estimates to any significant extent. 


--------------------
\17 No conclusive information is available pertaining to this issue. 
One study does suggest that blacks have a higher prevalence of severe
AD than whites.  See B.  S.  Schoenberg, D.  W.  Anderson, and A.  F. 
Haerer, "Severe Dementia:  Prevalence and Clinical Features in a
Biracial U.S.  Population," Archives of Neurology, Vol.  42 (1985),
pp.  740-43.  Another, however, indicates that the AD rate for blacks
differs little from typical rates from white populations.  See H.  C. 
Hendrie and others, "Prevalence of Alzheimer's Disease and Dementia
in Two Communities:  Nigerian Africans and African Americans,"
American Journal of Psychiatry, Vol.  152 (1995), pp.  1485-92. 

\18 See Corrada, Brookmeyer, and Kawas, "Sources of Variability in
Prevalence Rates of Alzheimer's Disease," pp.  1000-5. 


ADJUSTMENTS BASED ON ALTERNATIVE
CONVENTIONS
========================================================= Appendix VII

All prevalence studies are based on conventions that may be
questioned.  When a convention is judged to be inappropriate for a
given purpose, it may lead to biased prevalence estimates.  Our use
of quantitative integration to generate estimates and projections is,
in part, an attempt to get around some of the inappropriate
conventions in individual studies by diluting them, if not by
canceling them out. 

Two common conventions, in particular, seem inappropriate for our
purposes, although they were reasonable to those who adopted them in
the 1980s and early 1990s, when most of the reviewed work was done. 
One is that people with mixed dementia (AD and another kind of
dementia) should not be counted as AD cases.  Most of the studies we
reviewed have counted as cases of AD only people with "pure" AD, AD
in the absence of other dementia.  Although this convention was
useful for isolating those with no known cause of dementia--those
with AD only--it is illogical if one wants to know how many people
have AD.  A person with both AD and another dementia is logically a
person with AD.  If we drop the usual convention and instead treat
all cases of AD the same, regardless of other dementias, we can then
infer that the estimates presented are too low.  This is because only
four of the studies include mixed cases in the age- and
gender-specific AD rates; therefore, our estimates, based on an
integration of the published rates, underestimate the true rates of
all AD cases. 

Given the available data, it is not possible to derive, on the basis
of our revised convention, age- and gender-specific estimates of the
true rates of all AD.  We can provide rough overall ones, however. 
Ten of the studies enable us to estimate the overall percentage
increase in the number of AD cases that would be obtained if mixed
cases are added.  These estimates vary, with a median (middlemost)
value of 20 percent.  If none of the studies reviewed included mixed
cases, then it would be reasonable to assume that this 20 percent is
the adjustment factor needed to increase our estimates by taking into
account the mixed cases.  However, the 11 studies of mild or more
severe dementia that do not include mixed cases in their AD counts
also include only 29 percent of the participants in the studies of
any AD; in addition, these 11 studies happen to use, on the average,
relatively small samples.  Thus, the value of 20 percent greater
prevalence with mixed cases can only be applied to this 29 percent of
the participants in the studies of any AD, yielding an overall
percentage bias of 5.8 percent (20 percent times 29 percent). 

The 20-percent adjustment for mixed cases can also be applied to 90
percent of the participants in studies of moderate or severe AD since
this is the percentage of participants in studies of similar
populations that exclude mixed cases from their AD counts.  The
resulting adjustment factor is 18 percent (20 percent times 90
percent).  When this factor is applied to the estimates given above,
the findings for 1995 are 2 million people with AD of any kind and
1.3 million cases of moderate or severe AD, rather than the 1.9
million and 1.1 million, respectively, that were originally reported. 

The other common convention that now seems inappropriate is that
prevalence counts are not corrected for the likely number of people
with AD missed by initial screening tests (and therefore not given
workups for dementia).  In some of the studies we reviewed, we found
a potential problem:  Some AD cases were missed as a result of the
use of insufficiently sensitive screens for which no corrections in
the age- and gender-specific rates were made.  When such a problem is
likely, it is illogical to present prevalence rates without
correcting the prevalence estimates for the expected number of people
with AD missed by the screen.  Because studies that do not avoid the
problem of missed cases generate rates that are too low, our
estimates, to the extent they are based on such studies, are
underestimates of the true rates. 

As with the adjustment for mixed cases, it is not possible to derive
age- and gender-specific rates adjusted for missed cases, but a rough
overall adjustment factor can be derived.  Of the 15 studies we
reviewed that investigated mild dementia, 5 can be identified as
having a known or possible problem with missed cases.\19 Of these
five studies, two present their overall percentage increases as a
result of missed cases (although they do not correct the age- and
gender-specific rates for these); the higher of these two increases
is 7.2 percent.  The correction rate undoubtedly varies with the
specific screen used, but we adopted this 7.2 percent as a
representative value.  If all the studies we reviewed had a possible
problem with missed cases, then it would be reasonable to assume that
this 7.2 percent is the adjustment factor needed to increase our
estimates so as to take into account the expected missed cases.  The
five studies known to have a possible problem with missed cases,
however, use relatively large samples, including 50 percent of all
participants in the studies of any AD.  Thus, the value of 7.2
percent greater prevalence with missed cases can be applied only to
the 50 percent of the participants in the studies of any AD, yielding
an overall percentage bias of 3.6 percent (7.2 percent times 50
percent). 

The 7.2-percent adjustment for missed cases can also be applied to 71
percent of the participants in studies of moderate or severe AD. 
This is because 71 percent is the percentage of the participants in
studies of severe AD without correction to all or part of the data
for missed cases.  The resulting adjustment factor is 5.1 percent
(7.2 percent times 71 percent).  These adjustments for missed cases,
when made in addition to the adjustment for mixed dementia that was
already made, yield an increase of 10 percent in 1995, for an
adjusted estimate of 2.1 million cases for any AD.  The corresponding
adjustment for moderate or severe AD is 24 percent, for a 1995 total
of 1.4 million cases at these levels of severity. 

Appendix VIII NUMBER OF AD CASES FOR PERSONS 75 TO 89 YEARS OLD BASED
ON INDIVIDUAL STUDIES AND ON THE META-ANALYSIS

Table VIII.1 facilitates the comparison of the individual studies
reviewed in the meta-analysis of the prevalence of any AD.  The basis
of comparison is the age- and gender-specific numbers of cases of any
AD projected for the U.S.  population of 1995 aged 75 to 89 years. 
As explained in the report, only nine of the studies we reviewed
provide the appropriate data; the meta-analysis is based on these 9
plus an additional 6, for a total of 15. 



                                       Table VIII.1
                         
                          Millions of U.S. AD Cases as Projected
                          From Individual Studies and From Meta-
                                     Analytic Results

                                      Study number
              -------------------------------------------------------------
                                                                                    Meta-
                                                                                 analytic
Group            3      6     7     8    10      11      12      13      15       results
------------  ----  -----  ----  ----  ----  ------  ------  ------  ------  ------------
Men 75-79     0.15   0.11  0.60  0.06  0.09    0.04    0.15    0.13    0.07          0.11
Men 80-84     0.09   0.00  0.53  0.07  0.09    0.13    0.09    0.14    0.15          0.13
Men 85-89     0.08   0.05  0.29  0.07  0.08    0.11    0.18    0.25    0.08          0.11
Women 75-79   0.31   0.25  0.30  0.15  0.45    0.11    0.33    0.09    0.24          0.20
Women 80-84   0.22   0.29  0.78  0.13  0.38    0.23    0.24    0.22    0.28          0.28
Women 85-89   0.30   0.58  0.75  0.20  0.33    0.26    0.44    0.45    0.32          0.30
=========================================================================================
Total         1.16   1.28  3.25  0.68  1.43    0.88    1.42    1.28    1.14          1.13
-----------------------------------------------------------------------------------------
Note:  The study numbers are keyed to the list in appendix I. 
Summations may not equal totals because of rounding. 



(See figure in printed edition.)Appendix IX COMMENTS FROM THE
NATIONAL INSTITUTE ON AGING



(See figure in printed edition.)

Now on p.  1. 



(See figure in printed edition.)



(See figure in printed edition.)



(See figure in printed edition.)



(See figure in printed edition.)


The following are GAO's comments on NIA's June 4, 1997, letter. 

GAO COMMENTS

1.  We disagree with NIA's comment about the countries in which the
prevalence studies were conducted.  Studies of European and other
countries with predominantly white populations are legitimately used
to arrive at prevalence rates characteristic of the white population
in the United States.  None of the populations studied, including
those from the United States, is representative of the white U.S. 
population with respect to ethnic, linguistic, and socioeconomic
variables, but these have not been shown to determine AD prevalence
rates.  The relatively high rates of the combined U.S.  studies are
driven by the contributions of a single study that focused on the
population of East Boston.  This study, the one that NIA bases its
estimates on, yields rates that are higher than those of all other
studies, including the other U.S.  studies.  The difference between
the U.S.  and non-U.S.  studies reduces to a difference between East
Boston and all other studies. 

2.  The report has been changed to reflect the likely limiting role
of screen cut-off scores. 

3.  We disagree with NIA's point about "questionable" dementia.  In
accordance with the conventions of the field, we defined people with
AD as those rated as having mild or more severe dementia.  While it
is true that a certain proportion of those with questionable dementia
will develop AD, prevalence estimates are traditionally given for
persons who have a disease and do not include those who may get the
disease at a later time. 

4.  The report has been changed to reflect the likely role of
excluding mixed dementia. 

5.  To examine the role of heterogeneity among studies in our
meta-analysis, we took the following steps:  We analyzed the data for
any AD, focusing on each study as a possible source of variation. 
Then we dropped the one study (East Boston) found to be a
statistically significant source of variation relative to the set of
studies as a whole, and we reanalyzed the remaining ones.  No other
study was a statistically significant source of variation, and thus
the heterogeneity among studies was eliminated with the elimination
of that single outlier.  If the remaining studies are used to
generate prevalence estimates, these are somewhat lower than the
original ones, for example, by 5 percentage points for men at the age
of 95.  One can debate about which estimates are better.  However,
although the heterogeneity of the data can be eliminated by dropping
the one outlier, we are reluctant to exclude the data from a major
American study of prevalence. 


--------------------
\19 The others (1) are one-phase studies, including all elderly in
the population without use of an initial screen, (2) use a sensitive
screen so that no cases are missed at the prediagnostic phase, or (3)
statistically correct for the expected rate of missed cases. 


GAO CONTACT AND STAFF
ACKNOWLEDGMENTS
=========================================================== Appendix X

GAO CONTACT

Donald M.  Keller, Evaluator-in-Charge, (202) 512-2932

STAFF ACKNOWLEDGMENTS

This report was prepared under the direction of George Silberman. 
Sushil K.  Sharma, Assistant Director, and L X.  Hy were responsible
for much of the research and analysis. 


*** End of document. ***