Web Documents



Blaxill, 2004

Created By: Riley Quijano
Font: small Large
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1497666/pdf/15504445.pdf
SYNOPSIS
Increases in the reported prevalence of autism and autistic spectrum disorders in
recent years have fueled concern over possible environmental causes. The author
reviews the available survey literature and finds evidence of large increases in
prevalence in both the United States and the United Kingdom that cannot be
explained by changes in diagnostic criteria or improvements in case ascertainment.
Incomplete ascertainment of autism cases in young child populations is the largest
source of predictable bias in prevalence surveys; however, this bias has, if anything,
worked against the detection of an upward trend in recent surveys. Comparison of
autism rates by year of birth for specific geographies provides the strongest basis
for trend assessment. Such comparisons show large recent increases in rates of
autism and autistic spectrum disorders in both the U.S. and the U.K. [1] Reported rates
of autism in the United States increased from <3 per 10,000 children in the 1970s
to >30 per 10,000 children in the 1990s, a 10-fold increase. [2]  In the United King-
dom, autism rates rose from >10 per 10,000 in the 1980s to roughly 30 per 10,000
in the 1990s.
Reported rates for the full spectrum of autistic disorders rose from the
5 to 10 per 10,000 range to the 50 to 80 per 10,000 range in the two countries. A
precautionary approach suggests that the rising incidence of autism should be a
matter of urgent public concern.
Time Trends in Autism
?
537
Public Health Reports / November–December 2004 / Volume 119
Since the 1960s, autism researchers have published more
than 50 surveys that provide estimates of the frequency of
autism in defined populations. This survey base provides a
significant opportunity for analysis, but also poses many dif-
ficulties with respect to interpretation. In recent years, these
difficulties have come into sharp relief as reports of higher
and rising prevalence rates have fueled concern over pos-
sible environmental causes.1,2 This survey literature has been
the subject of several reviews and commentaries,3–11 many of
them by a single author.3–8 This small set of reviewers have
shared a common interpretation regarding the trend evi-
dence, emphasizing the notion that autism is “not an ex-
tremely rare disorder,”10 while arguing that there is no evi-
dence for an increase over time in the prevalence of autism.5
In order to reconcile the full body of evidence with this
interpretation, several authors have suggested mechanisms
that might explain the reported increases in the apparent
absence of a secular trend, including diagnostic substitu-
tion,12–15 changing diagnostic standards,16 and improved
detection.17
The controversy over the true rates of autism might ap-
pear a simple matter, one that is best resolved prospectively,
with improvements in reporting procedures. In reality, the
persistence of the trend controversy reflects a vigorous de-
bate over causality in the broadest sense. Is autism primarily
a genetic disorder, as many would claim,18 or do environ-
mental factors play a stronger role than previously acknowl-
edged? In comparison to the methodological hypotheses,
the search for possible environmental causes has generated
far greater controversy, particularly as iatrogenic hypotheses
have been advanced and challenged.19–24
In the context of these controversies, a careful review of
the available literature is essential. Assessment of trend evi-
dence bears directly on the relative explanatory power of
environmental and genetic theories. Causal theories that
emphasize genetic inheritance carry greater weight if dis-
ease frequency is unchanged over time, whereas rising inci-
dence demands environmental explanations. Reliable trend
assessment requires a comprehensive synthesis of the best
available evidence.25,26
This article reviews the available survey evidence on the
prevalence of autism in order to demonstrate that the under-
lying rates—and not merely the reported rates—of autism
have risen sharply in the U.S. and the U.K. This discussion is
divided into two sections. The first section provides an analy-
sis of the impact of survey design choices on frequency and
trend estimates. The second section offers an in-depth quan-
titative and qualitative comparison of 11 U.K. and eight U.S.
surveys published since the mid-1960s (see Table 1).
IMPACT OF DIFFERENCES IN SURVEY DESIGN
Definitions
The determination of disease frequency requires, first, a
definition of the disease event. In the case of autism, this
determination has long been based on subjective psycho-
logical assessments. [3]Because autism has no well-defined bio-
logical markers, the condition itself is an hypothesis, “a sug-
gestion that behind the behavioral description [lies] a disease
entity.”
27 Not surprisingly, over the six decades since Leo
Kanner first described the condition,28 the standard definition
has been updated many times (see Table 2), incorporating
changes in nomenclature, diagnostic criteria, age of onset,
and disease categories. While the general intent has been to
make diagnostic decisions more consistent, these revisions
complicate the challenge of comparing findings across and
within studies.
Nomenclature. [4] The word autism has remained in constant use
since Kanner adopted the term to describe early infantile
autism or infantile autism in 1943
,28 yet the accompanying
modifiers have changed with time. In Rutter’s influential
modernization of the definition in 1978,27 three related
terms—autism, infantile autism, and childhood autism—were
used somewhat interchangeably. The third edition of the
Diagnostic and Statistical Manual of Mental Disorders (DSM-
III)29 used infantile autism as the core descriptor, but also
placed autism in the context of the pervasive developmental
disorders (PDDs) for the first time. In 1987, a revised edition,
DSM-III-R,30 abandoned the term infantile autism, in part to
recognize cases in which the onset of symptoms did not
occur in early infancy; DSM-III-R distinguished between au-
tistic disorder and pervasive developmental disorder, not otherwise
specified (PDD-NOS). In 1993, the World Health Organiza-
tion published the International Classification of Diseases, 10th
revision (ICD-10) definitions,31 which used the term childhood
autism. One year later, the fourth DSM edition,32 in most
respects identical to ICD-10, used the term autistic disorder.
Diagnostic criteria. Kanner and Eisenberg provided the first
formal set of criteria for the diagnosis of autism in 1956; the
criteria focused on two dimensions of the condition: “a pro-
found lack of affective contact” and “repetitive, ritualistic
behavior, which must be of an elaborate kind.”33 Practitio-
ners found these criteria somewhat problematic over the
years, in particular because they omitted the unusual lan-
guage and communication patterns that seemed to many a
core element of the disease. Rutter’s refinement introduced
the concept of simultaneous deficits in three behavioral
domains: impaired social relationships, impaired language
and communication skills, and “insistence on sameness.”27
In one form or another, these three domains have deter-
mined the operational definition of autism ever since.
Some evidence suggests that the shift away from the
Kanner criteria may have effectively broadened the scope of
the diagnosis. One group of investigators in Finland applied
both the Kanner criteria and the ICD-10 criteria in a popu-
lation survey.34 These researchers interpreted the Kanner
criteria as more restrictive and reported a lower rate for
what they called classic autism of 5.6 per 10,000, compared
with 12.2 per 10,000 for childhood autism. The mean preva-
lence rate for 11 studies using the Kanner criteria was two
per 10,000.35–45 This compares to a mean of seven per 10,000
for 13 surveys of infantile autism that applied either the Rutter
or DSM-III criteria or similar “post-Kanner” clinical criteria.46–58
Age of onset is the one criterion that has changed most
measurably over time. Kanner and Eisenberg implied an age
assumption by using the term early infantile autism. Rutter27
(followed by DSM-III) set a specific age of onset limit of
earlier than 30 months. The DSM III-R criteria relaxed this
limit, requiring only that the age of onset occur “during
538
?
Viewpoint
Public Health Reports / November–December 2004 / Volume 119
Table 1. Overview of 54 published reports of studies that provide disease frequency statistics
on autism, autistic spectrum disorders (ASDs), and related disorders.
Number of cases
Author(s)
Location
Year of publication
per 10,000 children
Lotter35
England
1966
4.1
Treffert36
U.S.
1970
0.7
Yamazaki et al.37
Japan
1971
2.6
Tanino38
Japan
1971
1.1
Haga and Miyamoto39
Japan
1971
1.1
Nakai40
Japan
1971
1.7
Brask41
Denmark
1972
4.3
Wing and Gould42
England
1979
4.9
Bohman et al.46
Sweden
1981
6.1
Hoshino et al.43
Japan
1982
2.3
Ishii and Takahashi47
Japan
1983
16.0
McCarthy et al.44
Ireland
1984
4.3
Gillberg48
Sweden
1984
2.0
Steinhausen et al.49
Germany
1986
5.8
Steffenburg and Gillberg50
Sweden
1986
4.6
Burd et al.45,a
U.S.
1987
1.2
Matsuishi et al.51
Japan
1987
15.5
Tanoue et al.52
Japan
1988
13.9
Bryson et al.86
Canada
1988
10.1
Ritvo et al.53
U.S.
1989
2.5
Aussilloux et al.54
France
1989
4.7
Sugiyama and Abe55
Japan
1989
13.0
Cialdella and Mamelle56
France
1989
5.1
Gillberg et al.60
Sweden
1991
5.8
Fombonne and du Mazaubrun57
France
1992
4.9
Herder88
Norway
1993
5.5
Rumeau-Rusquette et al.58,79
France
1994
3.1
Deb and Prasad82
U.K.
1994
9.0
Honda et al.66
Japan
1996
21.1
Fombonne et al.79
France
1997
5.4
Wignyosumarto et al.87
Indonesia
1997
11.7
Webb et al.89
U.K.
1997
7.2
Arvidsson et al.93
Sweden
1997
10.0
Sponheim and Skejdal94
Norway
1998
3.8
California Department of Developmental Services2,b
U.S.
1999/2003
31.2 (peak)
Taylor et al.20
U.K.
1999
5.3
Kadesjo et al.85
Sweden
1999
24.0
Irie90
Japan
1999
10.4
Kielinen et al.34
Finland
2000
5.6
Baird et al.78,c
U.K.
2000
30.8
Hillman et al.84
US
2000

Chakrabarti and Fombonne61
U.K.
2001
16.8
Fombonne et al.62
U.K.
2001
26.1 (ASDs)
Powell et al.67
U.K.
2001

Kaye et al.68
U.K.
2001
16.3
Bertrand et al.80
U.S.
2001
40.0
Sturmey and James83
U.S.
2001
16.0
Davidovitch et al.91
Israel
2001
9.9
Magnusson and Saemundsen92
Iceland
2001
8.6
Croen et al.12
U.S.
2002
11.0
Scott et al.95
U.K.
2002
57.0 (ASDs)
Lingam et al.63
U.K.
2003
14.9
Gurney et al.73
U.S.
2003
3.0–52.0 (ASDs)
Yeargin-Allsopp et al.75
U.S.
2003
34 (ASDs)
aA later survey77 examined the same population. The authors consider that the findings of the second survey confirm the original findings.
bAn earlier survey1 examined the same population. The authors consider the later report to contain the more accurate estimates.
cAn earlier survey74 examined the same population. The authors consider the later report to contain the more accurate estimates
Time Trends in Autism
?
539
Public Health Reports / November–December 2004 / Volume 119
infancy or childhood.” ICD-10 and DSM-IV set an age limit
at 36 months, in part a reaction to the problem posed by the
expanded definition introduced in DSM-III-R.59
Other autistic spectrum diagnoses. DSM-III introduced, in a
broad sense, the concept of a spectrum of autistic disorders
through PDDs, and specifically introduced the term atypical
PDD, which some found useful,45,77 while others applied the
term autistic-like condition.50,60 DSM-III-R introduced the term
PDD-NOS, which was retained in DSM-IV. By contrast, ICD-
10 opted to use the term atypical autism.
In recent years, researchers have begun adopting the
phrase autistic spectrum disorders (ASDs), used interchange-
ably with PDDs. The ASDs include autism and PDD-NOS, as
well as a larger list of related disorders, including Asperger’s
syndrome, childhood disintegrative disorder, and Rett’s syndrome.
Asperger’s cases represent approximately 14% to 19% of the
ASD population,20,61–63 while reported rates are very low for
childhood disintegrative disorder64 and Rett’s syndrome (a
well-characterized genetic disorder).65
Methodological issues: nomenclature and diagnostic criteria.
[5]The core definition of autism has remained relatively
stable since Rutter’s introduction of the three main
behavioral domains in 1978,
27 facilitating comparison
of rate estimates over time.
• Early surveys that applied the Kanner criteria may
have reported lower autism rates than later studies
because the Kanner criteria were more restrictive. This
modest effect may contribute to a perception of an
increasing trend over time.
• The three sets of DSM criteria for autism have varied
modestly in breadth. Some have argued that the move
from DSM-III to DSM-III-R broadened the concept of
Table 2. Changes in nomenclature and major shifts in diagnostic criteria in widely accepted definitions of autism
Kanner and
Eisenberg33
Rutter27
DSM-III29
DSM-III-R30
ICD-1031/DSM-IV32
Date published
1956
1978
1980
1987
1994
Larger category


PDD
PDD
PDD
Autism
Nomenclature
Early infantile
Infantile autism;
Infantile
Autistic disorder
Autistic disorder
autism; infantile
autism;
autism
autism
childhood autism
Age at onset
None
By 30 months
By 30 months
During infancy or
By 36 months
of symptoms
specified
childhood
Related disordersa

Other infantile
Infantile autism/
PDD-NOS
PDD-NOS;
psychoses
residual state;
Asperger’s syndrome;
atypical PDD
Rett’s syndrome;
childhood onset
disintegrative disorder
aA number of studies based on the diagnostic criteria outlined in DSM-III and DSM-III-R used the term “autistic-like condition” to refer to certain
related disorders, although this term was not included in the official definition of pervasive developmental disorders.
PDD = pervasive developmental disorder
NOS = not otherwise specified
autism, contributing to an apparent increase in preva-
lence over time; however, in the shift from the DSM-
III-R criteria to the DSM-IV/ICD-10 criteria, “a correc-
tive narrowing occurred.”59 Thus, differences over time
in the breadth of the diagnostic criteria are unlikely to
have had a meaningful effect on reported disease fre-
quency.
• Any comparison of surveys must take into account
differences in disease definitions, in particular whether
PDD-NOS and Asperger’s syndrome are included in
the definition of autism.
Measurements of disease frequency
The overwhelming majority of autism surveys have used
prevalence rates as the primary measure of disease frequency.
A small minority of studies have reported “incidence” rates
of various kinds,20,66–68 but the actual differences in assump-
tions, data sources, and methods make these “incidence”
calculations little different from reported prevalence rates.
The emphasis on prevalence reporting fits well with the
historical consensus among autism researchers that autism
is largely (if not exclusively) genetic in origin, unlikely to
vary in disease frequency over time, and unlikely to demon-
strate secular changes in incidence.3
In principle, incidence calculations are essential for trend
assessment. But developing incidence measures for autism
presents special difficulties. Measuring incidence rates re-
quires clearly identifiable incidence times. With respect to
autism, many researchers have conflated time of onset with
time of detection, due to both theoretical and practical
complexities. In practice, most researchers opt for simplicity
and report prevalence rates. Those few studies that report
incidence rates have used a wide range of incidence times,
540
?
Viewpoint
Public Health Reports / November–December 2004 / Volume 119
including age at diagnosis,20,67,68 year of birth,66,68 age at onset
of symptoms (sometimes called “age at parental concern”),20
and (although this measure is also used in calculating preva-
lence rates) age at entry into a services system.1,2
Prevalence calculations can vary from cumulative inci-
dence rates in numerous ways. First, they may underestimate
cumulative incidence by failing to count cases in which the
disease condition has lapsed. In autism, since the presence
of the disease condition is defined by its onset at a certain
stage of development, this concern is irrelevant in principle.
In practice, since the few follow-up studies have shown the
condition to be persistent,69,70 it may be unimportant. Sec-
ond, prevalence calculations may underestimate older popu-
lations if excess case mortality increases with age. Studies in
California have found excess mortality among autistic chil-
dren (largely due to accidental drowning deaths),71,72 but
not in autistic adults. In addition, this excess mortality was
extremely small relative to the population size. Third, most
prevalence surveys fail to adjust for net migration effects,
i.e., the effect of children with autism moving in and out of
the study area. These three issues have concerned some
reviewers,10,21 but there is little evidence in the literature for
a systematic large effect on either disease frequency or trend.
Finally, prevalence rates can vary from true disease rates
when case ascertainment is incomplete. This is an important
issue addressed further in the following section on case
finding.
The studies reviewed for the purposes of this article have
used four main approaches to reporting disease frequency:
1.
Point prevalence with no stratification by age or birth
year. This common approach was used in 23 of the
54 studies listed in Table 1.
2.
Point prevalence with stratification by birth year and/
or age. This approach has been used most frequently,
with 28 of 54 studies employing some form of strati-
fication. Studies reporting stratified prevalence rates
provide potentially useful trend information.
3.
Cumulative incidence, defined by many autism re-
searchers as the number of cases diagnosed within a
specified age-at-diagnosis window divided by the to-
tal population at risk.20,53,68,73 These studies are most
reliably informative with respect to trend, but they
are few in number.
4.
Incidence. Only one study claimed to measure inci-
dence.67 The authors gathered data on age at diag-
nosis but did not provide any data on birth years,
making comparison to other studies impossible.
Methodological issues: measurement of disease frequency.
• The common practice of using point prevalence mea-
sures facilitates the collection and comparison of sur-
vey data. Point prevalence measures that have been
stratified by age and/or birth year provide additional
information for trend assessment by allowing for the
comparison of frequencies and trends within as well
as across studies.
• Comparisons of prevalence estimates across studies
must take into account differences in study design
that affect frequency assessment, especially differences
in the scope of the autism definition. Comparisons of
birth cohort disease prevalence estimates within stud-
ies must take into account specific design choices that
affect trend assessment.
Case-finding: ascertainment issues
There are many practical obstacles to recognizing autism in
a young child. These problems start in the home, where
delays in recognition will produce ascertainment failures
under any survey method. The three behavioral domains of
autism are not medical problems and can be overlooked for
substantial periods of time: social interaction problems can
be interpreted as due to hearing impairments, and delays in
language development can be explained by the phenome-
non of “late-talking children.” Paradoxically, the “insistence
on sameness” of autistic children can also make them ap-
pear to be “easy” children, since they may make few de-
mands. Further delays can occur once concerned parents
have recognized that their child has serious developmental
issues; the process of scheduling a formal clinical assessment
can take many months.
While definitions of autism have generally specified the
time for the onset of symptoms as 30 to 36 months, the age
of diagnosis carries no similar requirement. Diagnoses can
come as early as 18 months with specialized methods,74,78 or
as late as 10 years of age. A number of autism surveys report
mean or median ages at diagnosis (see Table 3), with a wide
variation across the populations measured. A few surveys
reporting lower ages at diagnosis have based their calcula-
tions on truncated samples, removing later ages at diagnosis
from their calculations.20,61,67 Other studies report median
ages at diagnosis ranging from 3 to 6 years of age,1,2,63,68 or
means ranging from 3.9 to 6.9 years of age.8,36,75,76 This wide
variation and relatively late average age at diagnosis argues
for caution in interpreting surveys based on young children
and especially those based on survey populations that in-
clude children younger than 5 years of age.
Several recent surveys using administrative data have dem-
onstrated the disproportionate effect of ascertainment bias
on younger birth cohorts. Reports from states that have
updated their case files over a period of several years dem-
onstrate clearly the way in which young children can enter a
survey population gradually over time. This effect has been
described in Minnesota and California.2,73 Recent experi-
ence with California survey data has revealed how misinter-
pretation of trends based on mixed age groups in a survey
population can lead to erroneous conclusions.12–15
Arguably, however, ascertainment bias can be reduced
with aggressive case-finding methods, which accelerate the
age of diagnosis. One useful test of the impact of ascertain-
ment is to compare findings from repeated case-finding
efforts, i.e., pairs of surveys that cover identical geographic
areas and birth populations but in different time frames.
Three such pairs of surveys have been published (Table 4).
One such pair was initiated in North Dakota,45,77 where two
PDD prevalence surveys (neither stratified by age) were con-
ducted more than a decade apart by the same research
group. The first survey identified 98% of the cases found 12
years later; this high case ascertainment rate was likely due
to a high median age in the earlier survey, a wide age range,
Time Trends in Autism
?
541
Public Health Reports / November–December 2004 / Volume 119
and low overall prevalence in the survey population. A sec-
ond pair of surveys (with stratified age cohorts) was con-
ducted four years apart in a region in Sweden that includes
the city of Goteburg and the rural Bohuslan area.50,60 The
first survey found a wide variation in prevalence rates across
age cohorts, ranging from two per 10,000 to 11 per 10,000.
These researchers reported complete ascertainment after
Table 3. Age of autism diagnosis in selected U.K. and U.S. studies
Author(s)
Age at diagnosis
Time period
United Kingdom
Taylor et al. 199920
3.1 yearsa (median)
1979–1982 (birth year)
Powell et al. 200067
3.1 yearsa (median)
1991–1996 (birth year)
Chakrabarti and Fombonne 200161
3.4 yearsb (mean)
1998–1999 (year diagnosed)
Kaye et al. 200168
4.6 years (median); 4.0–6.0 range
1989–1999 (birth year)
Lingam et al. 200363
3.3 years (median)
1984–1995 (birth year)
United States
Treffert 197036
5.1 years (mean)
1962–1967 (year diagnosed)
California Department of Developmental Services 19991
5–9 years (median)
Up to 1987 (year entered
services system)
Croen et al. 200212
6.9 years (mean)
1987–1994 (birth year)
Mandell et al. 200276
6.3 years (mean)
1993–1999 (claim year)
California Department of Developmental Services 20032
?4 years (median)
Up to 2002 (year entered
services system)
Yeargin-Allsopp et al. 200375
3.9 years (mean)
1986–1993 (birth year)
aCase collection in this study was truncated to exclude children who were diagnosed after 60 months of age. This exclusion makes the median
age at diagnosis a low estimate of eventual age of diagnosis in this case population.
bCase collection in this study was truncated to exclude children who were diagnosed after 78 months of age. This exclusion makes the mean
age at diagnosis a low estimate of eventual age of diagnosis in this case population.
Table 4. Description of three survey pairs: identical geographic areas and populations
Case ascertainment
Location, description, and findings
Earlier study
Later study
rate in earlier study
North Dakota
Burd et al. 198745
Burd et al. 200077
Age range at time of study
2–18 years
14–30 years
Number of cases
59
60
98%
Prevalence of PDDs
3.26 per 10,000
3.51 per 10,000
Goteburg/Bohuslan, Sweden
Steffenburg and Gillberg, 198650
Gillberg et al. 199160
1975–1977
Age range at time of study
7–10 years
11–13 years
Number of cases
27
26a
100%a
1978–1980
Age range at time of study
4–6 years
8–10 years
Number of cases
19
26
73%
1981–1984
Age range at time of study
?3 years
4–7 years
Number of cases
6
22
27%
Prevalence of PDDs
6.6 per 10,000
9.5 per 10,000
South East Thames Health Region, U.K.
Baron-Cohen et al. 199674
Baird et al. 200078
Age at time of study
18 months
7 years
Number of cases
10
50
20%
Prevalence of autism
6.3 per 10,000
30.8 per 10,000
aLower number of cases in later study due to one child having moved out of study region.
PDD = pervasive developmental disorder
an interval of four years for the cohort aged 8 to 10 years in
the original survey. For the cohort aged 4 to 6 years, the
ascertainment rate declined to 73%, and for children younger
than age 4, the ascertainment rate was only 23%. In this
survey pair, what appeared to be a declining time trend in
the first survey population was not replicated in the second
survey. Finally, a third pair of surveys was initiated in the
542
?
Viewpoint
Public Health Reports / November–December 2004 / Volume 119
South East Thames Health Region in England to test the
efficacy of a newly designed Checklist for Autism in Toddlers
(CHAT).74,78 The researchers first carried out a careful sur-
vey of 18-month-old children to see if the CHAT could pre-
dict autism at an early age. Although the CHAT checklist
showed a high specificity rate (98%) for childhood autism,
the first survey failed to ascertain the majority of cases found
in the same population six years later. The autism ascertain-
ment rate for the combined medium and high risk groups
from the first survey was only 20%.
This pattern of ascertainment bias can also be detected
in the seven single surveys that provide annual birth cohort
prevalence estimates across wide age ranges.2,43,56,63,75,79,80 (Per-
sonal communication, Ron Huff, PhD, California Depart-
ment of Developmental Services, February 2003). Birth
cohort prevalence rates for children ages 6 or older show a
wide range of trends, from declining to flat to sharply in-
creasing, across the seven surveys. Yet each study shows a
pronounced and rapid decline in prevalence rates with de-
creasing age in age cohorts younger than 6 years. Interest-
ingly, these declines are comparable in administrative surveys
and active case-finding surveys.
If ascertainment lags in autism surveys have a predictable
effect on trend assessments, then this effect can only be
more pronounced for other autism spectrum diagnoses with
later median ages at diagnosis. One recent U.K. survey re-
porting age at diagnosis for the full autism spectrum re-
ported a median age at diagnosis for Asperger’s syndrome of
8.1 years, compared with a median age of 4.3 years for
atypical autism (PDD-NOS), and 3.3 years for childhood
autism.63 This pattern suggests that any survey reporting trend
evidence for populations of children younger than 10 years
of age may understate both the ASD trend and the relative
proportion of core autism cases vs. cases of less severe ASDs.
Methodological issues: ascertainment.
• Obstacles to recognizing autism in children younger
than 5 years of age can lead to ascertainment bias,
distorting trend evidence in surveys that do not con-
trol for age effects, either reducing a positive trend or
suggesting a negative trend where reported rates are
stable.
• Survey populations that have high proportions of
young children may yield reported prevalence signifi-
cantly below actual disease frequency.
• Aggressive case-finding may reduce ascertainment bias
by accelerating diagnoses in small populations, but
appears not to eliminate it. The ascertainment effect
holds with comparable strength in administrative sur-
veys and in surveys using aggressive case-finding meth-
ods.
• The later age at diagnosis for Asperger’s syndrome
and PDD-NOS increases the risk of misinterpretation
of trend evidence in surveys that attempt to measure
disease frequency of all ASDs.
Case finding: population restrictions, case
identification, and case validation methods
Population restrictions. Some autism researchers have made
specific choices that restrict their surveys’ population cover-
age. These include the following: limiting case finding to
mentally retarded children or children in special schools,81,82
a choice likely to downwardly bias reported rates; focusing
on males only,68 a choice likely to yield higher rates based on
the high male/female ratio in autism; or truncating the
case-finding period at an early age,61,68 a choice likely to
downwardly bias reported rates.
Case identification. Increasingly, autism surveys have been
conducted using cases identified entirely from centralized
databases. These databases are generally designed to facili-
tate the administration of service delivery and may be more
or less reliable as epidemiological resources. Often called
administrative surveys, efforts relying on central databases
are less costly to conduct than surveys that rely on indepen-
dent identification and validation of cases, often from mul-
tiple sources, especially when the former do not require
individual interviews for case validation. In the United States,
special education benefits are administered at the state level,
so several recent administrative surveys have been under-
taken at the state level.2,73,83,84 In the United Kingdom, health
care services databases have provided the case identification
for similar administrative surveys.20,63,68
When researchers rely on service providers rather than a
central database to identify potential cases, cases may be
missed through non-cooperation at many stages. Concerns
regarding the effectiveness of case-finding in such surveys
have been extensively reviewed elsewhere;3,4 issues include
institutional coverage and response rates, parent refusal rates,
diagnostic interview completion rates, and the quality of
diagnostic instruments.
One recent review has hypothesized that rising rates of
autism have resulted in part from improved case identifica-
tion in recent surveys based on small survey populations
(?50,000).11 These authors argue that small surveys have
generated a disproportionately high number of reports of
prevalence in excess of 10 per 10,000. Arguing against this
hypothesis is the observation that surveys of large popula-
tions in the U.S. and the U.K. have yielded ASD and autism
rates consistently greater than 10 per 10,000 and have also
reported upward trends during their survey periods.2,63,73,83
The same review also suggests that close coordination
with a “routine developmental check” in preschool children
can explain high levels of case identification, pointing for
support to several Japanese surveys that have found preva-
lence rates greater than 10 per 10,000.51,52,55,66 Arguing against
this suggestion are numerous earlier Japanese surveys that
reported lower autism rates,37–40,43,47 given no evidence of
changing practices in Japan. Additionally, none of the pub-
lished Japanese surveys has ever reported a rate of autism
greater than 25 per 10,000, the range of concern for recent
surveys in the U.S. and U.K.
Case validation. Methodological features of both administra-
tive and intensive case-recruitment surveys can affect the
quality of diagnoses and the accuracy of disease rates. These
include choice of screening instruments and instruments
for intensive assessment and/or record review, numbers of
informants, and methods to ensure inter-rater reliability.
These methods are reviewed in depth elsewhere.3,4
Recent experience has shown that surveys with aggressive
case validation methods can uncover more cases of autism
Time Trends in Autism
?
543
Public Health Reports / November–December 2004 / Volume 119
and PDD than surveys that work entirely from administrative
records. But even these experiences can provide support for
the usefulness of administrative databases. One recent study
in Atlanta estimated that only 41% of children with PDDs
had been given an autism diagnosis by school administra-
tors,75 but the survey’s definition of PDDs included Asperger’s
syndrome, whereas the school system may have applied a
more restrictive definition of autism. School sources were
the main source for case identification: only 3% of the chil-
dren in the sample who were classified by the researchers as
having PDDs were identified through databases other than
school records. This low percentage of missed cases under-
scores the value of central databases in identifying cases and
also the importance of specifying PDD categories for valida-
tion purposes.
Methodological issues: population restrictions, case identification,
and case validation method.
• Use of restricted populations based on age or receipt
of services may lead to underestimating true disease
frequency by reducing the potential for case identifi-
cation; alternatively, restricting the study population
to males may lead to overestimating the true frequency.
• Administrative surveys based on central databases may
provide lower estimates of disease frequency than sur-
veys that rely on independent case-finding and valida-
tion. The impact of this effect is hard to assess based
on current data and is influenced by the quality and
coverage of specific databases.
• Smaller survey populations may allow for better case
identification rates.
• Differences in case validation methods may explain
disease frequency differences across surveys.
REVIEW OF U.S. AND U.K. STUDIES
Previous autism reviews have consistently concluded that
autism incidence rates show no clear time trend.3–11 These
reviews, however, have shared three major flaws: use of flawed
meta-analytic methods, limited survey evidence, and inad-
equate correction for ascertainment bias.
Critical examination of methods for
cross-survey comparisons
A single author, Fombonne, has been the most active re-
viewer and interpreter of autism surveys.3–8 This reviewer has
conducted three meta-analyses of autism time trends.3,4,6 In
these cross-survey comparisons, he has correlated reported
prevalence estimates with the years in which the selected
surveys were published. The first meta-analysis showed an
increase in autism rates over time, but the correlation was
not statistically significant.3 The later analyses showed a ris-
ing trend that reached statistical significance.4,6 In each case,
however, the author attributed any positive trend in reported
prevalence rates to methodological changes. Other reviewers
have supported him in this opinion.10,11
Until recently, this was a reasonable inference to draw
from the literature. Reviewers have emphasized the diffi-
culty of assessing trends against a background of regular
changes in diagnostic criteria. When dates of publication
are used in a meta-analysis, an analysis of time trends will
reveal little beyond an apparent effect of changes in diag-
nostic standards. An assessment (not shown) of all published
prevalence studies by diagnostic criteria set (Kanner,35–45
Rutter/DSM-III,46–58 DSM-III-R,60,82,85–92 and DSM-IV through
20012,12,20,34,61–63,66–68,73,75,78–80,92–96) reveals that prevalence was
relatively stable until the most recent set of diagnostic crite-
ria was introduced.
Relying on the date of publication can obscure large
differences in the study populations observed. For example,
three Scandinavian studies published in a three-year period
used markedly different populations: a Swedish study pub-
lished in 1997 measured the prevalence of autism in a 3- to
6-year-old population born from 1988 through 199193; an-
other Swedish study published two years later, in 1999, fo-
cused on 7-year-olds who were born in 1985, several years
before the children in the first group85; while a Norwegian
study published between the two covered a wide range of
age cohorts and measured prevalence rates going back as far
as 1978.94
In addition, combining surveys from locations as widely
separated as Yokohama, California, and Goteburg greatly
increases the potential for non-comparability. None of the
three meta-analyses chose surveys based on country group-
ings, therefore risking the introduction of confounding en-
vironmental factors that are geographically specific. With
the publication since 1999 of six U.S. studies and seven U.K.
studies, the opportunity for useful geographic comparison
has improved greatly.
An alternative approach is to synthesize evidence from
autism surveys using the following methods:
• Analyze surveys from a single country separately in-
stead of grouping separate regions together.
• Use the most common measure of disease frequency:
point prevalence.
• Compare prevalence reports based on birth dates of
children in the survey population, not the publication
date.
• Highlight differences in diagnoses, distinguishing (us-
ing DSM-IV nomenclature) autistic disorder from PDD-
NOS and Asperger’s syndrome.
• Identify all possible sources of bias resulting from spe-
cific survey design choices.
Cross-survey comparisons: U.K. and U.S. studies
Tables 5 and 6 show how factors likely to affect comparisons
across surveys (factors affecting frequency estimates) and
within surveys (factors affecting trend estimates) impact 11
U.K. studies and eight U.S. studies that provide usable au-
tism rate data.
Frequency factors include the following: use of the Kanner
criteria, reliance on administrative databases, early age trun-
cation of the survey population, restrictions of the survey
population based on school or mental status, and gender
restrictions. Surveys applying one of these factors may re-
port lower autism rates than other surveys; however, in the
case of gender restrictions, relative rates will be higher when
only males are included. Although scant evidence supports a
population effect, the sizes of the survey populations are
544
?
Viewpoint
Public Health Reports / November–December 2004 / Volume 119
Table 5. Description of 11 U.K. surveys of autism prevalence, with factors affecting the assessment of disease frequency and time trends
Lotter
Wing and Gould
Deb and Prasad
Webb et al.
Taylor et al.
Fombonne et al.
196635
197942
199483
199789
199920
200162
Geographic area
Middlesex
Camberwell
Northeastern Scotland
Wales
North Thames area
Multiple
Birth year range
1953–1955
1955–1970
1969–1983
1977–1989
1979–1992
1984–1994
Survey population
78,000
35,000L
101,814
73,301
490,000
10,438L
Number of cases per
10,000 children
Autism
4.1 (95% CI 2.7, 5.5)
4.9 (95% CI 2.9, 7.8)
9 (95% CI 7.2, 11)
7.2 (95% CI 5.3, 9.3)
5.3

PDD, not including Asperger’s
6.9a



8.7
21.1
ASD, including Asperger’s




10.1
26.1 (95% CI 16.2, 36.0)
Factors affecting prevalence
Diagnostic criteria
Kanner33M
Kanner33M
DSM-III-R30
DSM-III-R30
ICD-1031
ICD-1031
Central database
No
No
No
No
Yes
No
Age truncation
No
No
No
No
No
No
Population restriction
No
No
Yes
No
No
No
Gender restriction
No
No
No
No
No
No
Factors affecting trend
Year of birth breakdowns
Nob
Nob
Yesb
Yesc
Yes, but NAb
Yesb
Age variation (range)
No (8–10)
No (?15)
Yes (5–19)M
Yes (3–15)M
Yes (5–16)
Yes (5–15)M
Ages ?5 years
No
Yes
No
YesM
No
No
Asperger’s included
No
No
No
No
Yes
YesM
continued on p. 545
Time Trends in Autism
?
545
Public Health Reports / November–December 2004 / Volume 119
Table 5 (continued). Description of 11 U.K. surveys of autism prevalence, with factors affecting the assessment of
disease frequency and time trends
Chakrabarti and
Lingam et al.
Kaye et al
Scott et al
Fombonne
Baird et al.
200363
200168
200295
200161
200078
Geographic area
North London
U.K.
Cambridgeshire
Staffordshire
South East Thames
Health Region
Birth year range
1985–1994
1988–1993
1988–1994
1992–1995
1993
Survey population
186,206
70,000
43,472
15,500L
16,235L
Number of cases per 10,000 children
Autism
14.9
16.3

16.8 (95% CI 11.0, 24.6)
30.8 (95% CI 22.9, 40.6)
PDD, not including Asperger’s
25.4


52.9
57.9 (95% CI 46.8, 70.9)
ASD, including Asperger’s
30.5

57 (95% 49.5, 65.8)
62.6 (95% CI 50.8, 76.3)

Factors affecting prevalence
Diagnostic criteria
ICD-1031
ICD-1031
ICD-1031
ICD-1031
ICD-1031
Central database
YesM
YesM
No
No
No
Age truncation
No
Yes (2–5)M
No
Yes (2.5–6.5)M
No
Population restriction
No
No
No
No
No
Gender restriction
No
 Yes (male)L
No
No
No
Factors affecting trend
Year of birth breakdowns
Yesd
Yesd
Nob
Nob
Nob
Age variation (range)
Yes (1–16)M
No
No (5–11)
No (2–6)
No (7)
Ages <5 years
YesM
NA
No
Yes
No
Asperger’s included
YesM
No
Yes
Yes
No
NOTE: Arrows indicate study design features with potential impact on reported outcomes, with downward-pointing arrows indicating a possible downward effect on frequency or trend
estimates and upward-pointing arrows indicating a possible upward impact.
aCombined prevalence of 6.9 per 10,000 for “autistic” children and children with “similar but less marked features.”
bAutism rates by age cohort and times of data collection provided by authors, allowing alignment of age cohort with birth years.
cAutism rates by year of birth provided by authors.
dAutism rates by year of birth provided by authors graphically.
CI = confidence interval
PDD = pervasive developmental disorder
ASD = autistic spectrum disorder
NA = not applicable
546
?
Viewpoint
Public Health Reports / November–December 2004 / Volume 119
Table 6. Description of eight U.S. surveys of autism prevalence, with factors affecting the assessment of disease frequency and time trends
California
Sturmey
Yeargin-
Ritvo
Burd
Department of
and
Allsopp
Bertrand
Gurney
Treffert
et al.
 et al.
Developmental
James
et al.
et al.
et al.
197036
198953
198745
Services 20032
200183
200375
200180
200373
Geographic area
Wisconsin
Utah
North
California
Texas
Atlanta,
Brick
Minnesota
Dakota
 GA
Township, NJ
Birth year range
1949–1969
1960–1984
1967–1991
1970–1997
1983–1994
1986–1993
1988–1995
1989–1993
Survey population
899,750
769,620
180,986
14,200,000
3,565,000
289,456
8,896
1,600,000
Number of cases
per 10,000 children
Autism
0.7
2.47
1.16
12.6
16

40

(95% CI 0.6, 0.9) (95% CI 2.1, 2.8)
(95% CI 28, 56)
PDD, not including
3.1

3.26





Asperger’s
(95% CI 2.5, 4.2)
ASD, including





34
67
20-66
Asperger’s
 (95% CI 32, 36)
(95% CI 51, 88)
Factors affecting
prevalence
Diagnostic criteria
Kanner33M
DSM-III29
Kanner33M
DSM-IV32

DSM-IV32
DSM-IV32
DSM-IV32
Central database
YesM
No
No
YesM
YesM
No
No
YesM
Age truncation
No
No
No
No
No
No
No
No
Population restriction
No
No
No
No
No
No
No
No
Gender restriction
No
No
No
No
No
No
No
No
Factors affecting trend
Year of birth
breakdowns
Noa
Yesc
Yesb
Yesc
Yes, but NAa
Yesd
Yesa
Yesc
Age variation (range)
No (3–12)
3–27M
1–1M
Yes (5–32)
Yes (6–17)M
Yes (3–10)M
Yes (3–10)M
No
Ages ?5 years
Yes
YesM
YesM
No
No
YesM
YesM
No
Asperger’s included
No
No
No
No
No
YesM
Yes/No
YesM
NOTE: Arrows indicate study design features with potential impact on reported outcomes, with downward-pointing arrows indicating a possible downward effect on frequency or trend
estimates and upward-pointing arrows indicating a possible upward impact.
aAutism rates by age cohort and times of data collection provided by authors, allowing alignment of age cohort with birth years.
bNumber of autism cases by year of birth provided by authors graphically, allowing calculation of rates using state census of live births.96
cAutism rates by year of birth provided by authors.
dAutism rates by year of birth provided by authors graphically.
CI = confidence interval
PDD = pervasive developmental disorder
ASD = autistic spectrum disorder
NA = not applicable
Time Trends in Autism
?
547
Public Health Reports / November–December 2004 / Volume 119
shown in the Tables and may also be a factor affecting
frequency.
Factors affecting trend assessment relate to ascertainment
bias and include the following: the availability of trend infor-
mation for the survey period, the inclusion of a range of age
groups in the survey population, the inclusion of children
younger than 5 years of age in the survey population, and
the inclusion of Asperger’s syndrome in the survey scope.
Each of the last three factors may depress evidence of an
underlying trend. A large trend effect may also affect com-
parisons across surveys.
Figure 1 shows a comparison of reported prevalence rates
across surveys from the U.K. and U.S., using the mid-point
year of birth as the basis for comparing survey populations.
This comparison demonstrates that reported autism rates in
the U.K. rose from ?10 per 10,00019,35,42,83,89 to the 17 to 31
per 10,000 range61,78 over 40 years and that reported PDD
rates rose to approximately 60 per 10,000 in the same pe-
riod.61,78,95 According to the reported confidence intervals
(CIs), these changes are statistically significant. The magni-
tude of the increase appears larger in the U.S. than in the
U.K. The U.K. data show a strong inflection point around
1990, whereas autism rates in the U.S. appear to have begun
their rise in the 1980s.
The amount of useful comparative information expands
when one examines details of the published reports (Figure
2). In the U.K., starting with the earliest surveys based on
NOTE: These graphs show prevalence estimates from 11 U.K. and 8 U.S. studies. For studies with survey populations spanning a range of birth
years, the midpoint of the birth year range is used.
aLotter 196635
bWing and Gould 197942
cDeb and Prasad 199482
dWebb et al. 199789
eTaylor et al. 199920
fFombonne et al. 200162
gLingam et al. 200363
hKaye et al. 200168
iScott et al. 200295
jChakrabarti and Fombonne 200161
Figure 1. Reported prevalence of autism and autistic spectrum disorders (ASDs), by midpoint year of birth,
United Kingdom and United States, 1954–1994
Midpoint year of birth
Midpoint year of birth
United Statesl–s
United Kingdoma–k
Cases per 10,000
1950
1960
1970
1980
1990
2000
60
40
20
0
1950
1960
1970
1980
1990
2000
o ASD
• Autism
60
40
20
0
kBaird et al. 200078
lTreffert 197036
mRitvo et al. 198953
nBurd et al. 198745
oCalifornia Department of Developmental Services 20032
pSturmey and James 200183
qYeargin-Allsopp et al. 200375
rBertrand et al. 200180
sGurney et al. 200373
children born before 1990 in Middlesex,35 Camberwell,42
Scotland,82 and Wales,89 the prevalence rates for autism were
generally well below 10 per 10,000. In surveys including
children born after 1990, reported autism rates were both
higher20 and rising.20,68,78 Each of these surveys had unique
features. A survey in the North Thames health region showed
a remarkable rising trend in numbers of autism cases by
birth year.20 An administrative survey with one of the most
credible trend reports (this design controlled for age-at-
ascertainment bias) found a clear increase using a measure
of cumulative incidence by birth year.68 This result is difficult
to compare to those of other surveys since this study used
data on boys only (introducing an upward bias) and admin-
istrative data truncated to 2- to 5-year-olds (introducing a
downward bias). The South East Thames survey provided a
well-supported and high estimate of autism prevalence—
30.8 per 10,000—in a single age cohort with a small popula-
tion and effective case-finding.78 One survey in Staffordshire
stands out as an exception;61 the reported autism rate is
modestly lower than in other recent surveys, while the re-
ported PDD rate (including Asperger’s syndrome) is the
highest ever reported in the U.K.
Evidence from U.K. surveys of the full range of ASDs also
supports an increasing trend. A survey that sampled mul-
tiple areas showed a rising trend from approximately 1990.62
Four recent surveys found similarly high rates,61,63,78,95 three
with rates of approximately 60 per 10,00061,78,95 and a fourth
548
?
Viewpoint
Public Health Reports / November–December 2004 / Volume 119
with rates peaking at ?50 per 10,000.63 These recent survey
reports can be compared with two earlier reports.20,35 A sur-
vey of children born in the 1950s found a low prevalence
rate of only 2.8 per 10,000 for children who had “similar but
less marked features” than children diagnosed with autism.35
A survey of children born from 1979 through 1992 found a
rate of 3.4 per 10,000 for atypical autism.20
The results of seven U.S. surveys are also shown in Figure
2. These surveys support conclusions similar to those for the
U.K. The early autism surveys show a clear convergence in
disease frequency from the 1950s through 1980, with four
different studies reporting rates consistently lower than three
per 10,000.2,36,45,53 Such low rates held in administrative sur-
veys and active case-finding surveys and across four states.
More recent data show a marked (if gradual) increase for
autism in California2 and also in New Jersey,80 to levels ?30
NOTE: These graphs show prevalence estimates from 11 U.K. and 8 U.S. studies. Horizontal lines represent estimates for multi-year birth
cohorts. Squares represent estimates for single-year birth cohorts. Broken lines are used to link multiple estimates from the same study.
aLotter 196635
bWing and Gould 197942
cDeb and Prasad 199482
dWebb et al. 199789
eTaylor et al. 199920
fFombonne et al. 200162
gLingam et al. 200363
hKaye et al. 200168
iScott et al. 200295
jChakrabarti and Fombonne 200161
Figure 2. Reported prevalence of autism and autistic spectrum disorders (ASDs), by year of birth,
United Kingdom and United States, 1953–1997
Year of birth
Year of birth
Year of birth
Year of birth
United Kingdom: autism
United States: autism
United Kingdom: ASDs
United States: ASDs
Cases per 10,000
Cases per 10,000
60
40
20
0
1950
1960
1970
1980
1990
2000
1950
1960
1970
1980
1990
2000
60
40
20
0
80
60
40
20
0
80
60
40
20
0
1950
1960
1970
1980
1990
2000
1950
1960
1970
1980
1990
2000
a
b
c
d
e
h
j
k
l
m
p
o
r
r
s
q
n
j
i
k
g
e
f
a
kBaird et al. 200078
lTreffert 197036
mRitvo et al. 198953
nBurd et al. 198745
oCalifornia Department of Developmental Services 20032
pSturmey and James 200183
qYeargin-Allsopp et al. 200375
rBertrand et al. 200180
sGurney et al. 200373
per 10,000. This 10-fold increase holds both within2 and
across surveys. An increasing trend was also reported in a
survey of Missouri children.84
Three U.S. surveys report rates for ASDs of 43 per 10,000,75
66 per 10,000,73 and 80 per 10,000.80 These rates are both
considerably higher than the earliest PDD rates reported in
North Dakota,45 which peaked at ?7 per 10,000 around
1980, and all three survey populations showed higher fre-
quency in later birth cohorts during survey periods. The
Minnesota survey was well controlled with respect to ascer-
tainment, and reported rising prevalence rates among 8-year-
olds over a five-year period; the authors attributed the three-
fold rate increase to changing administrative practices.73
Surveys of 3- to 10-year-old populations in Georgia75 and
New Jersey80 shared nearly identical ascertainment bias prob-
lems, with likely higher rates of Asperger’s syndrome in the
Time Trends in Autism
?
549
Public Health Reports / November–December 2004 / Volume 119
older age cohorts and likely incomplete ascertainment of all
PDDs in the 3 to 5 year age group.75,80 These biases may have
the effect of dampening evidence of rising rates while also
skewing the shape of the trend. (This skew may be especially
relevant in the Georgia study, in which a large number of
children who had not received a prior autism diagnosis were
reclassified into the PDD category.75) Like the autism evi-
dence, the data for PDDs point to roughly a 10-fold increase
for all ASDs from the 1970s to the early-to-mid 1990s.45,73,75,79
CONCLUSIONS
The evidence supporting an increasing rate of autism in the
U.K. and the U.S. has gathered strength. Although both the
nomenclature and the criteria set used to define autism
have changed over the years, these changes are not so great
as to prevent comparative analysis and do not explain major
differences in reported prevalence over time. The largest
stable source of variability in reported autism rates comes
from incomplete ascertainment in young age cohorts, which
limits the ability to detect an underlying and rising secular
trend. Reviews that have downplayed the rising trend have
overemphasized unimportant methodological problems,
employed flawed meta-analytic methods, and failed to take
into account the most relevant biases in survey methodolo-
gies. Point prevalence comparisons made within and across
surveys conducted in specific geographic areas, using year of
birth as a reference for trend assessment, provide the best
basis for inferring disease frequency trends from multiple
surveys. A comparison of U.K. and U.S. surveys, taking into
consideration changing definitions, ascertainment bias, and
case-finding methods, provides strong support for a conclu-
sion of rising disease frequency. The rate of autism in the
U.S., once reported as ?3 per 10,000, has now risen to ?30
per 10,000, a 10-fold increase. The rate of autism in the
U.K., once reported as ?10 per 10,000, has risen to roughly
30 per 10,000. Reported rates for ASDs in both countries
have risen from the 5 to 10 per 10,000 range to the 50 to 80
per 10,000 range. This review has found little evidence that
systematic changes in survey methods can explain these in-
creases, although better ascertainment may still account for
part of the observed changes. A precautionary approach
therefore suggests that increased rates of autism and related
disorders be accepted as an urgent public health concern.
Category: Second Semester Research Sources | Comments: 0 | Rate:
0 Votes
You have rated this item.
huluhub.com
umraniye escort pendik escort
sikis
bedava bahis