Epidemiological Surveys of Autism and Other Pervasive Developmental Disorders : An Update

Journal of Autism and Developmental Disorders, Vol. 33, No. 4, August 2003 (© 2003)
Epidemiological Surveys of Autism and Other Pervasive
Developmental Disorders: An Update
Eric Fombonne1
This paper was commissioned by the committee on the Effectiveness of Early Education in
Autism of the National Research Council (NRC). It provides a review of epidemiological stud-
ies of pervasive developmental disorders (PDD) which updates a previously published article
(The epidemiology of autism: a review. Psychological Medicine 1999; 29: 769–786). The de-
sign, sample characteristics of 32 surveys published between 1966 and 2001 are described. Re-
cent surveys suggest that the rate for all forms of PDDs are around 30/10,000 but more recent
surveys suggest that the estimate might be as high as 60/10,000. The rate for Asperger disorder
is not well established, and a conservative figure is 2.5/10,000. Childhood disintegrative disor-
der is extremely rare with a pooled estimate across studies of 0.2/10,000. A detailed discussion
of the possible interpretations of trends over time in prevalence rates is provided. There is
evidence that changes in case definition and improved awareness explain much of the upward
trend of rates in recent decades. However, available epidemiological surveys do not provide an
adequate test of the hypothesis of a changing incidence of PDDs.
KEY WORDS: Autism; pervasive developmental disorder; epidemiology; prevalence; incidence; childhood
disintegrative disorder; Asperger disorder.
INTRODUCTION
or as components of the lesser variant of autism thought
to index genetic liability to autism in relatives. With
Epidemiological surveys of autism started in the
the exception of recent studies, other pervasive devel-
mid-sixties in England (Lotter, 1966) and have since then
opmental disorders (PDD) falling short of diagnostic
been conducted in many countries. All epidemiologi-
criteria for autism (PDD-NOS, Asperger syndrome)
cal surveys have focused on a categorical-diagnostic
were generally not included in the definition used in
approach to autism that has relied over time on differ-
the earlier surveys although several epidemiological
ent sets of criteria; however, all surveys used a defin-
investigations yielded useful information on the rates
ition of autism which comprised severe impairments in
of these particular types of PDDs. These data are sum-
communication and language, social interactions, and
marized separately. The aims of this article are to pro-
play and behavior. This paper is therefore concerned
vide an up-to-date review of the methodological
with autism defined as a severe developmental disor-
features and substantive results of published epidemi-
der and not with more subtle autistic features or symp-
ological surveys. This article updates our previous
toms which occur as part of other, more specific,
review (Fombonne, 1999) with the inclusion of 9 new
developmental disorders, as unusual personality traits,
studies made available since then. A key feature of
the review was to rely on summary statistics through-
out in order to derive quantitative estimates for rates
1Correspondence should be addressed to Eric Fombonne, Canada
and correlates of autism-spectrum disorders. The spe-
Research Chair in Child and Adolescent Psychiatry, McGill
cific questions addressed in this review update are:
University, Department of Psychiatry, The Montreal Children’s
a) what is the range of prevalence estimates for autism,
Hospital, 4018 St. Catherine St. W., Montreal, QC H3Z 1P2,
Canada; e-mail: eric.fombonne@mcgill.ca
and related disorders? b) what proportion of autism
365
0162-3257/03/0800-0365/0 © 2003 Plenum Publishing Corporation

---

366
Fombonne
cases is attributable to specific associated medical dis-
with an overall median age of 8.0 across the 32 studies.
orders? c) is the incidence of autism increasing? d) what
Similarly, there is huge variation in the size of the pop-
are the other correlates of autistic-spectrum disorders,
ulation surveyed (range: 826–899,750), with a median
particularly with respect to race and ethnicity? e) what
population size of 65,300 subjects (mean = 153,700)
is the role, if any, of cluster reports in causal investi-
and about half of the studies relying on targeted popu-
gations of autism?; what are the directions for future
lations ranging in size from 15,000 to 152,000. The
epidemiological studies?
total number of children surveyed is just approaching
the 5 million figure (N = 4,918,000).
SELECTION OF STUDIES
STUDY DESIGNS
The studies were identified through systematic
searches from the major scientific literature databases
Most investigations have relied on a two-stage or
(MEDLINE, PSYCINFO) and from prior reviews
multi-stage approach to identify cases in underlying
(Wing, 1993; Zahner & Pauls, 1987; Fombonne, 1998,
populations. The first screening stage of these studies
1999). Only studies published in the English language
often consisted of sending letters or brief screening
were included in this review. This led to the exclusion
scales requesting school and health professionals to
of several questionnaire-based studies and of small-
identify possible cases of autism. Each investigation
scale investigations published in the national literature
varied in several key aspects of this screening stage.
of the relevant countries (Aussilloux et al., 1989; Haga
First, the coverage of the population varied enor-
& Miyamoya, 1971; Herder, 1993; Ishii & Takahashi,
mously from one study to another. In some (i.e., stud-
1983; Nakai, 1971) and, most certainly, of other simi-
ies 3, 17, 20, and 24), only cases already known from
lar studies unknown to the author. Overall, 32 studies
educational or medical authorities could be identified,
published between 1966 and 2001 were selected which
whereas in other surveys an extensive coverage of the
surveyed autism in clearly demarcated, non-overlap-
entire population, including children attending normal
ping samples. They are listed in Table I by order of their
schools (studies 1 and 25) or children undergoing sys-
appearance in the literature. Studies are numbered from
tematic developmental checks (studies 13, 19, 22, and
1 to 32, and these numbers are used subsequently to
32) was achieved. In addition, the surveyed areas var-
index each study. For several studies, the publication
ied in terms of service development as a function of
listed in Table I is the most detailed account or the ear-
the specific educational or health care systems of each
liest one; however, other published articles were used
country and of the year of investigation. Secondly, the
to extract relevant information from the same study,
type of information sent out to professionals invited
when appropriate.
to identify children varied from simple letters in-
cluding a few clinical descriptors of autism-related
symptoms or diagnostic checklists re-phrased in
SURVEY DESCRIPTIONS
non-technical terms, to more systematic screening
based on questionnaires or rating scales of known re-
The surveys were conducted in 13 countries
liability and validity. Thirdly, participation rates in the
and half of the results have been published during the
first screening stages provide another source of vari-
last decade (Table I). Details on the precise socio-
ation in the screening efficiency of surveys. Refusal
demographic composition and economical activities of
rates were available for 11 studies (studies 1, 5, 6, 9,
the area surveyed in each study were generally lack-
12, 14, 19, 20, 23, 25, and 30); the rate of refusal
ing; most studies were, however, conducted in pre-
ranged from 0% (study 25) to 29.4% (study 5), with
dominantly urban or mixed areas, with only 2 (studies
a median value of 10%. Fewer studies could examine
6 and 11) surveys carried out in predominantly rural
the extent to which refusal to participate or unco-
areas. The proportion of children from immigrant fam-
operativeness in surveys is associated with the likeli-
ilies was generally not available and very low in 5
hood that the corresponding children have autism.
surveyed populations (studies 11, 12, 19, 23, and 26);
Bryson et al. (1988), however, provided some evi-
only in study 4 was there a substantial minority of chil-
dence that those families who refused cooperation in
dren with an immigrant West-Indian background living
the intensive assessment phase had children with ABC
in the area. The age range of the population included
scores similar to other false positives in their study,
in the surveys is spread from birth to early adult life,
thereby suggesting that these children were unlikely

---

Epidemiological Surveys of Autism and Other Pervasive Developmental Disorders
367
)
2.7
5.9
4.1
16.2
2.8
1.6
1
21.1
95% CI
2.7; 5.5
2.4; 6.2
0.6; 0.9
2.1; 7.5
1.9;
3.9; 7.4
2.7;
1.4; 2.4
2.4;
1.3; 19.8
1.5;
3.4; 5.6
2.1;
4.1; 5.7
1
1
5.8; 14.4
6.7; 19.4
7.3;
2.3;
(
Continued
b
4.1
4.3
0.7
4.8
5.6
1.9
3.26
4.5
4.9
1.7
15.5
10.1
13.0
1
Prevalence
Rate /10,000
2.6
1.4
16
9.9 2.33
1.6
2.7
4.7
2.5
—
2.3
2.7 9.5
2.1
2.0
(M:F)
(23/9)
(12/7)
3.06
1.33 4.3
2.25
4.07 13.8
3.73 2.47
(4/2)
(52/17)
(16/1)
(42/9)
(15/6)
(129/13)
(24/15)
(16/12)
(36/16)
(43/16)
(106/26)
(190/51)
(54/20)
(105/49)
Gender ratio
0
15.6
—
—
30
—
20.5
—
55.8
—
—
—
23.8
—
—
34
18
13.3
normal IQ
Percent with
like
’
s
Criteria
rating scale
of Lotter
criteria
criteria
Diagnostic
Rating scale
Clinical
Kanner
24 Items
Kanner
Rutter
Kanner
Rutter
DSM-III
DSM-III
DSM-III
New RDC
DSM-III
DSM-III like
DSM-III
DSM-III-R
Clinical-
CARS
Autism
a
6
32
20
69
17
39
28
52
59
51
21
16
61
74
autism
142
132
241
154
Number of
subjects with
get
Prevalence Surveys of
5,120
78,000
46,500
25,000
69,000
65,000
32,834
95,394
20,800
12,263
78,106
899,750
609,848
279,616
135,180
769,620
274,816
able I.
population
T
Size of tar
g
est
Area
isconsin
Ken
Västerbotten
est Berlin
Dakota
Ibaraki
Nova-Scotia
(Rhône)
Gothenbur
+ Bohuslän
County
départements
ogyakarita
(SE of Jakarta)
Middlesex
Aarhus County
W
Camberwell
Fukushima-
County of
East
W
North 180,986
Kurume City
Southern
Part of
Nagoya
1 département
Utah
South-W
4 régions
14 ICD-10
Y
Country
UK
Denmark
USA
UK
Japan
Sweden
Ireland
Germany
USA
Japan
Japan
Canada
Japan
France
USA
Sweden
France
d
et al.
et al.
et al.
et al.
et al.
et al.
et al.
et al.
et al.
et al.
g
Authors
et al.
fert
ref
ing
anoue
ignyosumarto Indonesia
Lotter
Brask
T
W
Hoshino
Bohman
McCarthy
Steinhausen
Burd
Matsuishi
T
Bryson
Sugiyama et
Abe
Cialdella et
Mamelle
Ritvo
Gillber
Fombonne et du
Mazaubrun
W
et al.
ear of
1966
1970
1970
1976
1982
1983
1984
1986
1987
1987
1988
1988
1989
1989
1989
1991
1992
1992
Y
publication
1
2
3
4
5
6
7
8
9
1
No.
10
1
12
13
14
15
16
17
18

---

368
Fombonne
9.3
76.6
6.9
130.6
40.6
10.5
56.0
95% CI
1.4; 30.8
4.6; 6.1
5.3;
3.4;
7.9; 9.5
1
16.1;
5.8;
9.8; 16.6
14.7;
22.9;
10.5; 14.0
28.0;
16.2; 36.0
10.3; 23.2
5.35
8.7
7.8
21.08
46.4
72.6
30.8
12.2
26.1
13.2
16.8
Prevalence
Rate/10,000
1)
1)
1)
(5/1)
(M:F)
2.6
3.5
—
—
2.2 40.5
8.0
4.2
3.3
(13.5)
1.81
12/62)
6.57 7.2
(46/7)
(7/2)
2.09 5.2
15.7
(47/3)
4.12
(24/3)
(20/6)
.
(1
(23/1
(156/50)
(25/1
(46/1
Gender ratio
5.0
together
c
rs.
—
—
—
normal IQ
50.0
12.1
22.2
47.1
50.0
60
49.8
36.7
55.5
15.8
29.2
Percent with
s

g’
ger
Criteria
ICD-10-like
R/ICD-10
Gillber
criteria
(Asper
syndrome)
ICD10/
DSM-IV
ICD-9/
ICD-10
ICD-10
10
DSM-IV
Diagnostic
1991) a detailed examination showed that there was overlap between the
ICD-10
Clinical
DSM-III-R
ICD-10
ICD-10
ICD-10
DSM-III-
ICD-10
Clinical/
ICD-8/
DSM-IV
DSM-IV/
Mostly ICD-
ICD10/
et al.,
)
g
9
6
18
53
34
50
62
36
27
57
26
174
427
187
autism
Number of
(
Continued
subjects with
g, 1986; Gillber
able I.
get
T
826
8,537
1,941
8,896
73,301
65,688
10,438
43,153
15,500
population
325,347
490,000
152,732
Size of tar
g & Gillber
gan,
fenbur
Area
ales
ownship,
fordshire
okohama
départements
Glamor
W
County
Thames
Thames
est 25,377
Midlands
& Lapland)
T
New Jersey
Pays de
Galles
Y
3
South
Akershus
North
South-East 16,235
W
North (Oulu
Brick
Angleterre &
Whole Island
Staf
ing et Gould (1979).
W
g, 1984; Stef
(Gillber
Country
est coast)
Japan
France
UK
Sweden Mölnlycke
(W
Norway
UK
Sweden Karlstad
(Central)
UK
UK
Finland
USA
UK
Iceland
UK
(Midlands)
et al.
g
et al.
et al.
et al.
et al.
et al.
et al.
et al.
et al.
et al.
et al.
et al.
Authors
ebb
aylor
Honda
Fombonne
W
Arvidsson
Sponheim and
Skjeldal
T
Kadesjö
Baird
Powell
Kielinen
Bertrand
Fombonne
Magnússon and
Saemundsen
Chakrabarti et
Fombonne
g surveys by Gillber
, mild mental retardation was combined with normal IQ, whereas moderate and severe mental retardation were grouped
ear of
Y
1996
1997
1997
1997
1998
1999
1999
2000
2000
2000
2001
2001
2001
2001
publication
This number corresponds to the sample described in
This rate corresponds to the first published paper on this survey and is based on 12 subjects amongst children aged 5 to 14 yea
For the Gotebor
samples included in the 3 surveys; consequently only the last survey has been included in this table.
No.
19
20
21
22
23
24
25
26
27
28
29
30
31
32
In this study
a
b
c
d

---

Epidemiological Surveys of Autism and Other Pervasive Developmental Disorders
369
to have autism. By contrast, in a Japanese study
The assessments were conducted with various di-
(Sugiyama & Abe, 1989; study 13) where 17.3%
agnostic instruments, ranging from a classical clinical
of parents refused further investigations for their
examination to the use of batteries of standardized mea-
18-month old children who had failed a developmen-
sures. The Autism Diagnostic Interview (Le Couteur
tal check, follow-up data at age 3 suggested that half
et al., 1989) was used in 5 of the most recent surveys.
of these children still displayed developmental prob-
The precise diagnostic criteria retained to define case-
lems. Whether or not these problems were connected
ness vary according to the study and, to a large extent,
to autism is unknown, but this study points to the pos-
reflect historical changes in classification systems.
sibility of having higher rates of developmental dis-
Thus, Kanner’s criteria, Lotter’s and Rutter’s defini-
orders among non-participants to surveys. Similarly,
tions were used in the first 8 surveys (all conducted be-
in Lotter’s study (1966; study 1), 58 questionnaires
fore 1982), whereas DSM-based definitions took over
covering schools for handicapped children were
thereafter as well as ICD-10 since 1990. Some studies
returned out of the 76 forms sent out, and an inde-
have relaxed partially some diagnostic criteria such as
pendent review of the records showed that 4 of the 18
the requirement of an age of onset before 30 months
missing forms corresponded to autistic children.
(study 6) or that of the absence of schizophrenic-like
Therefore, it is difficult to draw firm conclusions from
symptoms (studies 13 and 14). However, most surveys
these different accounts. Although there is no consis-
have relied on the clinical judgment of experts to ar-
tent evidence that parental refusal to co-operate is as-
rive at the final case groupings. It is worth underlining
sociated with autism in their offspring, it appears that
that field trials for recent classifications such as DSM-
a small proportion of cases may be missed in some
III-R (Spitzer & Siegel, 1990) or DSM-IV/ICD-10
surveys as a consequence of non-cooperation at the
(Volkmar et al., 1994) have also relied upon the
screening stage. No survey included a weighting pro-
judgment of clinical experts, taken as a gold standard
cedure to compensate for non-response.
to diagnose autism. Therefore, the heterogeneity of
Only two studies (studies 1 and 30) provided an
diagnostic criteria used across surveys is somewhat mit-
estimate of the reliability of the screening procedure.
igated by reliance on expert clinical judgment to de-
The sensitivity of the screening methodology is also
termine final caseness. It is furthermore difficult to
difficult to gauge in autism surveys. The usual epi-
assess the impact of a specific diagnostic scheme or
demiological approach which consists of sampling at
of a particular diagnostic criterion on the estimate of
random screened negative subjects in order to estimate
prevalence since other powerful method factors con-
the proportion of false negatives, has not been used in
found between-studies comparisons of rates. Surpris-
these surveys for the obvious reason that, due to the
ingly, few studies have built in a reliability assessment
very low frequency of the disorder, it would be both
of the diagnostic procedure; reliability during the in-
imprecise and very costly to undertake such estima-
tensive assessment phase was high in 6 surveys (stud-
tions. The consequence of these remarks is that preva-
ies 4, 13, 16, 23, 24, and 32) and moderate in another
lence estimates must be seen as underestimates of ‘true’
one (study 14).
prevalence rates because cases are being missed due ei-
ther to lack of cooperation or to imperfect sensitivity
of the screening procedure. The magnitude of this un-
CHARACTERISTICS OF IDENTIFIED
derestimation is unknown in each survey.
SAMPLES
Similar considerations about the methodological
variability across studies apply to the intensive assess-
A total number of 2,380 subjects assessed in the
ment phases. Participation rates in these second stage
second stage of the 32 surveys were considered to suf-
assessments were not always available, either because
fer from autism, this number ranging from 6 to 427
they had simply not been calculated, or because the de-
across studies (median: 51). An assessment of intel-
sign and/or method of data collection did not lead eas-
lectual function was obtained in 20 studies. These as-
ily to their estimation. When available (studies 1, 5, 8,
sessments were conducted with various tests and
12, 13, 15, 22, 23, 25, 29, 30, 32), they were generally
instruments; furthermore, results were pooled together
high, ranging from 76.1% (study 12) to 98.6% (study
in broad bands of intellectual level which did not share
25). The source of information used to determine case-
the same boundaries across studies. As a consequence,
ness usually involved a combination of informants and
differences in rates of cognitive impairment between
data sources, with a direct assessment of the person
studies should be interpreted with caution. With these
with autism in 19 studies.
caveats in mind, some general conclusions can

---

370
Fombonne
nevertheless be reached (Table I). The median propor-
surveys, i.e., those published since 1987. We decided
tion of subjects without intellectual impairment is 30%
to further exclude 2 studies with a target population of
(range: 0%–60%).1 The corresponding figures are 30%
less than 10,000 children as estimates deriving from
(range: 6.6%–100%) for mild to moderate intellectual
smaller size studies were associated with excessively
impairments, and 40% (range: 0%–81.3%) for severe
large confidence intervals and were influencing unduly
to profound level of mental retardation. Gender repar-
the results. For the 19 remaining studies, the prevalence
tition among subjects with autism was reported in 29
estimates ranged from 2.5 to 30.8/10,000 (average 95%
studies and the male/female sex ratio varied from 1.33
CI width: 6.4), with an average rate of 11.1/10,000 and
(study 7) to 16.0 (study 4), with a mean sex ratio of
a median rate of 9.5/10,000. Similar values were ob-
4.3. Thus, no epidemiological study ever identified
tained when slightly different rules and time cutpoints
more girls than boys with autism, a finding which par-
were used. From these results, the best estimate for the
allels the gender differences found in clinically referred
current prevalence of autism is most consistent with
samples (Lord et al., 1982). Gender differences were
values lying somewhere between 9/10,000 and
more pronounced when autism was not associated with
11/10,000. For further calculations, we arbitrarily
mental retardation. In 12 studies (841 subjects) where
adopted the mid-point of this interval as the working
the sex ratio was available within the normal band of
rate for autism prevalence, i.e., the value of 10/10,000.
intellectual functioning, the median sex ratio was
5.75:1. Conversely, in 11 studies (789 subjects), the
ASSOCIATED MEDICAL CONDITIONS
median sex ratio was 1.9:1 in the group with autism
and moderate to severe mental retardation.
Rates of medical conditions associated with autism
were reported in 15 surveys and the findings are sum-
marized in Table II. It will be appreciated that these med-
PREVALENCE ESTIMATIONS
ical conditions were investigated by very different means
FOR AUTISTIC DISORDER
ranging from questionnaires to full medical work-ups.
Conditions such as congenital rubella, and PKU
Prevalence estimates ranged from 0.7/10,000 to
account for almost no cases of autism. Prior studies
72.6/10,000 (Table I). Confidence intervals were com-
suggesting an association of congenital rubella (Chess,
puted for each estimate; their width (difference between
1971) and PKU (Knobloch & Pasamanick, 1975;
the upper and lower limit of the 95% confidence inter-
Lowe et al., 1980) with autism were conducted before
val) indicates the variation in sample sizes and in the
implementation of systematic prevention measures.
precision achieved in each study (range: 0.3–115.9;
Likewise, our nil estimate of 0% for autism and neuro-
mean = 12.0). Prevalence rates were negatively cor-
fibromatosis is comparable to the 0.3% rate found in a
related with sample size (Spearman r = ?0.77;
large series of 341 referred cases (Mouridsen et al.,
p < 0.01); small-scale studies tended to report higher
1992) and, contrary to earlier claims (Gillberg &
prevalence rates.
Forsell, 1984), it does not exceed the rate expected
When surveys were combined in two groups ac-
under the assumption of independence of the two
cording to the median year of publication, the median
prevalence rate for 16 surveys published in the period
1966–1991 was 4.4/10,000, and the median rate for the
Table II. Medical Disorders Associated with Autism in Recent
Epidemiological Surveys
16 surveys published in the period 1992–2001 was
12.7/10,000. Indeed, the correlation between preva-
No. of studies
Median rate
Range
lence rate and year of publication reached statistical
significance (Spearman r = 0.70; p < 0.01); and the
Cerebral palsy
6
2.0
0–4.8
Fragile X
8
0.3
0–8.1
results of the 18 surveys with prevalence rates over
Tuberous sclerosis
10
1.2
0–3.8
7/10,000 were all published since 1987. These findings
Phenylketonuria
7
0
0–0
point towards an increase in prevalence estimates in the
Neuro?bromatosis
6
0
0–1.4
last 15 years. In order to derive a best estimate of the
Congenital rubella
10
0.3
0–5.9
current prevalence of autism, it was therefore deemed
Down Syndrome
11
1.3
0–16.7
At least one disorder
14
6.4
0–16.7
appropriate to restrict the analysis to the 21 most recent
Epilepsy
11
16.8
0–26.4
Hearing de?cits
7
1.7
0–5.9
1Study 23 which relied upon different IQ groupings has been
Visual de?cits
5
1.3
0–11.1
excluded.

---

Epidemiological Surveys of Autism and Other Pervasive Developmental Disorders
371
disorders. Similarly, bearing in mind the high rate of
surveys of autism, the attributable proportion of cases
mental retardation among samples of autistic subjects,
of autism would not exceed the 10% figure for any med-
the rates found for cerebral palsy and Down’s syndrome
ical disorder (excluding epilepsy and sensory impair-
equally suggest no particular association. The recogni-
ments). Although this figure does not incorporate other
tion that Down’s syndrome and autism co-occur in
medical events of potential etiological signi?cance, such
some individuals has been the focus of attention in re-
as encephalitis, congenital anomalies, and other rare
cent reports (Bregman & Volkmar, 1988; Ghazziudin
medical syndromes, it is similar to that reported in a re-
et al., 1992; Howlin et al., 1995); the epidemiological
cent review of the question (Rutter et al., 1994). It is
findings give further support to the validity of these
worth noting that epidemiological surveys of autism in
clinical descriptions (i.e., that the 2 conditions co-occur
very large samples (i.e., studies 15, 17, and 20) provided
in some children), although they do not suggest that the
estimates in line with our conservative summary statis-
rate of comorbidity is higher than that expected by
tics. By contrast, claims of average rates of medical con-
chance once the effects of mental retardation are taken
ditions as high as 24% appear to apply to studies of
into account. For fragile X, the low rate available in
smaller size and relying on a broadened definition of
epidemiological studies is most certainly an underesti-
autism (Gillberg & Coleman, 1996).
mate due to the fact that fragile X was not recognized
Rates of epilepsy are high among autism samples.
until relatively recently and that, in the most recent sur-
The proportion suffering from epilepsy tends also to be
veys, systematic screening for fragile X was not always
higher in those studies which have higher rates of se-
undertaken. In line with prior reports (Smalley et al.,
vere mental retardation (as in studies 16, 17, and 20).
1992), tuberous sclerosis (TS) has a consistently high
Age-specific rates for the prevalence of epilepsy were
frequency among autistic samples. Assuming a popu-
not available. The samples where high rates of epilepsy
lation prevalence of 1/10,000 for TS (Shepherd et al.,
were reported tended to have a higher median age, al-
1991; Hunt & Lindenbaum, 1984; Ahlsen et al., 1994),
though these rates seemed mostly to apply to school-
it appears that the rate of TS is about 100 times higher
aged children. Thus, in light of the increased incidence
than that expected under the hypothesis of no associa-
of seizures during adolescence among subjects with
tion. The rate of TS in autistic samples is, however,
autism (Rutter, 1970; Deykin & MacMahon, 1979), the
much lower in these epidemiological studies than the
epidemiological rates should be regarded as underesti-
9% minimum rate claimed in a recent study (Gillberg
mates of the lifetime risk of epilepsy in autism. These
et al., 1994). Whether or not the association between TS
rates are nonetheless high and support the findings of
and autism is mediated by epilepsy, localized brain
a bimodal peak of incidence of epilepsy in autistic
lesions or direct genetic effects is a matter for future re-
samples, with a first peak of incidence in the first years
search (Smalley, 1998).
of life (Volkmar & Nelson, 1990).
The overall proportion of cases of autism which
could be causally attributed to known medical disorders
RATES OF OTHER PERVASIVE
therefore remains low. From the 14 surveys where rates
DEVELOPMENTAL DISORDERS
of one of 7 clear-cut medical disorders potentially
causally associated with autism (cerebral palsy, fragile
Unspeci?ed PDDs—PDD-NOS
X, TS, PKU, neurofibromatosis, congenital rubella, and
Down’s syndrome) were available, we computed the
Several studies have provided useful information
proportion of subjects with at least one of these recog-
on rates of syndromes similar to autism but falling short
nizable disorders. Because the overlap between these
of strict diagnostic criteria (Table III). Because the
conditions is expected to be low and because the infor-
screening procedures and subsequent diagnostic as-
mation about multiply-handicapped subjects was not
sessments differed from one study to another, these
available, this overall rate was obtained by summing di-
groups of disorders are not strictly comparable across
rectly the rates for each individual condition within each
studies. In addition, as they were not the group on
study; the resulting rate might, therefore, be slightly
which the attention was focused, details are often lack-
overestimated. The fraction of cases of autism with a
ing on their phenomenological features in the available
known medical condition potentially etiologically sig-
reports. Different labels (see Table III) have been used
nificant ranged from 0% to 16.7%, with a median and
to characterize them such as the triad of impairments
mean values of 6.4% and 6.0% respectively. Even if
involving impairments in reciprocal social interaction,
some adjustment was made to account for the underes-
communication and imagination (Wing & Gould,
timation of the rate of fragile X in epidemiological
1979). These groups would be overlapping with current

---

372
Fombonne
psychotic”
s
“borderline
ger Disorder
or
Asper
s
s
psychoses”
“other
).
Case de?nition for other PDD’
with
autistic children
like” symptoms, not meeting DSM-III criteria for IA,
COPDD or atypical PDD.
of “infantile psychosis” than autism, or a broadened
de?nition of DSM III.
Children with some behaviour similar to
Children
Socially impaired (triad of impairments)
Autistic mental retardation
Children referred by professionals with “autistic-
Children meeting criteria for other forms
Children with mixed developmental disorders
Children with mixed developmental disorders
Children with other PDD’
Children with other PDD’
Children with PDD-NOS and
Children with PDD-NOS
c
0.90
0.44
3.33
1.25
2.39
1.04
1.43
2.05
0.9
1.7
0.7
2.15
Rumeau-Rouquette et al. 1994
Prevalence
rate ratio
s
a
other PDD’
7.8
6.2
5.25
1.05
9.2
1.2
+
21.2
1
1
16.3
57.9
20.8
67.4
52.9
>
Combined rate
autism
a
Informative Studies on Rates of Nonautism Pervasive Developmental Disorders
3.3
1.9
2.92
7.79
4.7
6.6
16.3
>
10.94
27.1
13.0
27.0
36.1
Prevalence rate
of other PDD
able III.
T
4.1
4.3
4.9
2.33
3.26
4.5
4.6
5.3
7.8
30.8
40.5
16.8
Rates of
Autism
b
.
(1997)
(1982)
(2001)
(2000)
(1976)
(1987)
et al.
(2000)
et al.
et al.
et al.
et al.
et al.
et al.
ing
Mamelle (1989)
du Mazaubrun (1992)
Chakrabarti and
Fombonne (2001)
1. Lotter (1966)
2. Brask (1970)
4. W
5. Hoshino
9. Burd
Computed by the author
These rates are derived from the complete results of the survey of 3 birth cohorts of French children (
No. Study
14. Cialdella and
17. Fombonne and
20. Fombonne
26. Baird
27. Powell
29. Bertrand
32.
Other PDD rate divided by autism rate.
a
b
c

---

Epidemiological Surveys of Autism and Other Pervasive Developmental Disorders
373
diagnostic labels such as atypical autism and PDDNOS,
but a ratio of 4 to 1 would appear an acceptable, albeit
which does little to understand their relationship with
conservative, conclusion based on this limited avail-
a narrower definition of autism. Twelve of the 32 sur-
able evidence. This translates into a rate for AS which
veys yielded estimates of the prevalence of these de-
would be one fourth that of autism. We therefore used
velopmental disorders, with 8 studies showing higher
for subsequent calculations an estimate of 2.5/10,000
rates for the non-autism disorders than the rates for
for AS.
autism. The ratio of the rate of non autistic PDD to the
Few surveys have provided data on childhood dis-
rate of autism varied between from 0.44 to 3.33 (Table
integrative disorder (CDD), also known as Heller syn-
III) with a mean value of 1.5, which translates into an
drome, disintegrative psychosis (ICD-9) or late-onset
average prevalence estimate of 15/10,000. In other
autism (see Volkmar, 1992). In 4 studies (studies 9, 23,
words, for 2 children with autism assessed in epidemi-
31, 32), prevalence estimates ranged from 11.1 to 64.5
ological surveys, 3 children were found to have severe
per million (Fombonne, 2002). Most of the upper lim-
impairments of a similar nature but falling short of strict
its of the 95% confidence intervals were consistent with
diagnostic criteria for autism. This group has been
an upper bound limit of 1/10,000, and the pooled esti-
much less studied in epidemiological studies but it
mate was 1.7/100,000. As cases of CDD were both rare
should be clear from these figures that they represent
and already included in the numerator alongside autism
a very substantial group of children whose treatment
cases in most surveys, we do not provide separate es-
needs are likely to be as important as those of children
timates of the numbers of subjects suffering from CDD
with autism.
in subsequent calculations.
Asperger Syndrome and Childhood
PREVALENCE FOR COMBINED PDDs
Disintegrative Disorder
The reader is referred to recent epidemiological
Taking the aforementioned conservative estimates,
reviews for these two conditions (Fombonne, 2001a;
the prevalence for all PDDs is at least 27.5/10,000 (i.e.,
Fombonne, 2002). In brief, epidemiological studies of
the sum of estimates for autism (10/10,000), PDDNOS
Asperger syndrome (AS) are sparse, probably due to
(15/10,000), and AS (2.5/10,000)). This global estimate
the fact that it was acknowledged as a separate diag-
is derived from a conservative analysis of existing
nostic category only recently in both ICD-10 and
data.
DSM-IV. Only two epidemiological surveys have been
However, 3 recent epidemiological surveys
conducted which specifically investigated its preva-
yielded rates about twice as high (Table IV). The com-
lence (Ehlers & Gillberg, 1993; Kadesjö et al., 1999).
mon features in the design of these epidemiological
However, only a handful (N < 5) of cases were identi-
enquiries are worth noting. First, the case definition
fied in these surveys, with the resulting estimates of 28
chosen for these investigations was that of a pervasive
and 48/10,000 being extremely imprecise. By contrast,
developmental disorder (PDD) as opposed to a nar-
other recent autism surveys have consistently identi-
rower approach focusing on autistic disorder. Investi-
fied smaller numbers of children with AS than those
gators were concerned with any combination of severe
with autism within the same survey. In 6 such surveys
developmental abnormalities occurring in one or more
(studies 23–27, 32), the ratio of autism to AS rates in
of the 3 symptomatic domains defining PDD and
each survey was above unity, suggesting that the rate
autism. Second, case finding techniques employed in
of AS was consistently lower than that for autism. How
these surveys were proactive, relying on multiple and
much lower is difficult to establish from existing data,
repeated screening phases, involving both different
Table IV. Newer Epidemiological Surveys of PDDs
Autism
PDDNOS + AS
All PDDs
Rate/
M/F
% IQ
Rate/
M/F
% IQ
Rate/
Age
10,000
ratio
normal
10,000
ratio
normal
10,000
Bertrand et al., 2001
3–10
40.5
2.2
37
27.0
3.7
51
67.5
Baird et al., 2000
7
30.8
15.7
60
27.1
4.5
—
57.9
Chakrabarti and Fombonne 2001
4–7
16.8
3.3
29
44.5
4.3
94
61.3

---

374
Fombonne
informants at each phase and surveying the same co-
whilst reviewing the evidence for a secular increase in
horts at different ages, which certainly maximized the
rates of PDDs.
sensitivity of case identification. Third, assessment
Five approaches to assess this question have been
were performed with standardized diagnostic measures
used in the literature.
(i.e., ADI-R and ADOS) which match well the dimen-
sional approach retained for case definition. Finally,
Referral Statistics
these samples comprised young children around their
fifth birthday, thereby optimizing sensitivity of case
Increasing numbers of children referred to spe-
finding procedures. Furthermore, the size of targeted
cialist services or known to special education registers
populations was reasonably small (between 9,000 and
have been taken as evidence for an increased incidence
16,000), probably allowing for the most efficient use
of autism-spectrum disorders. However, trends over
of research resources. Conducted in different regions
time in referred samples are confounded by many fac-
and countries by different teams, the convergence of
tors such as referral patterns, availability of services,
estimates (Table IV: right-hand column) is striking. Two
heightened public awareness, decreasing age at diag-
further results are worth noting. First, in sharp contrast
nosis and changes over time in diagnostic concepts and
with the prevalence for combined PDDs, the separate
practices, to name only a few. Failure to control for
estimates for autistic disorder and PDDNOS vary
these confounding factors is obvious in some recent re-
widely across studies, as if the reliability of the differ-
ports, such as the widely quoted report from Califor-
entiation between autistic disorder and PDDNOS was
nia educational services (Department of Developmental
mediocre at that young age, despite the use of up-to-
Services, 1999; Fombonne, 2001b). First, these reports
date standardized measures. Second, the rate of men-
apply to numbers rather than rates, and failure to relate
tal retardation was, overall, much lower than in
these numbers to meaningful denominators leave these
previous surveys of autism. While this should not be a
figures vulnerable to changes in the composition of
surprise for children in the PDDNOS/AS groups, this
the underlying population. Second, no attempt was made
trend was also noticeable within the samples diagnosed
to adjust the trends for changes in diagnostic concepts
with autistic disorder. To what extent this trend reflects
and definitions. However, major nosographical modi-
the previously mentioned differential classification is-
fications were introduced during the corresponding
sues between autism and PDDNOS or a genuine trend
years with a general tendency in most classifications to
over time towards decreased rate of mental retardation
broaden the concept of autism (as embodied in the
within children with autistic disorder (possibly as a
terms ‘autism spectrum’ or ‘pervasive developmental
result as earlier diagnosis and intervention) remains to
disorder’). Third, age characteristics of the subjects
be established.
recorded in official statistics were portrayed in a con-
fusing manner where the preponderance of young sub-
jects was presented as evidence of increasing rates in
TIME TRENDS
successive birth cohorts. The problems associated with
disentangling age from period and cohort effects in such
The debate on the hypothesis of a secular increase
observational data are well known in the epidemiolog-
in rates of autism has been obscured by a lack of clar-
ical literature and deserve a better statistical handling.
ity in the measures of disease occurrence used by
Fourth, the decreasing age at diagnosis leads in itself
investigators, or rather in their interpretation. In
to increasing numbers of young children being identi-
particular, it is crucial to differentiate prevalence (the
fied in official statistics or referred to already busy spe-
proportion of individuals in a population who suffer
cialist services. Earlier identification of children from
from a defined disorder) from incidence (the number
the prevalence pool may result in increased service ac-
of new cases occurring in a population over a period
tivity; however, it does not mean increased incidence.
of time). Prevalence is useful to estimate needs and plan
Fifth, upward trends were also reported for other dis-
services; only incidence rates can be used for causal re-
orders (such as cerebral palsy, epilepsy or mental re-
search. Both prevalence and incidence estimates will
tardation), therefore casting doubts on the specificity
be inflated when case definition is broadened and case
of the trends for autism. Evidence from these referral
ascertainment is improved. Time trends in rates can
statistics is therefore very weak (Fombonne, 2001b).
therefore only be gauged in investigations which hold
Accordingly, proper epidemiological studies are needed
these parameters under strict control over time. These
in order to assess secular changes in the incidence of a
methodological requirements must be borne in mind
disorder.

---