Systematic Review of Prevalence Studies of Autism Spectrum Disorders

8
Arch Dis Child: first published as 10.1136/adc.2004.062083 on 29 April 2005. Downloaded from http://adc.bmj.com/ on October 10, 2021 at UK NHS and HE Athens Access. Protected by
ORIGINAL ARTICLE
Systematic review of prevalence studies of autism

spectrum disorders
J G Williams, J P T Higgins, C E G Brayne
...............................................................................................................................
Arch Dis Child 2006;91:8–15. doi: 10.1136/adc.2004.062083
Aim: To quantitatively examine the influence of study methodology and population characteristics on
prevalence estimates of autism spectrum disorders.
Methods: Electronic databases and bibliographies were searched and identified papers evaluated against
See end of article for inclusion criteria. Two groups of studies estimated the prevalence of typical autism and all autism spectrum
authors’ affiliations disorders (ASD). The extent of variation among studies and overall prevalence were estimated using meta-
.......................
analysis. The influence of methodological factors and population characteristics on estimated prevalence
Correspondence to: was investigated using meta-regression and summarised as odds ratios (OR).
Dr J G Williams, Results: Forty studies met inclusion criteria, of which 37 estimated the prevalence of typical autism, and 23
Department of Public
Health and Primary Care, the prevalence of all ASD. A high degree of heterogeneity among studies was observed. The overall
University of Cambridge, random effects estimate of prevalence across studies of typical autism was 7.1 per 10 000 (95% CI 1.6 to
Forvie Site, Robinson 30.6) and of all ASD was 20.0 per 10 000 (95% CI 4.9 to 82.1). Diagnostic criteria used (ICD-10 or
Way, Cambridge CB2
2SR, UK; j.g.williams.97@ DSM-IV versus other; OR = 3.36, 95% CI 2.07 to 5.46), age of the children screened (OR = 0.91 per year,
cantab.net 95% CI 0.83 to 0.99), and study location (e.g. Japan versus North America; OR = 3.60, 95% CI 1.73 to
7.46) were all significantly associated with prevalence of typical autism. Diagnostic criteria, age of the
Accepted sample, and urban or rural location were associated with estimated prevalence of all ASD.
22 December 2004
Published Online First
Conclusions: Sixty one per cent of the variation in prevalence estimates of typical autism was explained by
29 April 2005 these models. Diagnostic criteria used, age of children screened, and study location may be acting as
....................... proxies for other study characteristics and require further investigation.
copyright.
T
he prevalence of autistic disorder is now considered to be Study selection
around 10 per 10 000, and the prevalence of pervasive Identified papers were examined against criteria for inclusion
developmental disorders, 27.5 per 10 000. These are (box 2). The paper itself was examined if the abstract was
derived from studies which have estimated prevalences of
autistic disorder ranging from 0.7 to 72.6 per 10 000.1 An
increase in prevalence estimates has been observed over time, Box 1: Search strategy for identifying
the reasons for which are not clear and may include: changes prevalence studies
in study methodology; a genuine rise in autism risk factors;
increase in services available, including diagnostic; increased MEDLINE (PubMed) (searched 13/04/04)
awareness among educational and clinical professionals; and Years (1966–2004):
growing acceptance that autism can coexist with a range of (‘‘Autistic-Disorder’’/all subheadings [MeSH] OR
other conditions.1–4 ‘‘Asperger-Syndrome’’/all subheadings [MeSH] OR
True variation in prevalence could generate aetiological ‘‘Schizophrenia-Childhood’’/all subheadings [MeSH] and
hypotheses for autism and it is vital to understand what (PY = 1966–1970) OR autis* (free text term)) AND
underpins the variation. Accurate estimates of the true (‘‘Prevalence-‘‘/all subheadings [MeSH] OR ‘‘Cross-
prevalence are of value in planning diagnostic and interven- Sectional-Studies’’/all subheadings [MeSH] OR ‘‘Mass-
tion services. Screening’’/all subheadings [MeSH] OR ‘‘Multiphasic-
Several narrative reviews have been conducted. This paper Screening’’/all subheadings [MeSH]).
uses systematic and quantitative methods to examine reasons
EMBASE (Excerpta Medica Database) (searched 13/04/04)
for variation in prevalence estimates. The aims are to assess
(BIDS EMBASE, via Ovid, copyright 2003)
the degree of variation among prevalence studies of autism,
Years (1980–2004):
and to provide an overall summary of prevalence diversity
(exp` autism/ OR exp infantile autism/ OR exp Asperger
taking into account among-study variance using meta-
syndrome/ OR autism.mp1 (as keyword) OR Asperger.mp
analysis. Aspects of study methodology and population
(as keyword)) AND (exp prevalence/ OR exp mass screen-
characteristics are then examined using meta-regression to
ing/ OR exp screening/ OR cross-sectional.mp (as key-
investigate their influence on prevalence estimates. word)) NOT (genetic screening/exp OR genetic screen.mp
(as keyword).
METHODS
Literature searches MeSH (Medical Subjects Headings), The National Library of
Two databases, MEDLINE and EMBASE, were systematically Medicine controlled vocabulary for indexing articles in
searched by the first author (box 1). In addition, bibliogra- PubMed; `exp, explode term (search under all subheadings);
phies of previous reviews1 5 6 were examined to identify 1mp, uses the database thesaurus search term.
published prevalence studies.
www.archdischild.com
Prevalence studies of autism spectrum disorders 9
insufficiently clear. Where there was more than one paper for the true prevalence was calculated as the mean of
published on a particular study, the most recent was included logits ¡ 1.96 t, where t is the among-study standard
in the review. deviation.11
Data extraction Investigation of sources of heterogeneity

Methods and population characteristics reported across most The potential influence of covariates on the prevalence
studies were selected for data extraction. The first author estimates was investigated using a random effects regression
extracted and coded the data. In studies using different model, thus taking account of among-study variance, using
diagnostic criteria, prevalence data based on the more the metareg command in STATA.12 The regression coefficients
recently published diagnostic criteria were extracted. The represent log odds ratios, which are presented as odds ratios
studies formed two groups: those that assessed the pre- with 95% confidence intervals.
valence of classic autism, or autistic disorder, known here as A multivariate meta-regression model was constructed to
‘‘typical autism’’; and those that assessed the prevalence of investigate which covariates were associated with prevalence
autism spectrum disorders (ASD) or all pervasive develop- estimates if there was adjustment for other study covariates.
mental disorders, known here as ‘‘all ASD’’. Assessments of The fit of each model was assessed using the percentage of
risk of bias included reporting of refusal rates and the among-study variance explained ((12(t2 in model/t2 in
reliability of screen and assessment procedures. model with no covariates))6100), together with a signifi-
cance test for each introduced variable (T = coefficient/SE,
Analysis related to the t-distribution). The models were constructed
In the basic tables, crude prevalence estimates (number of using a forward stepwise procedure as described in the
cases/sample size) were presented, along with standard results section. For each model a maximum number of
errors. For all meta-analyses and meta-regressions, preva- covariates was set at n/10 where n was the number of studies,
following standard recommendations for model size relative
lence estimates were transformed to logits to
to sample size.13
improve their statistical properties. These were later back-
transformed to prevalences and expressed as cases per 10 000
RESULTS
people.
Studies identified
Literature searches identified 670 papers (including dupli-
Description of heterogeneity among studies and cates). After exclusion through comparison of titles and
summary of prevalence abstracts against inclusion criteria, 77 papers were identified
Forest plots8 were used to visualise the extent of hetero- for detailed examination. Thirty seven papers were excluded,
geneity among studies. Two statistical methods were used to including ten on the basis of inclusion criterion 2 (box 2), ten
quantify the variation. A standard test for heterogeneity on the basis of criterion 4, nine on the basis of 6, and one on
copyright.
examined the null hypothesis that the true prevalences are the basis of 7. In addition, four papers did not have detailed
identical in every study. Since heterogeneity was expected a English summaries, one was not peer reviewed, and two were
priori, this was supplemented with a measure of the degree of untraceable. Of these seven potentially eligible studies, four
inconsistency across studies, I2.9 I2 describes the proportion of were conducted in Japan, one in the USA, one in France, and
variation in prevalence estimates that is due to genuine one in Sweden.
variation in prevalences rather than sampling error. It is Forty papers met inclusion criteria, of which 37 gave
expressed as a percentage, with 0% indicating consistency. estimates for typical autism and 23 for all ASD (table 1). The
The random effects model assumes the study prevalences study sample sizes ranged from 826 to 4 590 333 (med-
follow a normal distribution, allowing for among-study ian = 48 705). Only 17 (40%) studies reported the refusal rate
variation.10 The usual confidence interval for the mean in at the screen phase of the study, and 13 (33%) at the
the random effects model does not take among-study assessment stage. Six (15%) studies reported investigating
variance into account, so is deceptively narrow when there the reliability of their screen method, and 11 (26%) studies
is substantial variation across studies. Instead, a 95% interval stated that the inter-rater reliability for the diagnostic
assessment had been investigated. Many studies did not
report refusal rates and reliability, so these covariates could
Box 2: Inclusion criteria not be included in further analyses.
(1) Primary research Description of heterogeneity among studies and

(2) A geographically and temporally defined population summary of prevalence
(3) Cross-sectional study or data, or first phase of a There was clearly wide variation in the prevalence estimates
longitudinal study of typical autism (fig 1) and an increase in prevalence
(4) Defined diagnostic criteria stated for autism or autism estimates over time. The Q statistic was very large
spectrum disorder (Q = 1947.6, df = 36, p , 0.001; I2 = 98.2%), showing that
there was a great deal of variation among the studies. There
(5) Includes individuals under 18 years old
was also a high degree of heterogeneity among estimates of
(6) Initial selection in a wide range of children in the all ASD (fig 2) (Q = 1577.7, df = 22, p , 0.001; I2 = 98.6%).
general population, or in a clinical setting The back-transformed mean of the random effects dis-
(7) Final identification of cases based on clinical or other tribution for studies of typical autism was 7.1 per 10 000
diagnostic assessment of selected children (95% interval for true prevalences: 1.6 to 30.6) and for studies
(8) Published in English, or with detailed summaries in of all ASD was 20.0 per 10 000 (95% interval for true
English prevalences: 4.9 to 82.1).
(9) Peer reviewed paper or conference presentation
(10) Includes prevalence data Investigation of sources of heterogeneity
(criteria adapted from Wing7) Two studies were excluded from the regression analyses:
either the published paper was not available,16 or insufficient
10
Table 1 Summary of prevalence studies of autism spectrum disorders (42 studies)

Typical autism All ASD
Prospective(P)/ Prevalence
Publ. Sample Screen retrospective (R) Diagnostic estimate (per Diagnostic Prevalence estimate
Ref. year First author Country Sample size Area Age (years) screened1 methods assessment criteria used 10 000) (SE) criteria used (per 10 000) (SE)
14 1966 Lotter UK 78 000 Urban 8–10 P Q P Kanner 4.5 (0.76) – –
15 1970 Treffert USA 899 750 Mixed 0–12 C R R Kanner 0.7 (0.09) – –
16 1970 Brask Denmark 46 500 Urban 2–14 S – – Kanner 4.3 (0.96) – –
17 1979 Wing UK 35 000 Urban 0–15 C R P Kanner 4.9 (1.18) Triad 21.2 (2.46)
18 1982 Hoshino Japan 609 848 Mixed 0–18 P L P Kanner 2.33 (0.20) – –
19 1983 Ishii Japan 34 987 Urban 6–12 P L P Rutter 16.0 (2.14) – –
20 1983 Bohman Sweden 69 000 Mixed 0–20 P L P Rutter 3.0 (0.66) Rutter 5.6 (0.9)
21 1984 McCarthy Ireland 65 000 Mixed 8–10 S R R Kanner 4.3 (0.81) – –
22 1984 Gillberg Sweden 128 584 Mixed 4–18 P L P Rutter 2.0 (0.39) Rutter 3.9 (0.55)
23 1986 Steinhausen Germany 279 616 Urban 0–15 S L R Rutter 1.9 (0.26) – –
24 1986 Steffenburg Sweden 42 886 Mixed 0–9 P L P DSM-III 4.7 (1.05) DSM-III 6.9 (1.27)
25 1987 Burd USA 180 986 Mixed 2–18 C L P DSM-III 1.2 (0.26) DSM-III 3.3 (0.43)
26 1987 Matsuishi Japan 32 834 Urban 4–12 P R P DSM-III 15.5 (2.17) – –
27 1988 Tanoue Japan 95 394 Rural 7 P C P DSM-III 13.8 (1.20) – –
28 1988 Bryson Canada 20 800 Mixed 6–14 P Q P DSM-III 1.9 (0.96) Denkla Triad 10.1 (2.2)
29 1989 Ritvo USA 526 514 Mixed 3–17 P L P DSM-III 2.92 (0.24) – –
30 1989 Sugiyama Japan 11 320 Urban 1.5 P C P DSM-III 13.0 (3.37) – –
31 1989 Cialdella France 135 180 Urban 5–9 C L R DSM-III 5.1 (0.61) DSM-III 10.8 (0.89)
32 1991 Gillberg Sweden 78 102 Mixed 4–13 C L P DSM-III-R 7.0 (0.95) DSM-III-R 9.4 (1.1)
33 1992 Ohtaki Japan 35 366 Mixed 6–14 S R P DSM-III-R 13.9 (1.98) – –
34 1992 Fombonne France 274 816 Mixed 9 & 13 C R R ICD-9 4.9 (0.42) – –
35 1993 Herder Norway 50 909* – 1–17 C – – DSM-III-R 5.5 (1.04) – –
36 1996 Honda Japan 8 537 Urban 5 P C P ICD-10 21.1 (4.97) – –
37 1997 Arvidsson Sweden 1 941 Mixed 3–6 P C P ICD-10 31 (12.6) ICD-10 46 (15.36)
38 1997 Webb UK 73 300 Mixed 3–15 P L P DSM-III-R 7.2 (0.98) – –
39 1997 Fombonne France 325 347 Mixed 6–16 C R R ICD-10 5.35 (0.41) ICD-10 16.29 (0.71)
40 1998 Sponheim Norway 65 688 Mixed 3–14 P L P ICD-10 3.8 (0.76) ICD-10 5.2 (0.89)
41 1999 Kadesjo Sweden 826 Urban 6–7 P L P ICD-10 60 (26.87) Gillberg’s criteria 121 (38.04)
42 2000 Powell UK 16 049* Mixed 0 to ,5 S R R DSM-III-R or 16.2 (3.17) ICD-10 33.7 (1.48)
ICD-10**
43 2000 Kielinen Finland 152 732 Mixed 3–18 C R R ICD-10 5.6 (0.61) ICD-10 13.9 (0.95)
44 2000 Baird UK 16 235 Urban 1.5 P Q (CHAT) P ICD-10 30.8 (4.35) ICD-10 57.9 (5.95)
45 2001 Magnusson Iceland 43 153 Mixed 5–14` P C R ICD-10 8.6 (1.41) ICD-10 13.2 (1.75)
46 2001 Bertrand USA 8 896 Urban 3–10 P R P (where possible) DSM-IV 40.0 (6.69) DSM-IV 67 (8.65)
47 2001 Chakrabarti UK 15 500 Mixed 2.5–6.5 P C P DSM-IV 16.8 (3.29) DSM-IV 62.6 (6.34)
48 2002 Croen USA 4 590 333 Mixed 0–12 S R R DSM-III & 11.0 (0.15) – –
DSM-IV
49 2003 Lingam UK 186 206 Mixed 5–14 S R R ICD-10 14.9 (0.89) ICD-10 30.4 (1.28)
50 2004 Tebruegge UK 2 536 Mixed 8–9 P R R ICD-10 23.7 (9.7) ICD-10 82.8 (18.0)
ASD papers
51 2001 Fombonne UK 10 438 Mixed 5–15 P Q (DAWBA) P – – DSM-IV and ICD- 26.1 (5.0)
10 (excl Rett’s)
52 2002 Scott UK 43 472 Mixed 5–11 P L R – – ICD-10 57.0 (3.61)
53 2003 Yeargin-Allsopp USA 289 456 Urban 3–10 C R R – – DSM-IV 34 (1.08)
*Estimated from number of cases and prevalence estimate; Cohort first screened at 18 months, followed up until age 7–8 years; `Data from an older cohort also included in the study, but excluded from these analyses; 1Sample screened: P,
whole population; C, general clinical services; S, specialist clinical/educational services; Screen methods: C, routine checks; L, letter to elicit referrals; Q, questionnaires/interview; R, records (CHAT, Checklist for Autism in Toddlers; DAWBA,
Development and Well-Being Assessment); **ICD-10 used for analysis; DSM-IV used for analysis.
Williams, Higgins, Brayne
copyright.
Study Study
Lotter, 1966 Wing, 1979

Treffert, 1970 Bohman, 1983
Brask, 1970 Gillberg, 1984
Wing, 1979 Steffenburg, 1986
Hoshino, 1982 Burd, 1987
Ishii, 1983 Bryson, 1988
Bohman, 1983 Cialdella, 1989
McCarthy, 1984 Gillberg, 1991
Gillberg, 1984 Arvidsson, 1997
Steinhausen, 1986 Fombonne, 1997
Steffenburg, 1986 Sponheim, 1998
Burd, 1987 Kadesjo, 1999
Matsuishi, 1987 Powell, 2000
Tanoue, 1988 Kielinen, 2000
Bryson, 1988 Baird, 2000
Ritvo, 1989 Magnusson, 2001
Sugiyama, 1989 Chakrabarti, 2001
Ciadella, 1989 Bertrand, 2001
Gillberg, 1991 Fombonne, 2001
Ohtaki, 1992 Scott, 2002
Fombonne, 1992 Lingam, 2003
Herder, 1993 Yeargin-Allsopp, 2003
Honda, 1996 Tebruegge, 2004
Arvidsson, 1997
Webb, 1997
Fombonne, 1997 0.5 1 2.5 5 10 25 50 100
Sponheim, 1998 per 10 000 (log scale)
Kadesjo, 1999
Powell, 2000 Figure 2 Forest plot of prevalence estimates and 95% confidence
Kielinen, 2000 intervals from studies of all ASD, log transformed (n = 23).
Baird, 2000
Magnusson, 2001
study was on a population or clinic based sample, were not
Bertrand, 2001
Chakrabarti, 2001 significantly associated with prevalence. Urban location gave
copyright.
Croen, 2002 rise to higher prevalence estimates than studies carried out in
Lingam, 2003 rural or mixed locations (OR = 1.90, 95% CI 1.10 to 3.25;
Tebruegge, 2004 among-study variance explained = 53%). Studies that drew
on records of previous diagnostic assessments resulted in
lower prevalence estimates than those which included a
0.5 1 2.5 5 10 25 50 100 prospective diagnostic assessment (OR = 0.57, 95% CI 0.33 to
per 10 000 (log scale) 0.96; variance explained = 53%). Including region of study
provided the model that explained the most among-study
Figure 1 Forest plot of prevalence estimates and 95% confidence variance (variance explained = 61%) (table 3). In this final
intervals from studies of typical autism, log transformed (n = 37). model, using ICD-10 or DSM-IV led to prevalence estimates
three times those using other diagnostic criteria. The odds
information on study methodology was included in the ratio for age was 0.91 (95% CI 0.83 to 0.99), showing that an
English abstract of a Swedish paper.35 increase of one year in the age of the children screened led to
a significant reduction in prevalence estimates. For example,
Studies of typical autism when the odds ratio is taken to approximate a relative risk, if
The associations between study covariates and prevalence prevalence was estimated to be 10 per 10 000 in a sample of 5
estimates of typical autism from univariate meta-regression year olds, it would be expected to be around 9.1 per 10 000 in
analyses are shown in table 2. Taking account of the age of a sample of 6 year olds. Studies in Japan gave rise to
the children, for example, explained 23% of the among- prevalence estimates that were 3.6 times those in North
studies variance. America.
Diagnostic criteria and decade of publication were the
covariates that explained the most variance among studies in Studies of all ASD
the univariate analyses. These two covariates are collinear The associations between study covariates and prevalence
and it was not possible to include both in a multivariate estimates of all ASD from univariate meta-regression
model. The diagnostic criteria used were entered first into the analyses are shown in table 4. Only three covariates were
multivariate model since this was considered to be more significantly associated with the prevalence estimates: age of
directly related to variation in prevalence estimates than the children screened, urban or rural study location, and the
decade of publication, which is a proxy for all time varying diagnostic criteria used. The screen method used was of
covariates. The binary categorisation of diagnostic criteria borderline significance. Of these, diagnostic criteria explained
was used, as it was not possible to use multiple categories of most among-study variance, and was therefore included in
diagnostic criteria in a multivariate analysis with so few the multivariate analyses. Each of the other covariates was
studies. Age of the children screened also explained much introduced into the model in turn to form models with two
among-study variance, and was entered next into the model. covariates. As in the analyses of studies of typical autism, as
Models with three covariates were constructed which decade and diagnostic criteria were collinear, only the
included age, diagnostic criteria, and each remaining covariate for diagnostic criteria was included in further
covariate in turn. Screening method used, and whether the analyses. When adjusting for diagnostic criteria, the only
12 Williams, Higgins, Brayne
Table 2 Results of meta-regression of studies of typical autism, univariate analyses (n = 35)
Categories of covariate Variance
Covariate (first listed used as baseline) No. studies Odds ratio 95% CI (odds ratio) p value t2 explained (%)
No covariates 35 – – – 0.98 –
Age Mid-point of age range 35 0.84 0.75 to 0.93 0.002 0.75 23
(continuous variable)
Decade 1960s and 1970s 3 1.00 – – 0.58 41
1980s 14 1.80 0.68 to 4.81 0.24
1990s 9 4.26 1.52 to 11.94 0.006
2000s 9 6.42 2.29 to 17.99 ,0.001
Region* N. America 6 1.00 – – 0.85 13
Japan 7 3.19 1.14 to 8.94 0.03
Europe and Scandinavia 22 2.05 0.87 to 4.85 0.10
Area Rural/mixed 24 1.00 – – 0.86 12
Urban 11 2.10 1.07 to 4.14 0.03
Sample screened Population based 21 1.00 – – 0.88 10
Clinic based 14 0.54 0.28 to 1.03 0.06
Screen method Routine checks 6 1.00 – – 0.77 21
Letter for referrals 13 0.28 0.12 to 0.68 0.005
Questionnaires 3 0.45 0.13 to 1.62 0.22
Records 13 0.53 0.22 to 1.27 0.16
Assessment Prospective 23 1.00 – – 0.95 3
Retrospective 12 0.73 0.37 to 1.46 0.38
Diagnostic criterion 1 Not ICD-10 or DSM-IV 21 1.00 – – 0.64 35
ICD-10 or DSM-IV 14 3.32 1.89 to 5.81 ,0.001
Diagnostic criterion 2 Kanner 5 1.00 – – 0.54 45
Rutter 4 1.38 0.51 to 3.71 0.53
DSM-III 8 1.95 0.84 to 4.57 0.12
DSM-III-R 3 3.29 1.13 to 9.58 0.03
ICD-9 1 1.82 0.37 to 8.85 0.46
ICD-10 11 5.10 2.29 to 11.47 ,0.001
DSM-IV 3 7.17 2.46 to 20.91 ,0.001
*Region = North America: USA and Canada; Japan; Europe (UK, France, Germany, Ireland) and Scandinavia (Denmark, Sweden, Norway, Finland, Iceland).
Sample screened = whole population versus clinic based (general clinical services and clinic specialist services).
Table 3 Multivariate meta-regression results, for studies of typical autism (n = 35)*
copyright.
Covariate Odds ratio 95% CI (odds ratio) p value
Diagnostic criteria Not ICD-10 or DSM-IV 1.00 – –

ICD-10 or DSM-IV 3.36 2.07 to 5.46 ,0.001
Age (years) 0.91 0.83 to 0.99 0.03
Region North America 1.00 – –
Japan 3.60 1.73 to 7.46 0.001
Europe and Scandinavia 1.67 0.92 to 3.02 0.09
Intercept 5.1561024 2.0661024 to 1.2961023 ,0.001
*As an example, to estimate the prevalence of typical autism in 8 year old European children using ICD-10, take
10 0006(x/(1+x)), where x = 5.156102463.3660.91861.67, that is, a prevalence of 13.6 per 10 000.
covariates that were significantly associated with the and 20.0 per 10 000 for all ASD are slightly lower than those
prevalence estimates were the age of the children screened estimated previously at 8.7–10.0 per 10 000 and 27.5 per
(variance explained = 50%) and urban or rural study location 10 000 respectively.1 3
(variance explained = 53%). Both these models are presented The covariate most strongly associated with prevalence
(table 5). Using ICD-10 or DSM-IV gave rise to prevalence estimates for typical autism and all ASD was the diagnostic
estimates that were over twice those in studies using other criteria used. This association has been recognised pre-
diagnostic criteria. When including age in the model, an viously.2–4 The time variation in prevalence is so closely
increase in the age of the sample by one year was associated linked to changes in diagnostic criteria, the two could not be
with a fall in prevalence by a factor of approximately 0.85, examined separately. Furthermore, it was not possible to
taking the odds ratio as an approximation of a relative risk. account entirely for the effect of the diagnostic criteria on the
Alternatively, when including study location, studies in prevalence estimates as the ICD-10 and DSM-IV diagnostic
urban areas gave rise to prevalence estimates over 2.5 times schema leave some scope for variation in their interpretation
those in rural or mixed urban and rural areas. and application.
The age of the children screened was strongly associated
DISCUSSION with the prevalence estimates. Manifestations of ASD may be
Main findings more obvious in younger children. Alternatively, some
As expected, a large amount of variation in prevalence across screening methods may be more sensitive for younger
studies was found by graphical representation of estimates children. Methods of screening were found to be significantly
and by indices of heterogeneity. Despite this wide variation, associated with the prevalence estimates in the univariate
pooled estimates are useful to indicate the public health analyses of typical autism, but not after adjusting for the age
burden of the disorder. The study variation is reflected in the of the children screened.
very large intervals on the summaries of overall prevalence. The multivariate model that explained most among-study
The estimates of around 7.1 per 10 000 for typical autism, variance in studies of typical autism included the region
Table 4 Results of meta-regression of studies of all autism spectrum disorders, univariate analyses (n = 23)
Categories of covariate Variance
Covariate* (first listed used as baseline) No. studies Odds ratio 95% CI (odds ratio) p value t2 explained (%)
No covariates 23 1.01
Age Mid-point of age range 23 0.82 0.72 to 0.94 0.005 0.74 27
(continuous variable)
Decade 1960s and 1970s 1 1.00 – – 0.43 57
1980s 6 0.29 0.07 to 1.19 0.09
1990s 5 0.93 0.22 to 3.94 0.92
2000s 11 1.75 0.44 to 6.89 0.42
Region N. America 4 1.00 – – 0.77 24
Japan 0 – – – – –
Europe 10 1.99 0.70 to 5.58 0.19
Scandinavia 9 0.72 0.25 to 2.05 0.54
Area Rural/mixed 17 1.00 – – 0.86 15
Urban 6 2.44 1.02 to 5.81 0.05
Sample screened Population based 14 1.00 – – 0.97 4
Clinic based 9 0.66 0.29 to 1.54 0.34
Screen method Routine checks 3 1.00 – – 0.72 29
Letter for referrals 9 0.31 0.10 to 0.98 0.05
Questionnaires 3 0.76 0.19 to 3.03 0.69
Records 8 0.93 0.30 to 2.97 0.91
Assessment Prospective 14 1.00 – – 0.96 5
Retrospective 9 1.52 0.66 to 3.49 0.32
Diagnostic criterion 1 Not ICD-10 or DSM-IV 9 1.00 – – 0.69 32
ICD-10 or DSM-IV 14 3.08 1.52 to 6.25 0.002
*Too few studies relative to the number of categories in diagnostic criterion 2 (criteria separately) were available to include this covariate in the analysis.
Table 5 Two multivariate meta-regression models for studies of all autism spectrum
disorders (n = 23)
Covariate Odds ratio 95% CI (odds ratio) p value
Model 1—including age

copyright.
ICD-10 or DSM-IV 2.61 1.40 to 4.85 0.003
Age (years) 0.85 0.85 0.76 to 0.96 0.006
Intercept 1.4561023 1.4561023 3.2561024 to 6.4561023 ,0.001
Model 2—including study area

ICD-10 or DSM-IV 3.48 1.92 to 6.33 ,0.001
Area Rural/mixed 1.00 – –
Urban 2.85 1.47 to 5.53 0.002
Intercept 2.0361024 7.1561025 to 5.7361024 ,0.001
studied, with studies from Japan having significantly higher method relied on records, these may have been more
estimates than North American studies. This could be due to complete in urban locations. If the screen method used
other study factors. For example, a higher proportion of the referrals from clinicians, it is possible that a higher
Japanese studies were from urban areas (4/7 (57%) studies) proportion of children were known to services in urban
compared to those in North America (1/6 (17%) studies). All locations. There may have been different diagnostic practices
the Japanese studies used prospective diagnostic assess- in urban locations where staff were more likely to be
ments, and all but one drew on whole population rather than employed at specialist healthcare centres than in rural
clinical samples. Due to the imposed limit of three covariates locations. It is easier to access the population in urban
in the model, it was not possible to adjust for further locations, and response rates may have been higher, but data
potential effect modifiers. Countries differ in their diagnostic on response were too limited to investigate this.
practice both in their theoretical background and their
training procedures for healthcare workers. This may, in Limitations and recommendations for future research
part, account for between-region variation in prevalence. Publication bias was not investigated in this review, as funnel
In an alternative model for typical autism, when adjusting plots were not considered appropriate due to the large degree
for age and diagnostic criteria, studies including prospective of variation across studies. It is unlikely that the set of papers
diagnostic assessments gave rise to higher prevalence published is biased with respect to prevalence reported.
estimates than those using retrospective records. This may However, it is possible that some studies were not identified
be linked to the use of different diagnostic methodology at in the searches if they were not published in mainstream
different times. Alternatively, an assessor taking part in journals. There may have been some time lag bias, with
prospective research studies might observe children more smaller studies, or studies with unremarkable results, coming
closely for symptoms of ASD. through to publication slower than larger studies.
When adjusting for diagnostic criteria, urban location was Of the papers identified for detailed examination, five
also observed to be associated with higher prevalence potentially eligible studies were excluded as they did not have
estimates for both typical autism and all ASD. If the screen a detailed English summary or were not peer reviewed. There
14 Williams, Higgins, Brayne
is no reason to suspect that the lack of availability of data 6 Gillberg C, Wing L. Autism: not an extremely rare disorder. Acta Psychiatr
Scand 1999;99:399–406.
from these studies is a direct consequence of the prevalences 7 Wing L. The definition and prevalence of autism: a review. European Child
they might have observed. and Adolescent Psychiatry 1993;2:61–74.
The choice of coding of the covariates may have affected 8 Lewis S, Clarke M. Forest plots: trying to see the wood and the trees. BMJ
2001;322:1479–80.
the model, such as using the midpoint of the age range or 9 Higgins JP, Thompson SG, Deeks JJ, et al. Measuring inconsistency in meta-
grouping diverse diagnostic criteria. Furthermore, it was only analyses. BMJ 2003;327:557–60.
possible to assess the impact of reported covariates, or easily 10 Whitehead A, Whitehead J. A general parametric approach to the meta-
analysis of randomized clinical trials. Stat Med 1991;10:1665–77.
quantifiable covariates. Qualitative influences on prevalence
11 Goodman SN. Meta-analysis and evidence. Control Clin Trials
such as awareness of autism in each population could not be 1989;10:188–204.
included. As more studies are published, it may be possible to 12 Sharp S. Meta-analysis regression. Stata Technical Bulletin Reprints
include new covariates or more precise coding of existing 2003;7(STB42):148–55.
13 Altman D. Practical statistics for medical research. London: Chapman & Hall,
covariates in such a model. It would be valuable to have even 1991:349.
more thorough recording of study characteristics in future 14 Lotter V. Epidemiology of autistic conditions in young children. Social
studies to facilitate meta-analyses of studies. Psychiatry 1966;1:124–37.
15 Treffert DA. Epidemiology of infantile autism. Arch Gen Psychiatry
It is unlikely that it would ever be possible to measure and 1970;22:431–8.
record all potentially important covariates. An alternative 16 Brask BH. A prevalence investigation of childhood psychoses. Nordic
approach to investigating trends in prevalence, through symposium on the care of psychotic children. Oslo: Barnepsychiatrist
Forening, Universitetsforlagets Trykningssentral, 1970.
ongoing monitoring of defined school aged populations using
17 Wing L, Gould J. Severe impairments of social interaction and associated
standard methodology, has been recommended.1 This would abnormalities in children: epidemiology and classification. J Autism Dev
enable researchers to investigate changes in prevalence over Disord 1979;9:11–29.
time, and geographical variations while controlling for study 18 Hoshino Y, Kumashiro H, Yashima Y, et al. The epidemiological study of
autism in Fukushima-ken. Folia Psychiatr Neurol Jpn 1982;36:115–24.
methodology. 19 Ishii T, Takahashi O. The epidemiology of autistic children in Toyota, Japan:
prevalence. Japanese Journal of Child and Adolescent Psychiatry
1983;24:311–21.
Conclusions 20 Bohman M, Bohman IL, Bjorck PO, et al. Childhood psychosis in a northern
This review has contributed to explaining some of the Swedish county: some preliminary findings from an epidemiological survey.
influences on variation among prevalence estimates. Over In: Schmidt M, Schmidt R, eds. Epidemiological approaches in child
psychiatry. Stuttgart: George Thieme Verlag, 1983:164–73.
half of the variation among study estimates can be explained 21 McCarthy P, Fitzgerald M, Smith MA. Prevalence of childhood autism in
by the age of the children screened, the diagnostic criteria Ireland. Ir Med J 1984;77:129–30.
used, and the country studied. Other important factors were 22 Gillberg C. Infantile autism and other childhood psychoses in a Swedish
urban region. Epidemiological aspects. J Child Psychol Psychiatry
whether the study was in a rural or urban location and 1984;25:35–43.
whether cases were assessed prospectively or retrospectively. 23 Steinhausen HC, Gobel D, Breinlinger M, et al. A community survey of
The impact of these identified factors on prevalence estimates infantile autism. J Am Acad Child Psychiatry 1986;25:186–9.
24 Steffenburg S, Gillberg C. Autism and autistic-like conditions in Swedish rural
should now be further investigated as they may be acting as
copyright.
and urban areas: a population study. Br J Psychiatry 1986;149:81–7.
proxies for other influences on prevalence. For example, the 25 Burd L, Fisher W, Kerbeshian J. A prevalence study of pervasive
effect of geographical location on prevalence may be due to developmental disorders in North Dakota. J Am Acad Child Adolesc
the services available, or variation in awareness of the Psychiatry 1987;26:700–3.
26 Matsuishi T, Shiotsuki Y, Yoshimura K, et al. High prevalence of infantile
disorder. By taking this quantitative approach, this review autism in Kurume City, Japan. J Child Neurol 1987;2:268–71.
has shown that using meta-analytic techniques can be a 27 Tanoue Y, Oda S, Asano F, et al. Epidemiology of infantile autism in southern
valuable additional tool in deepening our understanding of Ibaraki, Japan: differences in prevalence in birth cohorts. J Autism Dev Disord
1988;18:155–66.
the influences of study and population characteristics on 28 Bryson SE, Clark BS, Smith IM. First report of a Canadian epidemiological
variation in prevalence estimates in autism spectrum study of autistic syndromes. J Child Psychol Psychiatry 1988;29:433–45.
disorders. 29 Ritvo ER, Freeman BJ, Pingree C, et al. The UCLA–University of Utah
epidemiologic survey of autism: prevalence. Am J Psychiatry
1989;146:194–9.
ACKNOWLEDGEMENTS 30 Sugiyama T, Abe T. The prevalence of autism in Nagoya, Japan: a total
This work forms part of a PhD thesis entitled ‘‘Screening for autism population study. J Autism Dev Disord 1989;19:87–96.
spectrum disorders’’ (University of Cambridge, 2003). Jo Williams 31 Cialdella P, Mamelle N. An epidemiological study of infantile autism in a
(née Johnson) held an MRC studentship, and subsequently received French department (Rhone): a research note. J Child Psychol Psychiatry
1989;30:165–75.
funding from the Shirley Foundation. We wish to thank Prof Patrick
32 Gillberg C, Steffenburg S, Schaumann H. Is autism more common now than
Bolton and Prof Simon Baron-Cohen for their comments on an ten years ago? Br J Psychiatry 1991;158:403–9.
earlier draft of this paper. 33 Ohtaki E, Kawano Y, Urabe F, et al. The prevalence of Rett syndrome and
infantile autism in Chikugo District, the southwestern area of Fukuoka
..................... prefecture, Japan [letter]. J Autism Dev Disord 1992;22:452–4.
34 Fombonne E, du Mazaubrun C. Prevalence of infantile autism in four French
Authors’ affiliations regions. Soc Psychiatry Psychiatr Epidemiol 1992;27:203–10.
J G Williams, C E G Brayne, Department of Public Health and Primary 35 Herder GA. [Infantile autism among children in the county of Nordland.
Care, University of Cambridge, UK Prevalence and etiology]. Tidsskr Nor Laegeforen 1993;113:2247–9.
J P T Higgins, MRC Biostatistics Unit, Cambridge, UK and Public Health 36 Honda H, Shimizu Y, Misumi K, et al. Cumulative incidence and prevalence of
Genetics Unit, University of Cambridge, UK childhood autism in children in Japan. Br J Psychiatry 1996;169:228–35.
37 Arvidsson T, Danielsson B, Forsberg P, et al. Autism in 3-6-year-old children
Competing interests: none declared in a suburb of Goteborg, Sweden. Autism 1997;1:163–73.
38 Webb EV, Lobo S, Hervas A, et al. The changing prevalence of autistic
disorder in a Welsh health district. Dev Med Child Neurol 1997;39:150–2.
39 Fombonne E, du Mazaubrun C, Cans C, et al. Autism and associated medical
REFERENCES disorders in a French epidemiological survey. J Am Acad Child Adolesc
1 Fombonne E. Epidemiological surveys of autism and other pervasive Psychiatry 1997;36:1561–9.
developmental disorders: an update. J Autism Dev Disord 2003;33:365–82. 40 Sponheim E, Skjeldal O. Autism and related disorders: epidemiological
2 Wing L, Potter D. The epidemiology of autistic spectrum disorders: is the findings in a Norwegian study using ICD-10 diagnostic criteria. J Autism Dev
prevalence rising? Mental Retardation and Developmental Disabilities Disord 1998;28:217–27.
Research Review 2002;8:151–61. 41 Kadesjo B, Gillberg C, Hagberg B. Brief report: autism and Asperger
3 Fombonne E. Epidemiological trends in rates of autism. Mol Psychiatry syndrome in seven-year-old children: a total population study. J Autism Dev
2002;7(suppl 2):S4–6. Disord 1999;29:327–31.
4 Charman T. The prevalence of autism spectrum disorders. Recent evidence 42 Powell JE, Edwards A, Edwards M, et al. Changes in the incidence of
and future challenges. Eur Child Adolesc Psychiatry 2002;11:249–56. childhood autism and other autistic spectrum disorders in preschool children
5 Fombonne E. The epidemiology of autism: a review. Psychol Med from two areas of the West Midlands, UK. Dev Med Child Neurol
1999;29:769–86. 2000;42:624–8.
43 Kielinen M, Linna SL, Moilanen I. Autism in Northern Finland. Eur Child 49 Lingam R, Simmons A, Andrews N, et al. Prevalence of autism and parentally
Adolesc Psychiatry 2000;9:162–7. reported triggers in a north east London population. Arch Dis Child
44 Baird G, Charman T, Baron-Cohen S, et al. A screening instrument for autism 2003;88:666–70.
at 18 months of age: a 6-year follow- up study. J Am Acad Child Adolesc 50 Tebruegge M, Nandini V, Ritchie J. Does routine child health surveillance
Psychiatry 2000;39:694–702. contribute to the early detection of children with pervasive developmental
45 Magnusson P, Saemundsen E. Prevalence of autism in Iceland. J Autism Dev disorders? An epidemiological study in Kent, U.K. BMC Pediatr 2004;4:4.
Disord 2001;31:153–63. 51 Fombonne E, Simmons H, Ford T, et al. Prevalence of pervasive
46 Bertrand J, Mars A, Boyle C, et al. Prevalence of autism in a United States developmental disorders in the British nationwide survey of child mental
population: the Brick Township, New Jersey, investigation. Pediatrics health. J Am Acad Child Adolesc Psychiatry 2001;40:820–7.
2001;108:1155–61. 52 Scott F, Baron-Cohen S, Bolton P, et al. Brief report: prevalence of autism
47 Chakrabarti S, Fombonne E. Pervasive developmental disorders in preschool spectrum conditions in children aged 5–11 years in Cambridgeshire, UK.
children. JAMA 2001;285:3093–9. Autism 2002;6:231–7.
48 Croen LA, Grether JK, Hoogstrate J, et al. The changing prevalence of autism 53 Yeargin-Allsopp M, Rice C, Karapurkar T, et al. Prevalence of autism in a US
in California. J Autism Dev Disord 2002;32:207–15. metropolitan area. JAMA 2003;289:49–55.
Clinical Evidence—Call for contributors
Clinical Evidence is a regularly updated evidence-based journal available worldwide both as

a paper version and on the internet. Clinical Evidence needs to recruit a number of new
contributors. Contributors are healthcare professionals or epidemiologists with experience in
evidence-based medicine and the ability to write in a concise and structured way.
Areas for which we are currently seeking contributors:
N Pregnancy and childbirth
N Endocrine disorders
N Palliative care
N Tropical diseases
We are also looking for contributors for existing topics. For full details on what these topics
are please visit www.clinicalevidence.com/ceweb/contribute/index.jsp
However, we are always looking for others, so do not let this list discourage you.
copyright.
Being a contributor involves:
N Selecting from a validated, screened search (performed by in-house Information
Specialists) epidemiologically sound studies for inclusion.
N Documenting your decisions about which studies to include on an inclusion and exclusion
form, which we keep on file.
N Writing the text to a highly structured template (about 1500-3000 words), using evidence
from the final studies chosen, within 8-10 weeks of receiving the literature search.
N Working with Clinical Evidence editors to ensure that the final text meets epidemiological
and style standards.
N Updating the text every 12 months using any new, sound evidence that becomes available.
The Clinical Evidence in-house team will conduct the searches for contributors; your task is
simply to filter out high quality studies and incorporate them in the existing text.
If you would like to become a contributor for Clinical Evidence or require more information
about what this involves please send your contact details and a copy of your CV, clearly
stating the clinical area you are interested in, to [email protected].
Call for peer reviewers
Clinical Evidence also needs to recruit a number of new peer reviewers specifically with an
interest in the clinical areas stated above, and also others related to general practice. Peer
reviewers are healthcare professionals or epidemiologists with experience in evidence-based
medicine. As a peer reviewer you would be asked for your views on the clinical relevance,
validity, and accessibility of specific topics within the journal, and their usefulness to the
intended audience (international generalists and healthcare professionals, possibly with
limited statistical knowledge). Topics are usually 1500-3000 words in length and we would
ask you to review between 2-5 topics per year. The peer review process takes place
throughout the year, and out turnaround time for each review is ideally 10-14 days.
If you are interested in becoming a peer reviewer for Clinical Evidence, please complete the
peer review questionnaire at www.clinicalevidence.com/ceweb/contribute/peerreviewer.jsp

Systematic Review of Prevalence Studies of Autism Spectrum Disorders

Uploaded by

Systematic Review of Prevalence Studies of Autism Spectrum Disorders

Uploaded by

8

Systematic review of prevalence studies of autism

Data extraction Investigation of sources of heterogeneity

(1) Primary research Description of heterogeneity among studies and

Table 1 Summary of prevalence studies of autism spectrum disorders (42 studies)

Lotter, 1966 Wing, 1979

Table 3 Multivariate meta-regression results, for studies of typical autism (n = 35)*

Diagnostic criteria Not ICD-10 or DSM-IV 1.00 – –

Model 1—including age

Model 2—including study area

Clinical Evidence—Call for contributors

Clinical Evidence is a regularly updated evidence-based journal available worldwide both as

Call for peer reviewers

You might also like