J Autism Dev Disord (2016) 46:1669–1685

DOI 10.1007/s10803-016-2696-6

ORIGINAL PAPER

Prevalence of Autism Spectrum Disorders in Guanajuato, Mexico:

The Leon survey
Eric Fombonne1 • Carlos Marcin2 • Ana Cecilia Manero3 • Ruth Bruno4 •
Christian Diaz3 • Michele Villalobos5 • Katrina Ramsay6 • Benjamin Nealy6

Published online: 21 January 2016

! Springer Science+Business Media New York 2016

Abstract There are no epidemiological data on autism
for Mexico. This study was conducted to generate a first
estimate of ASD prevalence in Mexico. We surveyed
children age eight in Leon (Guanajuato). The sample was
stratified in two strata: (1) children having special educa-
tion and medical records (SEMR; N = 432) and (2) chil-
dren attending regular schools (GSS; N = 11,684). GSS
children were screened with the SRS and those with the
highest scores were invited to a diagnostic evaluation. The
final sample comprised 36 children (80.6 % male) who had
confirmed ASD. A third had intellectual disability, 25 %
were non-verbal, 69 % had co-occurring behavioral prob-
lems. The prevalence overall was 0.87 % (95 % CI 0.62,
1.1 %). This survey provides an estimate for ASD preva-
lence in Mexico that is consistent with recent studies.
Keywords Autism spectrum disorders ! Epidemiology !
Screening ! Prevalence ! Cultural
Introduction
Autism spectrum disorder (ASD) or pervasive develop-
mental disorder (PDD; we use ASD throughout for sim-
plification) is a lifelong condition characterized by
pervasive impairments in social reciprocity and commu-
nication, and stereotyped behaviors and restricted interests
(APA 1994, 2013). Recent reviews have concluded that
current best estimates for the prevalence of ASD lie
between 0.7 and 1.0 % (Fombonne 2009a, b; Elsabbagh
et al. 2012; Hill et al. 2014). However, some studies have

1
& Eric Fombonne Department of Psychiatry, Institute for Development and
fombonne@ohsu.edu Disability, Oregon Health & Science University, 840 SW
Gaines Str., Mail Code: GH254, Portland, OR 97239, USA
Carlos Marcin 2
inata2001@hotmail.com CLIMA (Clı́nica Mexicana de Autismo – México City),
Universidad IberoAmericana Mexico City, Van Dyck 66
Ana Cecilia Manero Street, Colonia Mixcoax, Mexico, D.F, Mexico
acmanero@hotmail.com 3
CLIMA-L (Clı́nica Mexicana de Autismo León), Universidad
Ruth Bruno del Valle de Atemajac, Valle del Mezquital # 101, Colonia
ruth.bruno@muhc.mcgill.ca del Valle del Campestre, León, Guanajuato, Mexico
Christian Diaz 4
Montreal Children’s Hospital, McGill University, 4018 Ste-
christian100285@hotmail.com Catherine West, Montreal, QC H3Z 1P2, Canada
Michele Villalobos 5
TEACCH" Autism Program, University of North Carolina at
michele_villalobos@med.unc.edu; Chapel Hill, 31 College Place, Building D, Suite 306,
http://www.TEACCH.com Asheville, NC 28801, USA
Katrina Ramsay 6
Division of Biostatistics, OHSU-PSU School of Public
ramseyk@ohsu.edu Health, Portland, OR, USA
Benjamin Nealy
ben.nealy@gmail.com

123
1670
yielded estimates over 2 % (Pelly et al. 2015; Randall et al.
2015) and even as high as 2.64 % (Kim et al. 2011). The
surveillance of ASD in the US conducted by the CDC on
8 year olds has led to several increasing estimates, the
latest being 1.47 % (CDC 2014). There is no standardized
methodology to conduct epidemiological surveys of ASD,
and the variation in prevalence proportions reported in the
literature reflects, at least in part, this lack of standardiza-
tion. Factors that explain the high variability in ASD
prevalence estimates have been discussed in details else-
where (Fombonne 2003, 2009a; Hill et al. 2014), and
include: the age range of participants (with lower rates in
preschoolers and teenagers or adults), sample size (with
smaller studies yielding higher rates), the comprehensive-
ness and sensitivity of case finding methods (cases already
diagnosed or not, reliance on existing records or more
proactive screening methods), the diagnostic concepts and
criteria used to define caseness (with higher rates in studies
based on ICD-10 and DSM-IV), and methods used to
confirm case status (expert clinical judgement of research
files or direct evaluation of subjects). Although there is no
doubt that the prevalence of ASD has increased steadily
over the last 30 years (Fombonne 2009a, b; Hill et al.
2014), the interpretation of this trend remains uncertain.
Survey methods variability, changes in diagnostic concepts
and broadening of definitions, diagnostic substitution,
improved identification and awareness all have contributed
to this trend (Fombonne 2009a, b; Hill et al. 2014). Yet, the
hypothesis that an increase in the underlying incidence of
ASD also accounts for some of the upward trend in
prevalence cannot be ruled out. If confirmed, it could point
to environmental risk factors playing an etiological role in
ASD alongside that of well-established genetic factors
(Willsey and State 2015; Chen et al. 2015). Consequently,
monitoring the prevalence of ASD has become a public
health priority in many countries.
With the exception of three preliminary studies in
Venezuela (Montiel-Nava and Peña 2008), Argentina
(Lejarraga et al. 2008) and Brazil (Paula et al. 2011), no
major survey has been conducted in Central or Latin
America. These studies differed widely in their method-
ology and their prevalence estimates, with figures of
0.17 % in Venezuela, 0.27 % in Brazil and 1.3 % in
Argentina. Two studies (Lejarraga et al. 2008; Paula et al.
2011) had small sample sizes (N \ 1500), included vari-
able age groups ranging from birth to age 12, used a two-
phase design with screeners specific to ASD or not, and
employed variable direct assessments to confirm caseness.
The third study had a larger sample size (N & 255,000)
but relied on reviewing medical and educational records of
children already diagnosed with ASD by a community
professional (Montiel-Nava and Peña 2008). The study
with the highest estimate (Lejarraga et al. 2008) surveyed a
123
J Autism Dev Disord (2016) 46:1669–1685
clinical sample recruited in health clinics, with uncertain
sampling and participation rates, and used screening and
diagnostic instruments that were not specific to ASD. No
study provided data on the reliability of the screening tools
or final diagnostic procedures. No epidemiological data on
ASD exists for Mexico.
In the US, the Centers for Disease Control and
Prevention (CDC) were given a mandate to develop a
surveillance system for the US as early as 1996. The CDC
piloted a survey methodology that was first applied to the
metropolitan area of Atlanta, Georgia, leading to a first
estimate of prevalence of 0.34 % among 3–10 year olds
with a peak prevalence of 0.47 % among 8 year olds
(Yeargin-Allsopp et al. 2003). The methodology devised
by the CDC relied on systematic screening of the child
population age eight using multiple data sources. Trained
lay surveyors reviewed medical, educational records and
data from other health professionals and sources. The cases
were then abstracted when the source contained some
evidence of a social ‘trigger’ (i.e. a description possibly
suggestive of a social deficit in the target child, whether or
not an ASD diagnosis was already reported). An expert
panel subsequently reviewed all available information in
order to derive a best estimate diagnosis for each subject
included in the survey (Van Naarden Braun et al. 2007).
There are two noticeable limitations to the CDC method-
ology. First, children are not directly assessed and case
status is inferred from record reviews only. A validation
study of the record review procedure on a small sample
indicated a low sensitivity (0.60) for the record-review
procedure when compared to an in-depth clinical evalua-
tion. Furthermore, substantial misclassification occurred
with 21 % of record-review cases not confirmed by clinical
evaluation, and 9 % of non cases by record review being
determined to be cases by the clinical method. As a result,
potential underestimation of prevalence by a factor of 32 %
could have occurred (Avchen et al. 2011). Second, no
attempt is made to survey children without record tracing
past or current contacts with services. Nevertheless,
although it has been criticized (Mandell and Lecavalier
2014; Lecavalier and Mandell 2015), the methodology of
combining developmental/behavioral data stemming from
different health and education sources irrespective of a pre-
existing ASD diagnosis has provided an efficient and cost
effective way to ascertain cases in large community sam-
ples (Durkin et al. 2015; Newschaffer et al. 2015).
The survey methodology that was calibrated as part of
the CDC initial surveillance efforts was used in sequential
US surveys of children recruited every 2 years at multiple
sites. To account for underascertainment in preschoolers
and long delays between first evaluations and time of actual
diagnosis (Wiggins et al. 2006; Van Naarden Braun et al.
2007), cohorts of children age eight were selected,
J Autism Dev Disord (2016) 46:1669–1685
representing approximately 10 % of the US population in
each survey. The surveys resulted in increasing prevalence
estimates, with values of 0.67 % (CDC 2007a), 0.66 %
(CDC 2007b), 0.9 % (CDC, 2009), 1.13 % (CDC 2012),
and 1.47 % (CDC 2014) in the 1992, 1994, 1998, 2000 and
2002 birth cohorts, respectively. There is a strong advan-
tage in relying on the same methodology in surveys repe-
ated over time, especially for surveillance of large samples
of a population and detection of time trends in rates.
In 2009, the principal investigators of the present study
organized two meetings with national health authorities
and various local professionals and administrators that led
to the planning, design and execution of the first epi-
demiological survey of ASD in one specific region of
Mexico. This study has features that are comparable to the
CDC surveys such as the age of the target population and
some methods employed to ascertain cases with existing
medical or educational records (see below). However, a
major difference lies in our additional inclusion of a gen-
eral school two-phase survey to ensure more complete case
ascertainment. The main objectives of this study were: (1)
to test the applicability of autism survey methodology
among 8 year olds in a Mexican context; (2) to generate the
first estimate for the prevalence of ASD in Mexico; and (3)
to describe the correlates and developmental characteristics
of ASD cases identified in a population survey in order to
inform future service planning around the needs of Mexi-
can families with children with an ASD.
Materials and Methods
Study Site
The survey was conducted in Leon, a city located in
Guanajuato, one of the 32 states of Mexico. The distribu-
tion of economic activities in this state is comparable to the
national average with 32.2 % (25.9 %) in manufacturing,
29.5 % (33.5 %) in the commercial sector, 30.2 %
(27.4 %) in private non-financial services, and 9.1 %
(13.2 %) in other sectors including transports, construction
and agriculture, for Guanajuato (Mexico) respectively
(Instituto Nacional de Estadistica y Geografia 2015a). Leon
metropolitan area is the seventh largest in the country with
little more than 1,750,000 inhabitants. The population has
an homogeneous ethnic composition and the lifestyle
reflects the mixed, rural and urban, economy. For most of
the twentieth century, Leon’s main economic activity
revolved around the production of leather goods, but most
recently it began reorienting itself toward automotive
industry. Other economic areas that have shown important
growth are health care, higher education, research, and
tourism. Although Leon is not the political state capital, it
1671
is considered the social, commercial and government ser-
vices ‘capital’ (Instituto Nacional de Estadistica y Geo-
grafia 2015b). On several socio-demographic indicators,
the city of Leon is close to the country average, including
the proportion of persons between the ages of 15 and 29
(28.0 vs 26.8 %), the average number of occupants in
private dwellings (4.4 vs 3.9), the average number of live
births in women over age 12 (2.3 vs 2.3), the mean years of
schooling in persons over age 15 (8.5 vs 8.6), and literacy
rates among persons between ages 15 and 29 (98.8 vs
98.4 %) (Instituto Nacional de Estadistica y Geografia
2015b). Besides being a middle-size representative city,
Leon was selected for logistic reasons as a local autism
team from the Clı́nica Mexicana de Autismo (CLIMA)
with expertise in ASD could provide the back-up necessary
to respond to the clinical work generated by the survey and
was used to validate the diagnosis in the case sample of the
main study.
Sample Selection and Case Finding Methods
The target population was 8 year-old children residents of
Leon, Guanajuato, who attended in 2011–2012 a regular or
a special education school. Of the 554 primary schools of
the 6 school districts in Leon, 134 schools included in 4
typical school districts were chosen by the Educational
Authority to meet our requirements of adequate represen-
tativeness of the schools and of a sample size of at least
10,000 students. Children included in the study were
residing in Leon and were born between January 1st and
December 31st 2003. The final sample included in our
study was selected in 24.2 % of Leon primary schools and
the participants (N = 12,116) represented 32.4 % of the
total school population of children age eight. The sample
was divided in two separate strata. First, children who had
been in contact with health or educational services for
developmental, medical, behavioral, or learning problems
were identified through systematic screening of records at
each of the data sources described later (special education
and medical records—SEMR; N = 432). Second, we
undertook a survey of the general school population of
children meeting the above criteria for inclusion and who
had not been identified through a search of records (general
school sample—GSS; N = 11,684). The case identification
process is illustrated in Fig. 1 and we provide further
details in the next two sections.
General School Sample (GSS)
In Mexico, school registration is compulsory at age six and
registration in the school system occurs in February of each
year for the school year starting in August of the same year.
School rosters are held at the State Educational Board and
123
1672 J Autism Dev Disord (2016) 46:1669–1685

updated regularly. Each school has the registration of its
own yearly contingent of pupils, including their names,
sex, date of birth, and address. In addition, at the time of
school registration, children undergo a brief developmental
assessment to evaluate their readiness for Grade 1 entry.
Children born in 2003 were attending either Grade 2 or 3.
We used a two-phase design to identify cases of ASD.
Screening Phase Four school districts were chosen to
participate in the study, including 11,684 children attend-
ing 134 schools. Only children meeting the age criterion
were included. After preliminary meetings with school
staff and parent associations, envelopes were dispatched to
the pupils and their families through the primary teachers.
Each envelope contained a letter of information outlining
the goals and methods of the study and the procedures
followed for informed consent and confidentiality. Dates
for surveying each school were set, and during those days,
research staff went to schools to distribute the screening
instrument [Social Responsiveness Scale (SRS); Con-
stantino and Todd 2003; see below], help parents to com-
plete them if they wished, and collect the completed
checklists. Parents were also asked to pass a second copy of
the screening instrument to the primary teacher of their
child who could then return it directly to the research team
after completion. Each SRS was scored and entered in a
database. As no cut-off on the SRS has been validated for
screening purposes in populations with a low (1 %) rate of
the target disorder, and given our limited resources, we
separated the screened population based on the SRS scores
statistical distribution. At first, we had planned to separate
children into three groups based on the percentile of the
SRS, namely having a screen negative group (\95th cen-
tile), and two groups of screen positives (95th–98th centile,

Fig. 1 Case identification

process
School sectors (1, 2, 3 & 4)
N=12,116 children age 8

General School Sample (GSS)

Medical Records Special Education and Medical Records
(SEMR)
N=11,684 SRS Sent
N=432

SRS returned:
-Parents: N=4,195
(response rate: 36.2%)
-Teachers: N=2,761 N screened, abstracted, confirmed ASD
-Parent or teacher: N=4,431 CLIMA: 13, 13, 13
DIF : 187, 9, 5
CAM: 16, 10, 1
USAER: 129, 11, 5
th Health Prof: 87, 26, 7
Screen positive (>99 centile)
Total: 432, 69, 31
N=43 invited to next phase

Diagnostic confirmation phase

N=25 assessed (58.1%)

20 non-ASD 5 ASD 401 non-ASD 31 ASD

123
J Autism Dev Disord (2016) 46:1669–1685
and C99th centile); the two screen positive and the screen
negative groups would have then been sampled with
unequal probabilities to assess the proportion of ASD cases
in each. However, constraints on the study duration and
funding did not allow to follow this protocol. Instead, we
concentrated on the highest scorers only, considering as
‘screen positives’ children with a score at or above the 99th
centile; children with a score below the 99th centile were
all considered ‘screen negatives’ but no random sample
from that group could be further evaluated.
Diagnostic Confirmation Phase In this second phase, all
children of the screen positive group and their parents were
invited to a diagnostic confirmation assessment. When
possible, we used the ADI-R (Rutter et al. 2003) and the
ADOS-G (Lord et al. 2002) and other assessments
including a test of cognitive functioning and a behavioral
assessment. The research staff at CLIMA-L were clinically
experienced and received ad hoc training in the use of the
Spanish ADOS and ADI-R at the beginning of the study. In
some instances, families could not attend a lengthy
appointment and the child’s development and functioning
was reviewed using more limited clinical techniques,
including brief parent and teacher interview and direct
child observation. All results and observations were
abstracted on the survey form designed for our study, the
LASI (Leon Autism Survey Inventory; see below).
Special Education and Medical Records—SEMR
After ethical approval was obtained (see below), we
screened the records of children meeting inclusion criteria
at the following sources.
Desarrollo Integral de la Familia (DIF: Integral Family
Development) The DIF provides early intervention ser-
vices for young children with developmental difficulties. It
also has one Rehabilitation Center (CEEM: Centro Espe-
cializado de Estimulación Múltiple) that provides services
for children with Cerebral Palsy, Intellectual Disability,
Down Syndrome, rare genetic syndromes, premature
babies at neurological risk, myelomeningocele, etc. Chil-
dren who have motor deficits tend to stay in the CEEM for
their primary education. DIF serves approximately 8000
children, 3000 of which are regular patients (50 % under
age 10) in a given year. Children seen in the preschool
period are referred to DIF by health professionals. Children
with developmental, behavioral, and social problems, are
referred to the Centro de Atención Psicopedagógica de
Educación Preescolar (CAPEP: Psychological and Educa-
tional Preschool Center Service) preschool program (there
is just one CAPEP in Leon). CAPEP provides assessments,
outpatient consultations and treatment, and they also have
1673
one on-site classroom for children with special needs.
Graduates of CAPEP can be subsequently sent to USAER
or CAMS (see below), or regular classes. DIF also inter-
venes in 6 day care centers (operating from age 3 months
to age 6 years) where preschoolers with special needs are
schooled together with typical children. A confidentiality
agreement was signed between the Leon DIF director and
the research team to access the DIF data. The DIF main-
tains a database that can be consulted and has the date of
birth of children, and the diagnosis (except when it is
simply a language delay). We used the DIF database to
identify children who may have ASDs. Medical records
were consulted and abstracted on the LASI.
Special Schools: Centro de Atención Múltiple (CAM;
Special School for Multiple Handicaps) The CAM has
been the Special Education School of the Education
Bureau since 1980, for children six to 12 years old. They
provide services for multiple handicapped children,
children with cerebral palsy, mental retardation, Down
syndrome, children with sensory handicaps (deafness),
and, more recently, for children with autism. The CAM
system provides primary school education with a cur-
riculum adapted to take into account the global and
specific learning disabilities of the students. A typical
CAM class has 15 children with one special education
teacher. The CAM children usually come with a psy-
chological assessment from another Health or Education
institution, and some of them also have a medical/neu-
rological diagnosis. Children with ASD have typically
been underdiagnosed, often mislabelled as mentally
retarded, reflecting the lack of special assessment tools
and expertise in autism in the special education system.
Medical and educational records from all children
meeting age criteria (N = 16) were screened and 10
were abstracted on the LASI.
Secretaria de Educación Publica (SEP: Sceretary of Public
Education) - Unidad de Servicios de Apoyo a la Educación
Regular (USAER;Units of Support Services to Regular
Education) The SEP manages the USAERs which are
special education teams supporting general education
schools. The SEP provides evaluation as well as direct and
indirect interventions to children with learning or behav-
ioral difficulties within their regular classroom or school.
The USAER teams are comprised of psychologists, edu-
cators, and speech pathologists that support several schools
within a given geographical area. Files of children are kept
centrally in the USAER. All children (N = 129) followed
by the USAER and eligible according to their year of birth
and school district were screened for ASD by a member of
the research team. When children’s records indicated the
possible presence of ASD symptoms (N = 11), data were
123
1674
abstracted on the LASI for later review by the research
diagnostic panel.
CLIMA The CLIMA is the only national organization in
Mexico that provides assessment and intervention services for
children with ASDs. CLIMA is a private non-profit organi-
zation that maintains 18 clinics across the country. CLIMA
has a regional clinic in Leon that has 10 permanent full-time
staff, five clinical psychologists (including two with a master
degree) and five educational psychologists with at least
8 years of clinical experience. The CLIMA-Leon (CLIMA-L)
team has been in operation since 1998 and has evaluated
several hundred children in recent years. The clinicians of
CLIMA-L administer routinely standardized diagnostic,
cognitive and adaptive behaviour measures such as the
Wechsler Intelligence Scale for Children (Weschler 2004), the
Neurodevelopmental Assessment NEPSY (Korkman et al.
1998), the Vineland Adaptive Behavior Scales (VABS;
Sparrow et al. 2005), the Inventario De Espectro Autista
(IDEA: Inventory of Autism Spectrum, Riviere 2002), the
Childhood Autism Rating Scale (CARS; Shopler et al. 2010),
the ADI-R (Rutter et al. 2003) and the ADOS-G (Lord et al.
2000, 2002). The CLIMA-L team also provides ongoing
intervention in the form of TEACCH classes, Picture Ex-
change Communication System (PECS), and Applied
Behavior Analysis (ABA) programs. CLIMA-L maintains a
database of its clients that is up-to-date. This database was
used to identify children residing in Leon, born during the
target year and having attended CLIMA-L for any reason at
any point during that time period. All children identified in
that way were included in the screening stage and their records
reviewed and abstracted on the LASI (N = 13).
Health Professionals, Hospitals and Private Clin-
ics Health professionals, hospitals and private clinics in
the Leon area were approached to help us identify children
in the target age range whom they knew and might have
ASD. A list of such professionals was generated by
including all practitioners accredited by their professional
organization for having skills and competency to assess
and manage children with ASD. This list included clinical
and educational psychologists, doctors, speech pathologists
and occupational therapists working in multidisciplinary
teams and clinics receiving referrals for ASD. The four
child neurologists of Leon who worked collaboratively
with CLIMA-L also participated in the study. Other health
professionals also contributed to the case finding. For
hospitals and clinics, discharge diagnoses following
admissions for pediatric patients were examined. ICD-9
codes of mental retardation, language delays, other devel-
opmental disorders, and autism (codes 299.x) were iden-
tified. Medical records were requested and data abstracted
from them. In total, 87 files of children in the age range
123
J Autism Dev Disord (2016) 46:1669–1685
were reviewed, and 26 files were abstracted corresponding
to records including a ‘trigger’ description for ASD.
Details of the sample recruitment for both the GSS and
the SEMR are summarized in Fig. 1.
Instruments
Screening Instrument (SRS)
For the screening phase of the GSS, we used the Spanish
version of the Social Responsiveness Scale. The SRS was
designed to measure autistic symptomatology and traits,
and the severity of the associated social impairment. It is
applicable to 4–18-year-olds and can be completed in about
15–20 min by a parent, a teacher, another caregiver or any
other informant knowledgeable about the child’s behavior
across contexts and over time. The SRS comprises 65 items
each scored on a Likert scale ranging from 1 (not true) to 4
(almost always true), with 17 items being reverse-scored.
When completed, a total raw score and five treatment
subscales scores (labeled social awareness, social cogni-
tion, social communication, social motivation, and autistic
mannerisms) can be generated. Only total raw scores were
used in our analyses.
In order to test the psychometric properties of the SRS in
Mexico, we had conducted a validation study before the
main survey (Fombonne et al. 2012). In brief, we recruited
a clinical sample of 200 children (81 % males; mean age:
7.4 years) with a confirmed diagnosis of ASD and a sample
of 363 control children (59.5 % males; mean age:
8.5 years) without ASD. The mean total and subscale raw
SRS scores were significantly different between the clinical
and control groups both for the parent and for the teacher
reports (p \ .001). Receiver operating characteristic
(ROC) analyses showed excellent discriminant validity of
the SRS in the Mexican sample (area under the curve:
0.962 for the parent, 0.960 for the teacher). Mexican SRS
scores were significantly higher than in the US and German
population for typically developing children but compara-
ble for clinically referred subjects. SRS psychometric
properties were excellent and comparable to those derived
from North American and other samples. Further details
can be found elsewhere (Fombonne et al. 2012).
To re-evaluate the test–retest reliability of the SRS in
this survey, we collected from a few parents and teachers
pairs of SRS administered on the same child a few weeks
apart. Informants were not informed initially that they
might be asked to complete the SRS on a second occasion.
Leon Autism Survey Inventory (LASI)
We developed a survey form to screen and abstract the
information from educational and medical records in a
J Autism Dev Disord (2016) 46:1669–1685
consistent format. The LASI provides an initial list of
descriptors that tap social interaction abnormalities that
may be noted in records and is a simplified version of the
list of social triggers used by the CDC (Van Naarden Braun
et al. 2007). For those children’s records endorsing one
such characteristic, the full LASI was completed with
sections containing socio-demographic information, medi-
cal and developmental history, data on service access and
service use from birth to present, co-occurring behavioral
problems, use of medications, comorbid medical and
genetic conditions, history of diagnoses and current diag-
nosis. Intellectual functioning was estimated with four
bands (normal: IQ C 85; borderline: IQ C 70 and \85;
mild intellectual disability (ID): IQ C 50 and \70; mod-
erate to severe ID: IQ \ 50) based on a clinical best esti-
mate derived from a review of all cognitive data available.
Language levels were assessed on 4-step scale, based on
current language proficiency (Fluent language; Simple
conversation; Limited language; Non-verbal). The LASI
was completed for all children’s records presenting a social
trigger in the SEMR sample, and for all children assessed
in phase two in the GSS survey. Each subject was assigned
a unique research ID number. Child’s gender and date of
birth, initials, and residence was tagged to it. As it was
expected that some children could be identified through
multiple data sources, redundant notifications were identi-
fied and the sample was cleaned up by eliminating dupli-
cate observations while consolidating all data on a given
child within one record only.
Expert Panel Review and Final Case Determination
The procedure to confirm caseness and establish that a
child had an ASD was based on a review of all docu-
mentation abstracted from school, medical records, SRS,
teacher and parent interviews, cognitive testing when
available, and standardized diagnostic tools such as the
ADOS and ADIS (Spanish versions) when they could be
administered. All data were transcribed on the LASIs that
were first reviewed independently by the two clinically
experienced research assistants. When the study was ini-
tiated, DSM-5 had not been released yet. Accordingly,
expert clinical judgement was guided by DSM-IV-TR
diagnostic criteria for Pervasive Developmental Disorders
(PDD) to establish case status (American Psychiatric
Association 2000). Despite the recognized lack of relia-
bility of diagnostic subtypes within the PDD umbrella,
subtyping of PDD according to DSM-IV-TR was per-
formed to facilitate comparisons with other studies. In the
text, we refer to all DSM-IV PDD diagnoses as ‘‘ASD’’, to
be consistent with the recent changes in the DSM (Amer-
ican Psychiatric Association 2013). Pairs of independent
diagnostic ratings were used to assess inter-rater
1675
agreement. In case of discrepancies between the initial
assessors, the cases were presented to the two lead authors
and a team consensus diagnosis was subsequently reached.
Data Collection Procedures
The study was conducted from the offices of CLIMA-Leon.
Two part-time (0.8 FTE) research assistants with clinical
training and secretarial support for CLIMA-L were hired.
The Mexican investigator led the local research team
throughout the survey. Under his supervision, the team
visited the various special education and general schools,
the medical centers and the health professionals acting as
informants to identify the study sample.
Before the data collection, the research team organized
meetings at each school for teachers where the study pro-
tocol and objectives were explained. As a small non-
financial incentive, teachers were presented with a Cer-
tificate of thanks for completing the SRS.
We formed an Advisory Committee (AC) including a
select group of local authorities, several teachers and child
neurologists, and family representatives. Three meetings of
the AC were organized in order to discuss the procedures
and logistics of the survey, and obtain feed-back and
recommendations.
The PIs met at the inception of the study with a number
of local authorities directors, including the Department of
Special Education in Guanajuato, the teams of CLIMA,
DIF and the head of SEP-USAER, the child neurologists
and the city hall authorities in Leon city. Additionally, they
met leading authorities at the Health System Surveillance
in Mexico City. In order to promote the study, the senior
authors gave two local TV interviews at peak viewing
times in 2011, and an interview with a local newspaper.
More than a dozen of additional media reports were
released on radios and newspapers at the initiative of the
local team.
Ethical Approval
The proposal was submitted to the Mexican Agency of
Public Health March 2010, and to the local educational and
medical authorities. An agreement was obtained between
DIF and CLIMA-L to have access to their records. A
similar agreement was established with the provincial
educational authority and private schools. The study pro-
tocol was approved by the Ethics Committee of McGill
University in October 2010.
Statistical Analyses
Survey data from the LASI were analysed in SPSS and
conventional statistical tests for categorical and
123
1676 J Autism Dev Disord (2016) 46:1669–1685

dimensional variables were used. In instances where scores the sensitivity of our results to that hypothesis, we repeated
were not normally distributed, we performed non-para- the estimation with a narrower (0.1–0.25 %) and broader
metric tests as well. As they yielded similar p values, we (0.1–0.75 %) ranges for the hypothesized prevalence.
only report results from parametric statistics. Items on the Over the 10,000 simulated populations, we calculated
LASI with over 30 % missing responses were omitted from prevalence as the number of cases divided by the entire
the analyses. Reliability was assessed with the intraclass population, then averaged these estimates for overall
correlation coefficient and the kappa statistic. Throughout, prevalence, using the 2.5th and 97.5th centiles as confi-
a p value of .05 was retained as level of statistical dence limits. The relative contributions of the SEMR and
significance. GSS groups to population prevalence were calculated as
In order to estimate the prevalence of ASD in the sur- the mean percentage of cases arising from each group over
vey, we used data from both the SEMR and GSS sub- all simulations.
samples. To calculate population prevalence and a 95 %
CI, we conducted simulations based on a combination of
sample findings and hypothesized values using the statis- Results
tical software package R, version 3.2.1. (R Development
Core Team 2008). Having no information about non-par- GSS Screening Phase
ticipants to the screening phase, we could only assume no
underlying difference between participants and non-par- A total of number of 11,684 SRS were distributed in phases
ticipants, which allowed us to calculate overall prevalence to the children attending Grades two or three in the 134
as a combination of the observed prevalences in the SEMR schools included in the survey and meeting age criteria for
and the GSS screen-positive groups, and a hypothesized inclusion. A total number of 4195 SRS were obtained from
prevalence in the GSS screen-negative children. In each parents (participation: 36.2 %), and 2761 from teachers.
simulation, a total sample was generated, and then subdi- Table 1 and Fig. 2 illustrate the distribution of scores on
vided into SEMR and GSS children; subgroup sizes and both parent and teacher SRS.
event count estimates for the SEMR group were generated Participation for teachers could not be estimated since
from a binomial distribution with a rounded event proba- we do not know how many parents passed to the teachers
bility based on the observed proportions. The GSS sub- of their child the second SRS that was provided to them. Of
group was further subdivided into the screen-positive the 2761 teacher SRS, 236 were received on children for
(C99th centile) and the screen-negative (\99th centile) whom parents did not respond themselves. A comparison
groups. Prevalence estimates for the screen positive group of mean SRS teacher scores showed no significant differ-
were generated using binomial distributions; estimates for ence (F = 0.59; p = .44) in scores between children with
the screen negative group were sampled from a uniform (N = 2525; mean = 41.4, SD = 27.9) and without
distribution spanning a reasonable range of potential (N = 236; mean: 39.9, SD = 29.3) an available parent
prevalences. There is no published data that allow us to SRS. On 2525 children with both a parent and a teacher
ascribe prevalence estimates for the proportion of ASD SRS, parent scores were significantly higher than teacher
children in a child population without any history of con- scores (mean difference: 3.03; SEM: 0.67; paired-sample
tact with services and screening results not in the top t test: t = 4.53, p \ .001). The correlation between the two
centile of a screening instrument. Moreover, this parameter scores was weak but significant (Pearson’s r = .18,
is likely to be population- and context-specific and no data p \ .001). Table 1 provides the mean scores for both
from Mexico was available to guide its calibration. Our
reviews (Hill et al. 2014) have shown that conservative Table 1 Screening phase results: total SRS scores by informant and
estimates for the overall prevalence of ASD are in the child gender
0.7–1.0 % range and closer to 1 % in school age samples. Informant All Boys Girls pa
It is reasonable to assume that at least 50 % of this
Parent N = 4195 N = 2074 N = 2120
prevalence proportion reflects ‘treated’ cases (children
already diagnosed and accessing services). Similarly, sur- Mean SD Mean SD Mean SD
vey data also suggest that some children with milder phe- 44.8 24.9 46.8 25.8 42.9 23.8 \.0001
notypes have never been diagnosed prior to their
participation into a study (Kim et al. 2011; CDC 2014). Teacher N = 2761 N = 1358 N = 1402
Thus, it may be inferred that this fraction of the overall Mean SD Mean SD Mean SD
prevalence is not nil. Accordingly, in our simulation, we
41.3 28.1 44.3 29.4 38.3 26.4 \.0001
allowed the hypothesized prevalence in the screen nega-
a
tives to fluctuate in the interval 0.1–0.5 %. In order to test Boys versus girls comparisons

123
J Autism Dev Disord (2016) 46:1669–1685 1677

9
Median=41.0 Mean 44.8
Median 41.0
8 SD 24.9

6
Sample Percent

0
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160

Parents Total Raw Scores

12
Median=34.0 Mean 41.3
Median 34.0
SD 28.1
10

8
Sample Percent

0
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160
Teachers Total Raw Scores

Fig. 2 Screening phase: distribution of total raw SRS scores, by informant

informants for the whole sample, and separately for each
gender. Boys scored consistently and significantly higher
than girls for both parents (mean difference: 3.8, t = 5.0;
p \ .001) and teachers (mean difference: 6.0, t = 5.7;
p \ .001) reports.
When we restricted the sample to children having both a
parent and teacher SRS, gender differences of the same
direction and magnitude were found with significant
(p \ .001) mean differences of 3.8 and 6.1, respectively.
When examined for each gender separately, parent and
teacher assessments were roughly comparable for boys
(46.4 vs 44.5; mean difference: 1.9; paired t test: 1.89,
p = .058) but differed for girls (42.5 vs 38.4; mean dif-
ference: 4.1; paired t test: 4.61, p \ .001).
Test–Retest Reliability
On subsamples of 16 parents and 12 teachers selected at
random during the screening phase, we obtained a second
SRS by the same informant within 2–4 weeks from the

123
1678
initial administration. In the parent subsample, the mean
SRS total score decreased slightly between the two
administrations, from 46.9 (SD = 26.44) to 42.2
(SD = 25.6), and the correlation between the total scores
at baseline and repeat administration yielded an intra-class
correlation coefficient of 0.76. For the 12 pairs of teacher
ratings, the scores decreased only slightly from 52.2
(SD = 27.0) to 50.4 (SD = 26.6), and the corresponding
intra-class correlation coefficient was 0.87. Pooling all
pairs of ratings together (N = 28), the mean SRS total
scores changed from 46.9 (SD = 26.4) to 42.2
(SD = 25.6), and yielded an intra-class correlation coeffi-
cient of 0.81.
Screen Positives
Screening data were arranged in a dataset where parent
SRS scores were assigned to each child surveyed with the
exception of the 236 children without a parental score for
whom the teacher value was instead selected; the total
sample of children with at least one SRS was N = 4431.
Children scoring in the higher percentile (score [ 99th
centile) were selected as screen positives. Limited resour-
ces did not allow to investigate separately children in
adjacent high risk centiles (e.g. 95th to 99th centiles);
similarly, we could not investigate a screen negative group
in low risk, lower centiles (\95th centile) which would
have been necessary to estimate the sensitivity of the
screening procedure.
Families of children in the top centile (N = 43) were
subsequently contacted by the research team and when
possible, invited for a diagnostic evaluation. Of these, 18
families could not be contacted or refused to participate.
The remaining 25 children were interviewed by a member
of the research team. There was no significant difference
(t = 0.73; df = 41; NS) between the mean SRS scores of
the 25 participants and the 18 non-participants (mean SRS:
122.2 vs 125.8, respectively) and no difference according
to gender (% male: 68 vs 66.7 %, respectively; v2 = 0.01;
df = 1; NS). This result suggested that participants in the
diagnostic phase were representative of all children scoring
at the top centile. When possible, we performed standard-
ized batteries of tests and those data were combined on the
LASI with results from the SRS, teacher and parent inter-
views, and direct observation.
SEMR Sample
Over the 2 years of the field work, the research team
members approached the medical and educational sites
holding records or having ongoing contact on children
meeting age and residence eligibility criteria (see ‘‘Mate-
rials and Methods’’ section above; Fig. 1). When the
123
J Autism Dev Disord (2016) 46:1669–1685
review of records or interviews with professional infor-
mants suggested that the child may have symptoms of
social impairment or an existing ASD diagnosis, data were
abstracted on the LASI. A total of 69 files were generated
during this process. The main data source for each of the 69
cases is shown in Fig. 1.
Final Case Determination and Interrater Agreement
A total of 94 LASIs (69 from the SEMR, 25 from the GSS)
were reviewed by the two clinicians of the research team
who evaluated independently if the child met or not criteria
for an ASD. ADOS was administered alone to 36 of the 94
participants (and in conjunction with the ADI for eight
subjects). Agreement occurred for 88 children (55 without
ASD, 33 with ASD). The interrater agreement between the
two assessors was very good (j = 0.865; 95 % CI
0.760–0.969). For the remaining six subjects, where
agreement had not been reached, files were reviewed by the
two lead researchers who concluded at the presence of an
ASD in three of the six children. Of the 25 children from
the GSS sample, five met criteria for an ASD after con-
sensus review by the team. Of the 69 files of the SEMR
sample, 31 children met criteria for ASD. Of the 36 con-
firmed ASD cases, 21 had been assessed with the ADOS
(three in the GSSS, 18 in the SEMR) and met criteria on
that instrument.
Cases Characteristics
The 36 children with ASD identified after expert clinical
review had a mean age: 8.1 years (SD = 0.27). The five
children from the GSS survey had a mean SRS score of
107.6. The 31 SEMR ASD children confirmed through
reviews of medical/educational records were identified at
all sources surveyed (see Fig. 1). Characteristics of the 36
children are summarized in Tables 2, 3 and 4.
Most subjects lived with their two parents (mean
paternal age: 37.8, SD = 6.8; mean maternal age: 34.5,
SD = 5.3). Consistent with other samples, the male:female
ratio was 4.1:1 (Table 2). The majority (88.9 %) of parents
noticed the first developmental anomalies before the 3rd
birthday of their child and although 61.2 % of parents had
a first contact with services before age three, a formal
diagnosis of ASD was not reached until age three in 77.8 %
of children, and even until age five in 22.2 % of them. Age
at diagnosis was significantly associated with the presence
of intellectual disability (ID) with 54.5 % of those with ID
diagnosed before age three compared to 4.0 % of those
without ID (v2 = 12.7; df = 2; p = .002). Almost all
subjects (96 %) subjects without ID were diagnosed after
age three (52 % between age three and five, and 44 % after
age five). In the presence of a co-occurring genetic
J Autism Dev Disord (2016) 46:1669–1685 1679

disorder, age at diagnosis was earlier with 66.7 % of those Table 3 Intellectual, medical and behavioral characteristics (N = 36
with and 15.2 % of those without a genetic disorder diag- children diagnosed with ASD)
nosed before age three although, due to limited statistical Characteristic N %
power, the difference fell short of significance (Fisher’s
Intellectual functioning
exact test: p = .09). Age at diagnosis was not associated
with gender (v2 = 3.05; df = 2; NS). The distribution of Normal IQ (C85) 12 33.3
DSM-IV diagnoses was typical of that reported in reviews Borderline IQ (70–85) 13 36.1
of population surveys (Fombonne 2009a, b), with Mild intellectual disability (C50 and \70) 4 11.1
PDDNOS being the most frequently used diagnosis, and Moderate/severe intellectual disability (\50) 7 19.4
autistic disorder being twice as frequent as Asperger dis- Language level
order. Diagnostic subtype was strongly associated with ID Fluent 10 27.8
(v2 = 11.65; df = 2; p = .003) with the rate of ID being Simple conversation 13 36.1
72.7 % in autistic disorder, 27.3 % in PDDNOS, and nil in Limited language 4 11.1
Asperger disorder. Non-verbal 9 25.0
Estimated level of cognitive functioning was in the Epilepsy 7 19.4
normal range for a third of the sample whereas another Associated genetic syndrome 3 8.3
third (30.5 %) had mild to severe intellectual disability Sensory deficit 2 5.6
(Table 3). The proportions of children with any intellectual Disruptive behavioral problems 25 69.4
disability (IQ \ 70) were 42.9 % in girls and 27.6 % in Emotional problems 16 44.4
boys although the difference was not statistically

Table 4 Case sample: treatment characteristics (N = 36)

Table 2 Case sample (N = 36): demographic, development and
diagnostic characteristics N %
N % Child neurologist ongoing care 25 69.4
Male gender 29 80.6 Currently on medication 15 41.7
Family Atypical neuroleptics 13 36.1
Lives with 2 parents 29 93.5 Stimulants 4 11.1
Education Antiepileptic 2 5.6
Secondary school or less 14 53.8 ABA program 8 22.2
Some college 4 15.4 Speech and language therapy 7 19.4
University level 8 30.8 Occupational Therapy 6 16.7
Age 1st symptoms noticed by parents Behavioral intervention 8 22.2
0–12 months 3 8.3 Individual counseling 4 11.1
12–18 months 4 11.1 Family counseling 3 8.3
18–24 months 9 25.0
24–36 months 16 44.4
[36 months 4 11.1
significant (Fisher’s exact test; NS). A fourth of the sample
Age 1st contact with services
had fluent language skills and another fourth was non-
0–24 months 7 19.4
verbal, with the remainder half of the sample showing
24–36 months 15 41.7
some limited language and conversational skills. A lifetime
[36 months 4 38.9
history of epilepsy was reported in about a fifth of the
Age formal diagnosis sample. There were three children with diagnosable genetic
0–24 months 1 2.8 disorders, including one boy with Down syndrome, one
24–36 months 7 19.4 boy with Fragile X disorder and one girl with Rett syn-
36–60 months 20 55.6 drome; all three had an autistic disorder diagnosis.
[60 months 8 22.2 There was a trend for non-verbal status to be associated
Diagnosis with the presence of ID (55.5 vs 22.2 %; Fisher’s exact
Autistic disorder 12 33.3 test: p = .09). Subjects with epilepsy tended to be more
Asperger 6 16.7 often non-verbal (57.1 vs 17.2 %; Fisher’s exact test:
PDDNOS 18 50.0 p = .05) and to have ID (57.1 vs 24.1 %; Fisher’s exact

123
1680
test: p = .17), but epilepsy was not associated with gender
(Fisher’s exact test: NS). There were high rates of behav-
ioral (aggression, impulsivity, attention deficits, irritability)
and emotional problems (anxiety, fears), and nine children
(25 %) endorsed both types of problems. The presence of
disruptive problems, of emotional problems or of the
combination of both problems was not associated with
gender, verbal status, or the presence of intellectual dis-
ability or of a genetic disorder (Fisher’s exact tests; all
p’s [ .20).
Over two-thirds of the sample was under the regular care
a child neurologist (Table 4). A total of 13 children were
taking either atypical neuroleptics (N = 13) or stimulants
(N = 4) for behavioral problems; the four children on
stimulants were on a combination of both medications. Use
of any of these two classes of medications was not asso-
ciated with gender, intellectual disability or the presence of
a genetic disorder (all p’s: [ .40).
Prevalence Estimation
We ran simulations with an approximate total sample size
of n = 12,200 per iteration. The SEMR and GSS sub-
group sizes were generated from a binomial distribution
with a rounded event probability (based on the observed
proportions) of 0.05 and 0.95, respectively. Within the
SEMR group, we randomly generated the prevalence of
ASD using a binomial distribution with event probability
matching the observed p = .07 (31/432). The GSS was
subdivided into the screen-positive (C99th centile) and
screen-negative (\99th centile) groups. We estimated the
screen-positive group prevalence using random draws
from a binomial series with p = .2, the observed pro-
portion (5/25) in this group. For the screen-negative
group, we hypothesized that the prevalence would be
somewhere between 0.001 and 0.005 (see ‘‘Materials and
Methods’’ section above) and varied the binomial proba-
bility in this range using a uniform distribution. This was
repeated with 10,000 random populations. Using this
approach, we estimated the prevalence of ASD to be
0.87 % [95 % CI (0.62, 1.1 %)]. Further calculations
indicated that 57.5 % of the prevalence proportion was
accounted for by the prevalence in the GSS sample
[0.50 %; 95 % CI (0.29, 0.72 %)] whereas 42.5 % of the
overall proportion was due to the prevalence in the SEMR
sample [0.37 %; 95 % CI (0.26 %, 0.49 %)]. When we
used two alternative ranges of possible values for the
screen negative hypothesised prevalence, the overall
estimate was 0.75 % [95 % CI (0.59 %, 0.91 %)] and
0.99 % [95 % CI (0.65, 1.4 %)] for the narrower and
broader intervals, respectively, a relative change of 14 %
in either direction.
123
J Autism Dev Disord (2016) 46:1669–1685
Discussion
This is the first epidemiological study conducted in Mexico
on autism spectrum disorders. Our estimate is in line with
best estimates of 0.6–1 % derived from recent reviews of
the epidemiological literature on ASD (Fombonne 2009a,
b; Elsabbagh et al. 2012; Hill et al. 2014). Our prevalence
is higher than that reported in two previous Latin American
studies and lower than a third one. The Argentina study
(Lejarraga et al. 2008) was primarily designed to assess a
screening instrument and referral process, and convenience
sampling in clinical settings together with low participation
rate and lack of autism specific diagnostic assessments
made the higher reported prevalence of 1.3 % difficult to
interpret. By contrast, Montiel-Nava and Peña (2008) in
Maracaibo county Venezuela, reported a prevalence pro-
portion of 0.17 % for ASD (0.11 % for autistic disorder,
and 0.06 % for Asperger disorder) in a population of
3-9 years old. In Brazil, a preliminary estimate of 0.27 %
was reported in a small survey of 1470 7–12 year old
children residing in one district of Atibaia, a South-Eastern
city (Paula et al. 2011). Both studies relied on case ascer-
tainment methods that mostly consisted to identify children
already diagnosed with an ASD; no specific screening
activity was deployed to review existing records of chil-
dren carrying other developmental or psychiatric diag-
noses, and no systematic survey of the general school
population was performed.
These differences in study design may have contributed
to the fourfold to fivefold difference in estimates. As well,
it is possible that public and professional awareness in the
countries surveyed previously may have been lower than
that in Mexico; our study benefited from having a local
specialized team (CLIMA) part of a national clinical net-
work with ASD expertise, the work of which in our local
area might have achieved positive changes in detection and
management of ASDs in Leon preceding our study.
The prevalence of ASD among US Hispanic children
has been estimated in successive CDC surveys (Pedersen
et al. 2012; CDC 2014) that provide a useful comparison to
our first Mexican estimation. In Arizona, the prevalence of
ASD in white non-Hispanic children at age 8 years rose
from 0.88 to 1.50 % in cohorts of children born between
1992 and 1998, whereas the prevalence in US Hispanic
children in the same survey rose from 0.27 to 0.79 %. The
trends show that the prevalence is consistently lower in US
Hispanic children compared to that in white non-Hispanic
children although over time the magnitude of this differ-
ence is attenuating. The same findings arose from analyses
of all CDC surveys. Thus, across all survey sites, the
prevalence of ASD at age eight among Hispanic children
rose from 0.37, 0.61, 0.79 to 1.08 % in the birth cohorts
J Autism Dev Disord (2016) 46:1669–1685
born in 1994, 1998, 2000 and 2002, respectively. The
prevalence in our study was estimated from children born
in 2003, and therefore, the most appropriate US Hispanic
comparison data is the 1.08 % figure derived from the 2002
birth cohort surveyed in 2010 (CDC 2014). As a compar-
ison, the prevalence figure for Black and White children in
that survey were 1.23 and 1.58 %, respectively. It is rec-
ognized that the lower figure for Hispanic children in the
US reflects poorer detection and access to diagnostic and
intervention services by these minority families (Pedersen
et al. 2012; CDC 2014). The true population prevalence is
therefore likely to be higher than 1.08 % amongst US
Hispanic children. To which extent our own finding of
0.87 % reflects a more pervasive difficulty in Mexico with
detection and access to services cannot be determined from
the data at hand, but it is a plausible explanation. However,
it is difficult to comment further on the US and Mexico
figures as the surveys differed in terms of methods, sample
size, and more importantly in terms of the underlying
organization of health and educational services.
The present survey generated data on a sample of cases
that should be representative of the underlying Mexican
population of ASD. Correlates of ASD in this study were
very similar to those known through previous surveys
(Fombonne 2003, 2009a). Thus, the family background
was unremarkable and the associations found with gender,
epilepsy, known genetic disorders, and comorbid behav-
ioral syndromes were consistent with previous knowledge.
The repartition between autistic disorder and other forms of
ASD (PDDNOS, Asperger disorder) in our study was
typical of the proportions reported across epidemiological
surveys of ASD (Fombonne 2009a, b); in the two previous
Latin American studies where it was reported, the pro-
portion of cases meeting strict criteria for autistic disorder
was 66.7 % (Montiel-Nava and Peña 2008) and 25 %
(Paula et al. 2011) whereas our figure was 33.3 %. This
variability reflects a lack of reliability in differentiating
diagnostic subtypes within the autism spectrum, providing
justification for adopting a unified concept of ASD fol-
lowing recent nosographical changes (American Psychi-
atric Association 2013). Interestingly, the degree of
associated intellectual disability (ID) (30.5 %) or of bor-
derline intelligence (36.1 %) was very similar to that
reported in recent surveys. For example, the rate of ID in
the latest CDC survey (2014) was 31 %, and 23 % of
children had borderline IQ (IQ 70–84). Of note, the pro-
portion of Hispanic children with ID in the CDC survey
was 38 %, slightly higher but comparable to that in our
sample, thereby suggesting that the predicament of His-
panic children with ASD is not different whether living in
the US or in Mexico.
Other important data were collected about the trajectory
of children with ASD. Although almost 90 % of parents
1681
noticed developmental atypicalities in their children before
the 3rd birthday, a formal diagnosis was attained only after
age three in 78 % of cases, and at age six or later in 22 %
of cases. These waiting periods are too protracted, resulting
both in unnecessary worries and stress in parents who are
waiting for answers and assistance and in lost opportunities
to intervene at an earlier age when maximal brain plasticity
permits more developmental gains to be achieved with
proper intervention (Fombonne 2009b; Dawson et al. 2010,
2012). Implications for developing systematic screening
programs in routine health care visits for Mexican toddlers
and for specific training of primary health care providers
should be drawn.
Several strengths and limitations of our study should be
kept in mind. Firstly, we used a methodology whereby
multiple sources, both from health and education services,
were screened to identify not only children having already
a diagnosis of ASD in their records but also those with
other diagnoses together with signs of social interaction
abnormality. In line with the CDC methodology (Van
Naarden Braun et al. 2007), casting such a broader net
increases the sensitivity of case ascertainment. Moreover,
to the best of our knowledge, this study is the second one to
include an additional ascertainment procedure in the form
of screening general education classrooms with an autism-
specific detection tool. Secondly, the reliability of assess-
ment tools and diagnostic procedures was very good both
for parent or teacher screening with the SRS and for final
case determination using clinical-diagnostic expert opin-
ions. Thirdly, in the screening phase, we obtained teacher
data on a subsample of children whose parents had not
participated. This allowed for a rare opportunity to test
whether or not non-respondent parents in the screening
phase of an autism survey systematically differed from
participants with respect to their child autism symptoma-
tology. We found no difference in SRS teacher scores
according to parent participation status, suggesting that
participation was not influenced by the presence or absence
of ASD features in the child. Fourthly, parent SRS scores
did not as well differ between screen positive participants
and non-participants in the diagnostic phase. In turn, these
results provided a rationale for ascribing equal weights to
participants and non-participants to diagnostic and
screening phases data for our prevalence calculations.
However, these positive features are mitigated by limita-
tions that may have jeopardized the internal validity of our
results. Firstly, by virtue of our study protocol, parents
agreed to allow teachers to complete the SRS. We have no
teacher scores for children whose parents completely opted
out of the survey. As a result, inferences about autism
symptoms in the majority of non-respondents still remain
uncertain. Secondly, our participation rate was relatively
low (36.2 %) and the potential bias in the prevalence could
123
1682
be substantial if our conservative assumptions on partici-
pation being independent of caseness were not verified,
even by a small amount. Similar issues have emerged in the
few surveys of ASD that included a general school
screening component. Thus, in the Korean study (Kim et al.
2011), non-participation occurred as well at different stages
of the study. The assumptions of equivalence between
participants and non-participants in the screening phase
may have contributed to overestimating the prevalence as
suggested in a recently published critical assessment of the
Korean study (Pantelis and Kennedy 2015). Thirdly, sur-
veys of relatively rare conditions that rely on a two-phase
design have infrequently included an assessment of an
appropriately sized sample of screen negative children. By
examining screen negatives, one can estimate the propor-
tion of false negatives and subsequently adjust the preva-
lence estimate based on the observed sensitivity of the
screens. Our initial plans to sample two groups of screen
negatives could not be executed due to limited resources.
This issue was addressed by assuming a non-zero preva-
lence among the screen negatives for which we assigned a
range of reasonable possible values. Our resulting estimate
of 0.87 % was robust and showed only a relatively minor
change when assumptions were varied in both directions.
However, we are fully aware that there was no empirically
derived knowledge to date to determine these parameters
with accuracy, and therefore our prevalence estimate
should be regarded as dependent upon the correctness of
the assumption. Replication surveys in other areas of
Mexico are therefore necessary to confirm our initial
results.
Additional limitations of the study need to be considered
with respect to the external validity of our study findings.
Although we do not anticipate that migration in or out of
the area would be a significant source of bias, we could not
track possible changes of residence before or during the
study period, and evaluate their differential association
with ASD. The survey was conducted at a single site and
generalization of results to the entire state of Guanajuato
and to other areas of Mexico should be prudent. However,
drawing a nationally representative sample of children is
not easy. Surveys of large representative samples of
households have been employed in ASD epidemiology (for
a recent example, see Zablotsky et al. 2015). However, the
merits of national representativeness are tempered by
methodological costs, including the lack of complete cov-
erage of the population (with homeless and institutional-
ized subjects being left out), response rates that rarely
surpass the 70–75 % mark, reliance on unvalidated par-
ental reports to measure diagnostic status, and exquisite
sensitivity of prevalence estimates to questionnaire design
and architecture (see Zablotsky et al. 2015 for a remarkable
example). Other surveys rely on multisite sampling to
123
J Autism Dev Disord (2016) 46:1669–1685
attain broader coverage of the underlying population.
Despite employing a uniform methodology across all
included study sites, a remarkable site-specific variability
in prevalence estimates is frequently observed. The CDC
surveys provide a striking illustration. For example, in the
latest ADDM-CDC survey (CDC 2014), the average
prevalence was 1.47 % but the state specific prevalence
ranged from a low of 0.57 % in Alabama to a high of
2.19 % in New-Jersey. Far from being interpreted as
reflecting true site differences in ASD prevalence, this 3.8
fold difference likely reflected differences in the sensitivity
of case ascertainment in states contrasting for their level of
awareness about ASD and development of service infras-
tructure. In this example, unless there are reasons to regard
the New-Jersey estimate as biased upward, it is clear that
the reported average prevalence estimate (1.47 %) is likely
to be an underestimate of the true population parameter. As
a consequence, investigators should weigh the risks and
advantages of relying on multisite or nationwide repre-
sentative sampling as opposed to concentrating resources
on one (or perhaps a few) well chosen areas where case
finding methods will be optimized as well as approaches to
validate case status. In Mexico, nationally or multisite
sampling would likely lead to downward bias in estimates
as seen in the US CDC surveys; nevertheless, our main
study finding is also unlikely to be free of bias and, as
stated above, replications are needed.
Consistent with prior investigations (Kim et al. 2011;
Avchen et al. 2011), our study confirms that in order to
achieve maximal sensitivity for case ascertainment a
comprehensive sampling approach that includes children
attending public schools is required. Two-phase surveys
have been used in mental health surveys in order to effi-
ciently identify cases in general population samples.
However, the limited experience of such designs in ASD
surveys has uncovered complex issues that will need res-
olution in future studies. Screening tools for ASD have
typically been developed following top-down approaches,
distilling clinical knowledge onto parent report question-
naires. Such screening tools may contain items not relevant
to the respondents and often require significant amounts of
time. The use of screeners raises similar issues with
teachers who may be overburdened by completing them on
entire classrooms. Moreover, it is unclear if and when
reliance on multiple informants should be preferred. Future
research should develop and test field methods for ASD
screening that are more respondent-friendly, quick and
efficient. For example, a three-phase survey comprising a
two-stage screening might be proposed that would use
teachers and parents in complementary sequences.
Screening out the majority of children could be achieved
initially with short questionnaires or semi-structured
interviews with teachers who usually know their pupils
J Autism Dev Disord (2016) 46:1669–1685
well and can calibrate their behavior against implicit
norms. Then, the limited proportion of children who dis-
play some unusual behavioral, learning or social charac-
teristic would enter a secondary screening targeting ASD
with parent-completed autism specific screens. Research is
needed to address these issues and develop novel adequate
tools and screening procedures.
There are several take home messages to our findings.
First, considering a prevalence of about 0.9 % and the
Mexico population size of 120 millions (Instituto Nacional
de Estadistica y Geografia 2015a, b) including 43.5 millions
of youths aged 0–19, it can be estimated that, in Mexico
today, there are about 94,800 subjects with ASD in the age
group 0–4 and an additional 298,000 subjects with ASD in
the age range 5–19. These two figures point, respectively, at
the magnitude of both early detection and diagnosis pro-
grams needed for preschoolers and of education and treat-
ment programs required at later ages for the Mexican youth
population. Counting the adult population would add many
more subjects in need of specialized social, medical and
vocational services. Service planners should bear in mind
the high individual and societal costs attached to ASD
across the lifespan (Buescher et al. 2014). Even though
initial implementation costs are high, early detection and
diagnostic programs and early intensive behavioral inter-
ventions are pivotal in improving developmental and health
trajectories in children with ASD and decreasing residual
impairments and disabilities, and thereby hold the promise
of reducing costs in the long run. Second, our study has
demonstrated that a methodology for surveying ASD can be
effectively deployed in the Mexican child population. We
validated a standardized ASD measure that can now be used
both for clinical and research purposes in the Hispanic
population, alongside existing Spanish versions of diag-
nostic tools such as the ADI-R and ADOS. We employed
case ascertainment methods that were comprehensive and
made no assumption on prior identification. Sampling was
inclusive of children with or without special needs or a
history of contact with an array of services reflecting the
local service infrastructure. The case confirmation was
attained using different approaches that were tailored to the
case ascertainment procedure, and ultimately driven by
expert clinical judgement with demonstrated reliability.
These study features may guide the design of future studies
in setting firm sampling and measurement principles that
need to be flexibly applied within each study context. As
mentioned above, complex issues remain to be addressed by
new research to improve ASD survey methodologies. Third,
the trajectories of children with ASD identified in this
survey suggest that delays in detection and diagnosis occur
and should be targeted by vigorous training and educational
programs of primary care providers, nurses, pre-school
educators and the public at large. The age of diagnosis has
1683
decreased in most countries in the last two decades but
identifying persisting factors associated with late diagnosis
remains a public health priority (Jo et al. 2015). Fourth, we
encourage an interpretation of our findings that includes a
full appreciation of the study limitations and of the uncer-
tainty attached to any population parameter estimate. In
particular, the point estimate of 0.87 % is likely to be sen-
sitive to various assumptions that we had to make for esti-
mating the prevalence proportion within the constraints of
our study design, and future studies aiming at replicating
these findings in the Mexican population are necessary to
shed light on the validity of this first result. Similarly, rather
than reifying the 0.87 % proportion (for example by
asserting that one child out of 115 has autism), we suggest
that the prevalence estimated in our study is always com-
municated alongside its confidence interval and that a range
of plausible prevalences be preferred over a static and
inflexible numerical fact. As a first step, the study provided
a first range of estimates and a study template that will
hopefully be applied and improved in larger scale studies
conducted in other states within Mexico and elsewhere.
Acknowledgments The study has been supported by a grant from
Autism Speaks (6424). The authors express their gratitude to Andy Shi
and Michael Rosanoff (Autism Speaks) who were instrumental in
facilitating the planning and execution of the study. We thank Dr
Kuri, Chairman of the Health System Surveillance in Mexico City,
the Educational authorities of Guanajuato state and of the city of
Leon, the city hall of Leon and the Mayor’s office, the child neu-
rologists, the school teachers, the director and staff at DIF and
CLIMA, especially Norma Alicia Manero Tinoco. We express our
special gratitude to those families and their children who participated
to the study and made it possible. The PI Eric Fombonne was at
McGill University when the grant was awarded. Ruth Bruno was
funded through the Canada Research Chair in Child Psychiatry
awarded to the PI. Katrina Ramsay and Benjamin Nealy were sup-
ported by Oregon Clinical and Translational Research Institute
(OCTRI), through a Grant UL1TR000128 from the National Center
for Advancing Translational Sciences (NCATS) at the National
Institutes of Health (NIH).
Author Contributions Eric Fombonne and Carlos Marcin designed
and funded the study, organized the data collection and wrote the
manuscript. Eric Fombonne, Ruth Bruno, Katrina Ramsay and Ben-
jamin Nealy performed data and statistical analyses. Ana Cecilia
Manero, Christian Diaz and Michele Villalobos collected data. All
authors reviewed and approved the manuscript.
References
American Psychiatric Association. (2013). Diagnostic and statistical
manual of mental disorders (5th ed.). Arlington, VA: American
Psychiatric Publishing.
American Psychiatric Association (APA). (2000). Diagnostic and
statistical manual of mental disorders (4th ed.). Washington,
DC: APA.
American Psychiatric Association (APA). (1994). Diagnostic and
statistical manual of mental disorders (4th ed.). (DSM-IV).
Washington, DC: APA.
123
1684
Avchen, R. N., Wiggins, L. D., Devine, O., Braun, K. V. N., Rice, C.,
Hobson, N. C., et al. (2011). Evaluation of a records-review
surveillance system used to determine the prevalence of autism
spectrum disorders. Journal of Autism and Developmental
Disorders, 41(2), 227–236.
Buescher, A. V., Cidav, Z., Knapp, M., & Mandell, D. S. (2014).
Costs of autism spectrum disorders in the United Kingdom and
the United States. JAMA Pediatrics, 168(8), 721–728.
CDC. (2007a). Autism and Developmental Disabilities Monitoring
Network surveillance year 2000 principal investigators; Centers
for Disease Control and Prevention. Prevalence of autism
spectrum disorders—Autism and Developmental Disabilities
Monitoring Network, six sites, United States, 2000. MMWR
Surveillance Summery, 6(1), 1–11.
CDC. (2007b). autism and Developmental Disabilities Monitoring
Network surveillance year 2002 principal investigators; Centers
for Disease Control and Prevention. Prevalence of autism
spectrum disorders—Autism and Developmental Disabilities
Monitoring Network, 14 sites, United States, 2002. MMWR
Surveillance Summery, 56(1), 12–28.
CDC. (2009). Autism and Developmental Disabilities Monitoring
Network surveillance year 2006 principal investigators; Centers
for Disease Control and Prevention (CDC). Prevalence of autism
spectrum disorders—Autism and Developmental Disabilities
Monitoring Network, United States, 2006. MMWR Surveillance
Summery, 58(10), 1–20. (Erratum in: MMWR Surveill Summ.
2010 Aug 6;59(30):956).
CDC. (2012). Autism and Developmental Disabilities Monitoring
Network surveillance year 2008 principal investigators; Centers
for Disease Control and Prevention. Prevalence of autism
spectrum disorders–Autism and Developmental Disabilities
Monitoring Network, 14 sites, United States, 2008. MMWR
Surveillance Summery, 61(3), 1–19.
CDC. (2014). Developmental Disabilities Monitoring Network
surveillance year 2010 principal investigators; Centers for
Disease Control and Prevention (CDC). Prevalence of autism
spectrum disorder among children aged 8 years—Autism and
Developmental Disabilities Monitoring Network, 11 sites,
United States, 2010. MMWR Surveillance Summery, 63(2), 1–21.
Chen, J. A., Peñagarikano, O., Belgard, T. G., Swarup, V., &
Geschwind, D. H. (2015). The emerging picture of autism
spectrum disorder: Genetics and pathology. Annual Review of
Pathology: Mechanisms of Disease, 10, 111–144.
Constantino, J. N., & Todd, R. D. (2003). Autistic traits in the general
population: A twin study. Archives of General Psychiatry, 60(5),
524–530.
Dawson, G., Jones, E. J., Merkle, K., Venema, K., Lowy, R., Faja, S.,
et al. (2012). Early behavioral intervention is associated with
normalized brain activity in young children with autism. Journal
of the American Academy of Child and Adolescent Psychiatry,
51(11), 1150–1159.
Dawson, G., Rogers, S., Munson, J., Smith, M., Winter, J., Greenson,
J., et al. (2010). Randomized, controlled trial of an intervention
for toddlers with autism: The Early Start Denver Model.
Pediatrics, 125(1), e17–e23.
Durkin, M. S., Bilder, D. A., Pettygrove, S., & Zahorodny, W. (2015).
The validity and usefulness of public health surveillance of
autism spectrum disorder. Autism, 19(1), 118–119.
Elsabbagh, M., Divan, G., Koh, Y. J., Kim, Y. S., Kauchali, S.,
Marcı́n, C., et al. (2012). Global prevalence of autism and other
pervasive developmental disorders. Autism Research, 5(3),
160–179.
Fombonne, E. (2003). Epidemiological surveys of autism and other
pervasive developmental disorders: An update. Journal of
Autism and Developmental Disorders, 33(4), 365–381.
123
J Autism Dev Disord (2016) 46:1669–1685
Fombonne, E. (2009a). Epidemiology of pervasive developmental
disorders. Pediatric Research, 65(6), 591–598.
Fombonne, E. (2009b). A wrinkle in time: From early signs to a
diagnosis of autism. Journal of the American Academy of Child
and Adolescent Psychiatry, 48(5), 463–464.
Fombonne, E., Marcin, C., Bruno, R., Manero, C., & Marquez, C. D.
(2012). Screening for autism in Mexico. Autism Research, 5(3),
180–189.
Hill, A. P., Zuckerman, K., & Fombonne, E. (2014). Epidemiology of
autism spectrum disorders. In F. R. Volkmar, S. J. Rogers, R.
Paul, & K. A. Pelphrey (Eds.), Handbook of autism and
pervasive developmental disorders. Diagnosis, development,
and brain mechanisms (4th ed., Vol. 1, pp. 57–96). New York:
Wiley.
Instituto Nacional de Estadistica y Geografia. (2015a). Censos
económicos 2014: Resultados definitivos, Julio Results from
2010 census. http://www.inegi.org.mx/est/contenidos/Proyectos/
ce/ce2014/doc/presentacion/pprd_ce2014.pdf.
Instituto Nacional de Estadistica y Geografia. (2015b). Results from
2010 census. http://www.inegi.org.mx/default.aspx.
Jo, H., Schieve, L. A., Rice, C. E., Yeargin-Allsopp, M., Tian, L. H.,
Blumberg, S. J., et al. (2015). Age at autism spectrum disorder
(ASD) diagnosis by race, ethnicity, and primary household
language among children with special health care needs, United
States, 2009–2010. Maternal and Child Health Journal, 19(8),
1687–1697.
Kim, Y. S., Leventhal, B. L., Koh, Y. J., Fombonne, E., Laska, E.,
Lim, E. C., et al. (2011). Prevalence of autism spectrum disorder
in a total population sample. American Journal of Psychiatry,
168(9), 904–912.
Korkman, M., Kirk, U., & Kemp, S. (1998). NEPSY: A developmental
neuropsychological assessment. San Antonio, TX: Psychological
Corporation.
Lecavalier, L., & Mandell, D. (2015). Autism developmental
disabilities monitoring network surveillance: A reply to Drs
Durkin, Bilder, Pettygrove, and Zahorodny. Autism, 19(1),
120–121.
Lejarraga, H., Menendez, A. M., Menzano, E., Guerra, L., Biancato,
S., Pianelli, P., et al. (2008). Screening for developmental
problems at primary care level: A field programme in San Isidro,
Argentina. Paediatrics and Perinatatal Epidemioliogy, 22(2),
180–187.
Lord, C., Risi, S., Lambrecht, L., Cook, E., Leventhal, B., DiLavore,
P., et al. (2000). The autism diagnostic observation schedule-
generic: A standard measure of social and communication
deficits associated with the spectrum of autism. Journal of
Autism and Developmental Disorders, 30(3), 205–223.
Lord, C., Rutter, M., DiLavore, P., et al. (2002). The autism
diagnostic observation scale (ADOS). Los Angeles, CA:
Western Psychological Services.
Mandell, D., & Lecavalier, L. (2014). Should we believe the Centers
for Disease Control and Prevention’s autism spectrum disorder
prevalence estimates? Autism, 18(5), 482–484.
Montiel-Nava, C., & Peña, J. A. (2008). Epidemiological findings of
pervasive developmental disorders in a Venezuelan study.
Autism, 12(2), 191–202.
Newschaffer, C. J., et al. (2015). Regarding Mandell and Lecavalier’s
editorial ‘‘should we believe the Centers for Disease Control and
Prevention’s autism spectrum disorders prevalence estimates’’
and subsequent exchange with Durkin. Autism, 19(4), 505–507.
Pantelis, P. C., & Kennedy, D. P. (2015). Estimation of the prevalence
of autism spectrum disorder in South Korea, revisited. Autism
[Epub ahead of print].
Paula, C. S., Ribeiro, S. H., Fombonne, E., & Mercadante, M. T.
(2011). Brief report: Prevalence of pervasive developmental
J Autism Dev Disord (2016) 46:1669–1685
disorder in Brazil: A pilot study. Journal of Autism and
Developmental Disorders, 41(12), 1738–1742.
Pedersen, A., Pettygrove, S., Meaney, F. J., Mancilla, K., Gotschall,
K., Kessler, D. B., et al. (2012). Prevalence of autism spectrum
disorders in Hispanic and non-Hispanic white children. Pedi-
atrics, 129(3), e629–e635. doi:10.1542/peds.2011-1145. Epub
2012 Feb 20.
Pelly, L., Vardy, C., Fernandez, B., Newhook, L. A., & Chafe, R.
(2015). Incidence and cohort prevalence for autism spectrum
disorders in the Avalon Peninsula, Newfoundland and Labrador.
CMAJ Open, 3(3), E276–E280.
R Development Core Team. (2008). R: A language and environment
for statistical computing. R Foundation for Statistical Comput-
ing, Vienna, Austria. ISBN:3-900051-07-0. http://www.R-pro
ject.org.
Randall, M., Sciberras, E., Brignell, A., Ihsen, E., Efron, D.,
Dissanayake, C., et al. (2015). Autism spectrum disorder:
Presentation and prevalence in a nationally representative
Australian sample. The Australian and New Zealand Journal
of Psychiatry [Epub ahead of print].
Riviere, A. (2002). IDEA—Inventory of autism spectrum. Buenos
Aires: FUNDEC, Fundacion para el Estudio del Desarrollo
Cognitivo.
Rutter, M., LeCouteur, A., & Lord, C. (2003). The autism diagnostic
interview-revised (ADI-R). Los Angeles, CA: Western Psycho-
logical Services.
Shopler, E., Reichler, R. J., & Rochen, B. (2010). CARS 2—
Childhood autism rating scale (2nd ed.). Torrance: WPS
Western Psychological Services.
1685
Sparrow, S. S., Cicchetti, D. V., & Balla, D. A. (2005). Vineland
adaptive behavior scales: Second edition, survey interview
form/caregiver rating form. Livonia, MN: Pearson Assessments.
Van Naarden Braun, K., Pettygrove, S., Daniels, J., Miller, L.,
Nicholas, J., Baio, J., et al. (2007). Evaluation of a methodology
for a collaborative multiple source surveillance network for
autism spectrum disorders—Autism and Developmental Disabil-
ities Monitoring Network, 14 sites, United States, 2002. MMWR
Surveillance Summery, 56(1), 29–40.
Weschler, D. (2004). Weschler intelligence scale for children (4th
ed.). Toronto, ON: Harcourt Assessment. WISC IV - Escala de
inteligencia Wechesler en Español, Manual Moderno 2007
copyright The Psychological Corporation 2003.
Wiggins, L. D., Baio, J., & Rice, C. (2006). Examination of the time
between first evaluation and first autism spectrum diagnosis in a
population-based sample. Journal of Developmental and Behav-
ioral Pediatrics, 27(2 Suppl), S79–S87.
Willsey, A. J., & State, M. W. (2015). Autism spectrum disorders:
From genes to neurobiology. Current Opinion in Neurobiology,
30, 92–99. Epub 2014 Nov 28.
Yeargin-Allsopp, M., Rice, C., Karapurkar, T., Doernberg, N., Boyle,
C., & Murphy, C. (2003). Prevalence of autism in a US
metropolitan area. JAMA, 289(1), 49–55.
Zablotsky, B., Black, L. I., Maenner, M. J., Schieve, L. A., Blumberg,
S. J. (2015). Estimated prevalence of autism and other devel-
opmental disabilities following questionnaire changes in the
2014 National Health Interview Survey. National Health Statis-
tics Reports, 13(87), 1–20.
123