Psychology
2012. Vol.3, No.11, 997-1003
Published Online November 2012 in SciRes (http://www.SciRP.org/journal/psych) http://dx.doi.org/10.4236/psych.2012.311150
In Search of Prediction Factors for Autism Spectrum Disorders:
An Impossible Task?
René Pry1, Arne F. Petersen2, Amaria Baghdadli3
1Université Paul Valéry & CHU-Montpellier, Montpellier, France
2Centre de Ressource Autism, CHU-Montpellier, Montpellier, France
3Child & Adolescent Psychiatry, CHU-Montpellier, Montpellier, France
Received September 26th, 2012; revised October 21st, 2012; accepted November 16th, 2012
Clinical work and prediction of development are closely linked in the practice of early detection, diagno-
sis and choice of modes of intervention in young children with autism. Variables are often defined in
terms of risk factors or of development, and may refer to general or specific phenomena. The purpose of
this paper was, using a generalized mixed model, to test ways of measuring development and its predic-
tion regarding joint attention (that is to say, response to and initiation of joint attention) in children with
autism. Over a period of one year, seventy-seven children were followed from the age of four and a half
years upwards. The results show that it is possible to identify general risk factors, but much more difficult
to pinpoint specific factors. In our current state of knowledge, prediction can only be of a global nature
and therefore requires the use of general markers.
Keywords: Autism; Joint Attention; Prediction; GLMM; Risk Factor; General Marker
Introduction
A major part of the research and clinical work on children
with pervasive developmental disorders is devoted to the ques-
tion of prediction. Prediction is already implied in the early
detection of the syndrome, but it becomes a special develop-
mental problem once the diagnosis has been established, and it
is finally evoked when treatment strategies are chosen. So far
experience has shown that within these three domains the
prognosis is not only uncertain but virtually impossible.
Possible Prediction Factors—An Overview
Prediction factors are usually considered in two perspectives:
1) From a developmental (temporal) viewpoint, opposing risk
factors and developmental factors; and 2) From a clinical
viewpoint which distinguishes global factors and specific fac-
tors.
1) Developmental perspective. Risk factors may be sub-
divided into primary, secondary, tertiary etc. factors—a classi-
fication which divides the pre-diagnostic period (for autism the
first 30 months) into periods during which new competences
known as normal development are expected to occur. Devel-
opmental factors are generally evoked from the time of diagno-
sis, which is also the starting point for most longitudinal stud-
ies.
2) Clinical perspective. Global or general markers for au-
tism relate to the diagnosis, which is itself an outcome of com-
posite evaluations like the ADI (Lord, Rutter, & Le Couteur,
1994, 2010), ADOS (Lord, Risi, Lambrecht, Cook, Leventhal,
DiLavore, Pickles, & Rutter, 2000, 2010), or refer to the se-
verity of the disturbance, which is also a result of composite
assessments (CARS: Schopler, Reichler, DeVellis, & Daly,
1980), and other indices of gravity, IQ, as well as general
markers for socio-adaptability like those in Daily Living Skills
(DLS) of the Vineland Scale (Sparrow, Balla, & Cicchetti,
1984). These global scores possess emergent qualities which
transcend the sub-scores they consist of (thus the whole does
not equal the sum of its parts). Being real synthetic scores they
express a clinical reality, that of the autistic syndrome. Specific
or elementary markers, which refer to certain competences
(such as synchronous imitation, joint attention, identification of
facial expressions, prosodic modulation, vocabulary, etc.), are
unfortunately more often than not coded in terms of “present/
absent”, or marked “positive/negative”, even though their de-
velopment may start discretely as precursor indices for global
markers (Nadel, 2009).
Here it is noteworthy that, if we attribute a certain predictive
value to these different factors, naturally we also have to as-
cribe to them a certain stability over time, or consider them as
stable characteristics for individuals with autism.
By crossing the two main classifications we obtain a combi-
nation of prediction factors from which it is possible to articu-
late the developmental and clinical perspectives (Table 1).
But before addressing this question it is appropriate to note
that various constraints complicate the very identification of the
different factors. Among these constraints we shall consider the
following: definition of the disorder (a complex and multifari-
ous phenomenon), its changing prevalence, the question of
time-independence of the prediction factors, and the choice of
statistical methodology.
A Global Definition
For the whole group of pervasive developmental deficits
(PDD) the World Health Organisation’s ICD-10 classification
(1994) is the most recommended. PDD includes various clinical

R. PRY ET AL.
Table 1.
Factors and variables used in prediction studies.
Factor Elementary Global
Explanatory variables Risk
Social smile
Reaction to call-name
Joint attention
Fantasy play
Productive lexicon
Genetic load
Neonatal history
Sensorial reactivity
Tonico-postural regulation
Spontaneous motor activity
Developmental Special treatment program
Delay/specific hyper competence
Diagnostic markers
Associated disorders
Intensity of the treatment
Intellectual level
Variables to be explained Prediction Linguistic activity Interactive competences
Behavioural disorders
Clinical development
Psycho-social development
Everyday autonomy
phenomena, which may be divided into seven categories (infan-
tile autism, atypical, Rett, disintegrative, hyperkinetic with
mental retardation, Asperger and others) or in a dimensional
respect (the notion of Autism Spectrum Disorders: ASD).
These two approaches cover the same clinical reality: they
identify qualitative and simultaneous impairments, reciprocal
social interactions, modalities of communication, as well as a
repertoire of interests and limited stereotyped, repetitive activi-
ties. This combination, which founds the notion of “disorder”,
cannot be merely fortuitous. As a consequence, this very asso-
ciation translates itself in reality by a multitude of formulations,
in terms of intensity or predominance of one impairment over
another, or again by a more or less discrete presence of diag-
nostic criteria (age, …).
It is evident that this phenomenon, being multifarious indeed,
may easily be the phenotypic expression of the same disorder
with different severity or signal several independent deficits.
Besides the ICD, which is an ongoing classification, foresees
that within one and the same category new deficits may appear,
and that in future editions other categories may be formed.
Weak Prevalence and Prediction Factors That
Change with Time
For a long time pervasive developmental disorders were con-
sidered for a long time considered to be a statistically rare phe-
nomenon (4 in 10,000). Today, however, the prevalence is 6 to
7 for 1000 individuals under 20 years of age, and for autism 2
to 4 for 1000 individuals (Charman, 2002; Fombonne, 2009).
The accuracy of identifying of risk factors depends on the
prevalence in so far as a weak prevalence produces a large
number of “false positives”. This phenomenon can be illus-
trated by some working data which we now have at our dis-
posal in a precocity survey for which there is high specificity
(=.98), but low sensitivity and a large number of “false posi-
tives” (=.38) as in the “Check-List for Autism in Toddlers”
(Baron-Cohen, Wheelwright, Cox, Baird, Charman, Swetten-
ham, Drew, & Doehring, 2001) and in the “Modified Check-
List for Autism in Toddlers” (Robins, Fein, Barton, & Green,
2001; Baghdadli, 2005).
While, in some cases, predicting a given illness or disorder
may not be too difficult, since the risk factors have been easily
identified (linked, say, to genes, nutritional problems, radiation
exposure), it is quite another matter with pervasive develop-
mental disorders for which many factors will be in interaction
and probably of a genetic, neurological or environmental na-
ture.
Since the ways in which the risk factors are expressed change
with the child’s development and only become somewhat stabi-
lised when the child approaches 36 months of age, it seems
only possible to pronounce a reasonable diagnosis at this time
of maturation. By this we mean that the general markers, which
are present from early on, undergo differentiation and seem-
ingly end up “coinciding” with the characteristics of the disor-
der. For practical reasons it may be helpful to break down and
analyse the risk factors in view of a temporal organisation for
normal development during the sensorimotor period: 1) The
neonatal and perinatal phase; 2) The 2-month-old phase (with
social smile, intentionality and interplay with a partner); 3) A
phase at nine months (when triadic competences and joint at-
tention appear); and 4) A phase at eighteen months (when the
symbol function emerges) (Rochat, 2001). For each of these
phases the particularities of children with PDD may then be
noted, such as missing or non-developed abilities, regression or
temporary disappearance of skills, etc.
The only risk factors for PDD bearing upon the neo- and
perinatal phase, which have been identified so far, are those of
sex-ratio (four times more common in boys than in girls), the
age of the parents (the risk is multiplied by 1.3 for mothers over
35 years of age and by 1.4 for fathers aged over 40), the pres-
ence of another affected child among the siblings (4% higher
risk if the affected child is a boy, and 7% higher risk if it is a
girl). The risk strongly increases (25% to 30%) if the family al-
ready includes two children with PDD, and the syndrome con-
cordance between monozygote twin boys varies from 70% to
90%. Finally, it seems that incidences of pre- and perinatal
antecedents are more common in individuals with PDD than in
the population as a whole.
Retrospective studies, especially those that include analyses
of family films, have also revealed unusual behavioural traits,
first and foremost in spontaneous motricity (Rogers & Beneto,
2002; Fournier, Hass, Naik, Lodha, & Cauraugh, 2010), but
also in variations of tonus and peculiarities in sensory and at-
tention processing. However, these markers remain very gen-
eral and not functionally specific, even though they may derive
from a particular sensorimotor functioning, and at present we
do not dispose of enough relevant statistical data on prevalence,
sensitivity and specificity. Results of longitudinal studies, from
the age of 12 months, on brothers and sisters, born after a sib-
ling with autism, attach a tentative predictive value to the ab-
sence or rareness of social smile, eye contact, and orientation
998

R. PRY ET AL.
when being called (Yirmiya, Gamliel, Pilowsky, Feldman,
Baron-Cohen, & Sigman, 2006).
On the other hand, research based on questionnaires has
identified tertiary risks between 18 and 24 months such as pas-
sivity and low level of reactivity to social stimuli, difficulties
with visual following and joint attention, retarded expressive
language (in particular, absence of proto-declarative and de-
monstrative gestures), no real fantasy play.
Stability and Instability of Developmental Factors
Even though the number of interacting factors of develop-
ment is considerable, it has nevertheless been possible to clas-
sify variables such as the conditions of appearance and preco-
ciousness of the disorders, their importance in terms of symp-
tom intensity, the presence of associated troubles (intellectual
deficiency, epilepsy, somatic pathologies), and environmental
factors. Among the latter, certain modalities of care and support
seem to be of importance, although, even today, it is difficult to
evaluate the exact impact of treatment on development; this
being said, some results lead us to think that precocity, intensity
of the proposed stimulation, and structural adaptation of the
environment may be of positive influence (Montreuil & Mage-
rotte, 1994; Recordon-Gaboriaud, 2009).
In fact, the factors that best account for socio-cognitive de-
velopment in autism are general markers derived from devel-
opment quotients, e.g. IQ, established by evaluations on differ-
ent levels of development regarding adaptive functions assessed
at the time of diagnosis. These general markers are composite
and come from very heterogeneous developmental domains
(motor, cognitive, social, linguistic, etc.). In every second child
with autism, the scores of the intelligence test increase signifi-
cantly with age (Pry, Juhel, Bodet, & Baghdadli, 2007). More-
over, between childhood and adulthood a tendency towards a
decrease in the “non-verbal” component can be observed in
tempo with a moderate increase in the “verbal” component
(Seltzer, Shattuck, Abbeduto, & Greenberg, 2004; Mawhood,
Howlin, & Rutter, 2000). It should also be noted that in chil-
dren with “IQ-extremes” like <50 and >100 one observes a
more disadvantageous prediction than in the population with
IQ’s < 100 and >50. Likewise, the presence of language in
production—even when starting very late—is always a positive
prognostic factor. However, as mentioned before, the only
variable which shows some stability is the diagnosis, but this
remains unchanged for life in 80% - 96% of the cases studied
(Seltzer et al., 2004).
Finally, we should not forget that often the choice of variable
to be explained also is of a general nature (“good”/“bad”) or the
variable is described in a categorical manner from criteria such
as attainment of everyday autonomy. Studies in which the ob-
jective is the onset of specific competences are much rarer.
Statistical Constraints
Three major classes of model are employed for making the
prediction: linear regression and generalised linear models
(logistic regression and Poisson regression). Linear regression
leads to a generalised notion of correlation and of t-test; logis-
tics regression results in a generalisation of the notion of odds-
ratio and the Chi-2 test. A third class of model appears (which
is the object of this paper), namely the generalised linear model
mixed with repeated measures (GLMM) (Breslow & Clayton,
1993; Fitzmaurice, Laird, & Ware, 2004).
In linear regression the nature of the explicative variables
(qualitative dichotomies) lend themselves very well to tech-
niques like ANCOVA. In this kind of approach the “regression
diagnosis” is important: the variable to be explained must be
quantifiable, the residua ei must follow a normal law and be
independent, the explicative variables must not be redundant,
and the robustness of the model must be tested (ruling out the
extreme cases). In practice, and more precisely in developmen-
tal psychopathology, these conditions are rarely united, and we
often see situations of multi-linearity between prediction fac-
tors.
In logistic regression the variables may be qualitative (in
general binary), multi-nominal (when there are more than two
classes), or enumerative (e.g. number of hours of treatment) as
in the Poisson regression model. These models are sufficiently
adapted to the complex problem of prediction for linking up a
diagnosis or a development (binary variable) with a group of
risk factors with which one tries to characterise the respective
weights. The strength of such links may be expressed in terms
of excess risk, relative risk, “odds ratio”, or attributive risk.
The characteristics of “regression diagnosis” are the inde-
pendence of residues, the detection of “likely” subjects, at the
origin of important variations (robustness of the model) and the
search for multi-linearity.
Generalised linear models mixed with repeated measures are
techniques with a theory of high content and much in use: these
techniques are applied in areas as diverse as forestry, medicine,
finance, economy, industry and so forth. They are most inter-
esting since they can be used to analyse diverse effects and
repeated measurements, and they are ready for use with differ-
ent probability laws with a view to modelling errors, and even
more so as their distributions belong to the exponential family
(which is often found in developmental psychology).
On the basis of all these remarks one may ask how the inter-
actions between the different classes of variables take place
over a given period (does a predictive variable at a given time
explain the same percentage of variance later on?)—i.e. the
developmental perspective—and to question whether the data
collected at the moment of the diagnosis are informative with
respect to a specific competence—i.e. the clinical perspective.
Methodology: Population Characteristics,
Developmental Factors and Target Variables
The present study was coordinated by the Languedoc-
Roussillon Autism Resource Centre at the CHU in Montpellier.
Seventy-seven children with autism were examined three times
at intervals of 6 months: T1, T2 (T1 + 6 months) and T3 (T1 +
12 months).
The diagnoses were worked out on the basis of pluridiscipli-
nary, clinical observations guided by the standardised version
of international classifications of mental and behavioural dis-
orders (ICD-10, WHO, 1994) and by using the “Revised Au-
tism Diagnostic Interview” (Lord, Rutter, & Le Couteur, 1994),
ADOS (Lord, Risi, Lambrecht, Cook, Leventhal, DiLavore,
Pickles, & Rutter, 2000) and the “Childhood Autism Rating
Scale” (Schopler, Reichler, DeVellis, & Daly, 1980). Children
with pronounced motor delay on levels less than 18 months
were not included, as the aspects of retardation, by increasing
the prediction, may hide the variability of the rest of the sam-
ple.

R. PRY ET AL.
1000
A descriptive analysis of the population of 77 children was
carried out by calculating the frequency of the quantitative
variables, the median and the interquartiles (75% and 25%) for
the qualitative variables (their distribution not always being
Gaussian, the normality of the distributions was analysed with
the help of the Shapiro-Wilks Test).
The sample presented two diagnostic categories: 87% (67) of
the children suffered from infantile autism and 13% (10) from
atypical autism. It included 66 boys (86%) and only 11 girls
(14%): thus the sex-ratio is 6/1 which was slightly above the
ratio found in epidemiological studies, whose only inclusion
criterion is the diagnosis of autism or that of other PDDs. In
return this result is in agreement with those obtained in popula-
tions with autism without mental retardation. The median age at
the beginning of the observation period was 52 months with an
interval included between 35 months for the youngest child and
60 months for the oldest child and an interquartile 25/75 be-
tween 48 and 59 months. The level of expressive language was
assessed with item 19 of the ADI-R: (0 = production of phrases;
1 = fewer than 50 words; 2 = fewer than 5 words). The two
main forms of motor development (global and fine) were ex-
plored with the two revised Brunet-Lézine subscales: oculo-
manual coordination and postural development. Regarding
motor development, the median level of oculo-manual coordi-
nation was 20 months with an interquartile included between 18
and 24.5 months, and the median postural level was evaluated
as being around 24 months with an interquartile included be-
tween 20 and 30 months. These data are presented in Table 2.
The levels of socio-adaptive development were assessed by
the Vineland Scale or VABS (Sparrow, Balla, & Cicchetti,
1984). This scale evaluates the children’s adaptation level in the
functional domains of communication, socialisation, autonomy
in everyday life, and motor development. The American norms
of the Vineland Adaptive Behaviour Scale also apply to the
French population (Fombonne & Achard, 1993; Pry, Guillain,
& Foxonet, 1996), and there is a specific standardisation for
sub-populations with autism (Carter, Volkmar, Sparrow, Wang,
Lord, & Dawson, 1998; Freeman, Delhomme, Guthrie, &
Zhang, 1999).
The variable to be explained was joint attention, which was
assessed by using the ECSP-Scale (Guidetti & Tourette, 1993),
adapted to the Early Social Communication Scale (Seibert &
Hogan, 1982). This last scale evaluates the development of
skills for establishing shared attention to the same object, per-
son, event or topic. Two kinds of reactions were studied: 1)
Response to joint attention. The aim here was to describe the
Table 2.
Characteristics of the sample at T1.
Variable N
77 % 100
Diagnosis Typical
Atypical 67 10
87
13
Age (months) Median
25% - 75% percentile
52
48 - 59
Age at detection* (months) Median
25% - 75% percentile
16
8.75 - 24
Sex Male
Female
66
11
85.7
14.3
SPC** Low Average
High
13
23
41
16.9
29.9
53.2
Median
25% - 75% percentile
24
18 - 31
CARS
ADI-R
Median
25% - 75% percentile
Communication (≥8)***
Social interactions (≥10)
Few interests (≤3)
Anomalies before 36 months (≥1)
33
26.5 - 36.5
69
71
65
74
93
96
88
100
ADOS-G
Module 1
Module 2
N
Social interactions: m (sd)
Repeated behaviour
N
Social interactions: m (sd)
Repeated behaviour
57
15.08 (4.8)
3.67 (1.58)
20
13.92 (4.1)
2.67 (1.4)
74
26
Expressive la nguage:
Item 19 ADI
Visual-manual coordinatio n****
Postural level
Good: Phrases: N (%)
Middle: >5 words
Poor: <5 words
Median
25% - 75% percentile
Median
25% - 75% percentile
20
20
37
20
18 - 24.5
24
20 - 30
26
26
48
*Age at which the first disturbances were recognized; **Parents’ socio-professional category: Low (workers, farm workers); Average (middle class, employees, farmers);
High (industrial, business, and intellectual professions, heads of company or business); ***Algorithm threshold; ****BL-R.

R. PRY ET AL.
development of the child’s understanding of the attempts by an
adult to direct his attention towards an object and his abilities to
understand and follow the adult’s indication. Thus, the right
response to most items was to look at and touch, to point to or
take hold of what the adult had been attending to. To name and
comment on objects in reply to an adult’s questions appear
when the child starts talking on the higher level in a series of 5
levels: simple, complex, conventional gestures, conventional
verbal and symbolic utterances. 2) Initiation of joint attention.
The aim of this observation series was, firstly, to describe the
child’s growing awareness that an adult may attend to the same
object or event as he himself is attending to, and, secondly, to
describe the development of means which a child may use to
direct an adult’s visual attention to an object or event and make
the adult attend to what he is himself just looking at. Such at-
tempts, which must appear spontaneously, can also be divided
into 5 levels.
In children with typical development the joint attention skill
may be located between 6 and 12 months implying an interac-
tion with others. It reveals itself in pointing and checking the
direction of the other’s glance, and among the precursors of
these abilities is the capacity to detect the direction and target
of another’s glance. In children with autism this initial capacity
is much retarded, and when they start walking the deficits they
manifest with joint attention are remarked the most (Baron-
Cohen, 1989; Jones & Carr, 2004; Nichols, 2005; Mundy,
2007). The response and initiation of joint attention are associ-
ated with the emergence of receptive and expressive language
and deserve particular interest since their later development
turns out to be a predictive variable in itself. The scores of the
Vineland Scale (in months) and those of joint attention, distrib-
uted according to the three observation periods, are shown in
Table 3.
Results
A technique for generalised linear models mixed with re-
peated measurements was used for data processing (Statistica,
v.9). The regression was carried out in an exhaustive search for
the best model and by crossing two criteria: the standardised
AKAIKE information criterion (AIC) and the Bayesian de
Schwarz-criterion. The explicative variables introduced into the
model were as follows: the category of diagnosis, age at T1,
age when the disorder was discovered, the parents’ socio-pro-
fessional category (SPC), sex, intensity of the disorder at T1,
initial language level at T1, and the levels of adaptive behav-
iour at T1, T2, and T3. The values of the two explicative vari-
ables, at the three times, were also introduced into the model,
and a descending procedure of a step-by-step nature was then
undergone.
Regarding the response to joint attention, the prediction values
brought about by the most satisfactory model (~ <5 words + >5
words. Autism + CARS + Phrases. T2 + Phrases. T3 +
(1/NUM)) are shown in Table 4. The model for the variable to
be predicted is not linear but of the Poisson family (log link)
(see Table 4). The variables which seem to have a positive in-
Table 3.
Median development of adaptive behaviour and joint attention (Interquartiles Q25-Q75).
T1 T2 T3
Communication 18* 19 24
(14 - 22) (15 - 30) (16 - 35.5)
Autonomy 27 29 31
(21 - 32) (21 - 35) (24 - 39)
Socialisation 19 22 25
(12 - 25) (16 - 35) (15.5 - 39)
Motor behaviour 32 37 41
(24 - 39) (29 - 43) (32 - 51)
Response to joint attention 8** 10 11
(36 - 14) (5 - 16) (5 - 17)
Initiation of j oint attention 5 6 6
(4 - 8) (4 - 9) (4 - 10)
*Developmental age; **Raw score.
Table 4.
Prediction and behavioural response to joint attention.
Estimation
Standard errors Value of z Pr(>/z/) Significance
Order at onset 3.85 .24 15.79 <2e – 16 .000
<5 words –.99 .15 –6.74 1.62e – 11 .000
>5 words*Autism –.65 .14 –4.74 2.21e – 06 .000
Intensity –.04 .01 –4.48 4.40e – 06 .000
<5 words*Tps2 .27 .11 2.44 .015 .05
<5 mots*Tps3 .24 .11 2.23 .02 .05
AIC: 359; BIC: 383.1; LogLik: –172.5; Deviance: 345.

R. PRY ET AL.
Table 5.
Prediction and behavioural initiation of joint attention.
Estimation Standard errors Value of z Pr(>/z/) Significance
Order at onset 2.79 .15 18.2 <2e – 16 .000
<5 words –.69 .08 –8.08 6.68e – 16 .000
>5 words*Autism –.46 .09 –5.28 1.27e – 07 .000
Intensity –.02 .00 –3.35 –.81e – 5 .000
AIC: 193.8; BIC: 211; LogLik: –91.9; Deviance: 183.8.
fluence on the development of joint attention are the following:
absence of language, alone or associated with the diagnosis of
typical autism (versus atypical autism), initial intensity of the
disorder, and, finally, a time effect which may be considered as
a factor uniting maturation with life experience. Three variables
(<5 words, Autism diagnosis and intensity) seem to restrain
development of joint attention (aggravating factors). On the
other hand, for children with an elaborate, initial linguistic ac-
tivity (presence of phrase production), we noted a significant
improvement in this skill at T2 and T3, whereas the improve-
ment between T1 and T2 was not significant (protection fac-
tors).
Regarding the abilities for initiating joint attention, the pre-
diction values brought about by the most satisfactory model
(~ <5 words + >5 words. Autism + CARS + (1/NUM)) are
shown in Table 5. The only variables retained turned out to
have a significantly negative effect on the development of ini-
tiation of joint attention (aggravating factors). These were: no
language production, autism diagnosis when accompanied by
language beginning (lexicon of >5 words), and initial intensity
of the disorder. On the other hand, time did not seem to have an
improving influence on the performances and no variables ap-
peared which could have played a positive role (protection
factors) appeared.
In these two versions of joint attention—the first, to under-
stand and respond to the initiation of shared behaviour, and the
second, to initiate the very same behaviour oneself—neither the
adaptive behaviour of communication activities and DLS, nor
the motor behaviour had any effect upon the development of
the two facets of this skill. It is appropriate to remember (see
Table 2) that these behaviours become complex with time (the
differences in performance between T3 and T1 are significant
in the four domains), and that for most of them it depends on
the behaviour being learnt. We may speculate that the con-
straints on these two developments—adaptive behaviour and
joint attention, and the mechanisms at their basis—will be of a
different nature, at least at this age and in children with autism.
Discussion
Several questions arise concerning the interpretation of these
results. Among the factors retained, the majority of them appear
to be risk or aggravating factors registered clinically as negative
signs which forecast a delay or “non-appearance” of one or
more skills: no productive language, most severe diagnosis,
high disorder intensity. Only one of these factors is really spe-
cific for autism: that of the “typical autism” diagnosis; the other
factors may be found in intellectual deficiency and/or in spe-
cific language disorders. Moreover, they all accompany global
developmental delay. They are, however, extremely general
markers and none of them belong in any specific way to the
chain of development that leads to the complex coordination
constituting joint attention (awareness of self, imitation, visual
face-perception, selective attention, sharing, etc.) Yet they en-
dorse the liaison between linguistic activity and joint attention.
Among the other variables introduced into the model, such as
sex, levels of communication, socialisation or everyday adapta-
tion, of which we might expect a somewhat close affiliation
with the target variable, none of them had any predictive char-
acter. Thus we must admit that whatever methodological so-
phistication is used, we are left with extreme generalities such
as: an overall delay “produces” certain specific delays and/or is
predictive of specific delays, except for small groups of chil-
dren with certain language competences.
Perhaps these kinds of delay, like the qualitative impairments
in development are, at the same time, the characteristics and the
specificity of the autistic disturbance at the time of diagnosis. It
is perhaps also the period in the development of the individual
when the disturbance is the most pervasive, as it simultaneously
affects major functions like communication and socialisation,
and limits the taste for novelty—and, in so doing, limits the
possibilities of prediction.
Conclusion
Should we then conclude that any prediction in autism is
impossible? That the current formulation of the disorder, which
is of a behavioural nature, is an epistemological obstacle for all
prognostic activity, since the pathology is complex and devel-
opmental? Perhaps the normal approach, which consists of
searching for precursor elements that may define PDD (limita-
tion of interest, motor expressivity, proto-language), is not the
best solution? It remains possible that today the very general
markers are the best synthesis of these characteristics to come.
The counterpart of this attitude is that we should not reduce
autism spectrum disorder to a mere formulation of deficits.
After all, autism is an original development, a queer construc-
tion, with astonishing ways of processing information, whether
social or not, which may also lead to hyper competences.
REFERENCES
Baghdadli, A. (2005). Recommandations pour la pratique profession-
nelle du diagnostic de l’autisme. Fédération française de Psychiatrie.
Haute Autorité de Santé, Paris, Saint-Denis la Plaine: FFP; HAS.
Baron-Cohen, S. (1989). Joint attention deficits in autism: Towards a
cognitive analysis. Development and Psychopathology, 1, 185-189.
doi:10.1017/S0954579400000377
Baron-Cohen, S., Wheelwright, S., Cox A., Baird, G., Charman, T.,
Swettenham, J., Drew, A., & Doehring, P. (2000). Early identifica-
tion of autism by the Checklist for Autism in Toddlers (CHAT).
Journal of the Royal Society of Medicine, 93, 521-525.
Breslow, N. E., & Clayton, D. G. (1993). Approximate inference in
1002

R. PRY ET AL.
generalized linear mixed models. Journal of the American Statistical
Association, 88, 9-25. doi:10.2307/2290687
Carter, A. S., Volkmar, F. R., Sparrow, S. S., Wang, J. J., Lord, C.,
Dawson, G. et al. (1998). The vineland adaptive behavior scales:
Supplementary norms for individuals with autism. Journal of Autism
and Developmental Diso r ders, 28, 287-302.
doi:10.1023/A:1026056518470
Charman, T. (2002). The prevalence of autism spectrum disorders.
Recent evidence and future challenges. European Child and Adoles-
cence Psychiatry, 11, 249-256. doi:10.1007/s00787-002-0297-8
Esbensen, A. J., Seltzer, M. M., Lam, K. S., & Bodfish, J. W. (2009).
Age-related differences in repetitive behaviors in autism spectrum
Disorders. Journal of Autism and Developmental Disorders, 39, 57-
66. doi:10.1007/s10803-008-0599-x
Fitzmaurice, G. M., Laird, N. M., & Ware, J. H. (2004). Applied longi-
tudinal analysis. Hoboken, NJ: Wiley-Interscience.
Fombonne, E. (2009). Epidemiology of pervasive developmental dis-
orders. Pediatric Research, 65, 591-598.
doi:10.1203/PDR.0b013e31819e7203
Fombonne, E., & Achard, S. (1993). The vineland adaptive behavior
scale in a sample of normal french children: A research note. Journal
of Child Psychology and Psychiatry, 34, 1051-1058.
doi:10.1111/j.1469-7610.1993.tb01108.x
Fournier, K. A., Hass, C. J., Naik, S. K., Lodha, N., & Cauraugh, J. H.
(2010). Motor coordination in autism spectrum disorders: A synthe-
sis and meta-analysis. Journal of Autism and Developmental Disor-
ders, 40, 1227-1240. doi:10.1007/s10803-010-0981-3
Freeman, B., Delhomme, M., Guthrie, D., & Zhang, F. (1999). Vine-
land adaptive behaviour scale scores as a function of age and initial
iq in 210 autistic children. Journal of Autism and Developmental
Disorders, 29, 379-384. doi:10.1023/A:1023078827457
Guidetti, M., & Tourrette, C. (1993). Evaluation de la communication
sociale précoce ecsp. Issy-Les-Moulineaux: Editions Scientifiques et
Psychologiques.
Jones, E. A., & Carr, E. G. (2004). Joint attention in children with au-
tism: Theory and intervention. Focus on Autism and Other Develop-
mental Disabilities, 19, 13-26. doi:10.1177/10883576040190010301
Lawer, L., Brusilovsliy, E., Salzer, M. S., & Mandell, D. S. (2009). Use
the vocational rehabilitative services among adults with autism.
Journal of Autism and Developmental Disorders, 39 , 487-494.
doi:10.1007/s10803-008-0649-4
Lord, C., Risi, S., Lambrecht, L., Cook Jr., E. H., Leventhal, B. L.,
DiLavore, P. C., Pickles, A., & Rutter, M. (2000). The autism diag-
nostic observation schedule-generic: A standard measure of social
and communicative deficits associated with the spectrum of autism.
Journal of Autism Developmental Disorders, 30, 205-223.
Lord, C., Rutter, M., & Le Couteur, A. (1994). Autism diagnostic inter-
view revised: A revised version of a diagnostic interview for care-
givers of individuals with possible pervasive developmental disor
ders. Journal of Autism and Developmental Disorders, 24, 659-685.
Mawhood, L., Howlin, P., & Rutter, M. (2003). Autism and develop-
mental receptive language disorder: a comparative follow-up in early
adult life. I: Cognitive and language outcomes. Journal of Child Psy-
chology and Psychiatry, 41, 547-559. doi:10.1111/1469-7610.00642
Montreuil, N., & Magerotte, G. (1994). Pratique de l’intervention indi-
vidualisée. Bruxelles: De Boeck.
Mundy, P. (2007). Attention, joint attention, and social cognition. Cur-
rent Directions in Psych ological Science, 16, 269-274.
doi:10.1111/j.1467-8721.2007.00518.x
Nadel, J. (2009). Où est le développement dans le syndrome develop-
mental d’autisme? Enfance, 1, 101-108.
doi:10.4074/S0013754509001128
Nichols, K. E., Fox, N., & Mundy, P. (2005). Joint attention, self-rec-
ognition, and neurogognitive function in todlers. Infancy, 7, 35-51.
doi:10.1207/s15327078in0701_4
Organisation Mondiale de la Santé (1994). CIM 10—Classification
internationale des troubles mentaux et des troubles du comportement:
Descriptions cliniques et directives pour le diagnostic. Paris: Mas-
son.
Pry, R., & Aussilloux, C. (2000). Le childhoodautism rating scales
(cars) chez l’enfant autiste jeune: Analyse des items, étude des traits
latents, validité concourante et généralisabilité. Psychologie et psy-
chométrie, 21, 33-47.
Pry, R., Bodet J., Pernon, E., Aussilloux, C., & Bagdadli, A. (2007).
Initial characteristics of psychological development and evolution of
the young autistic child. Journal of Autism and Developmental Dis-
orders, 37, 341-353. doi:10.1007/s10803-006-0161-7
Pry, R., Guillain, A., & Foxonet, C. (1996). Adaptation sociale et com-
pétences sociocognitives chez l’enfant de 4-5 ans. Enfance, 3, 315-
329. doi:10.3406/enfan.1996.3030
Recordon-Gaboriaud, S. (2009). Les prises en charge de l’autisme,
pratiques et recherches en France. Pratiques en Santé Mentale, 55,
11-17.
Robins, D. L., Fein, D., Barton, M. L., & Green, J. A. (2001). The
modified checklist for autism in toddlers: An initial study investigat-
ing the early detection of autism and pervasive developmental disor-
ders. Journal of Autism and Developmental Disorders, 31, 131-144.
doi:10.1023/A:1010738829569
Rochat, P. (2001). The infant’s world. Cambridge: Harward University
Press.
Rogers, S., & Benetto, L. (2002). Le fonctionnement moteur dans le cas
de l’autisme. Enfance, 1, 63-73. doi:10.3917/enf.541.0063
Schopler, E., Reichler, R. J., DeVellis, R. F., & Daly, K. (1980). To-
ward objective classification of childhood autism: Childhood Autism
Rating Scale (CARS). Journal of Autism and Developmental Disor-
ders, 10, 91-103. doi:10.1007/BF02408436
Seltzer, M. M., Shattuck, P, Abbetudo, L., & Greenberg, J. S. (2004).
Trajectory of development in adolescents and adults with autism.
Mental Retardation Developmental Disabilities Research Review, 10,
234-247. doi:10.1002/mrdd.20038
Sparrow, S. S., Balla, D. A., & Cicchetti, D. V. (1984). Vineland adap-
tive behavior scale. Circle Pines, MN: American Guidance Service.
Yirmiya, N., Gamliel. I., Pilowsky, T., Feldam, R., Baron-Cohen, S., &
Sigman, M. (2006). The development of siblings of children with au-
tism at 4 and 14 months: Social engagement, communication, and
cognition. Journal of Child Psychology and Child Psychiatry and
Applied Disciplines, 47, 511-523.
doi:10.1111/j.1469-7610.2005.01528.x