Original Article

Using Artificial Intelligence to Identify Factors Associated

with Autism Spectrum Disorder in Adolescents with
Cerebral Palsy
Carlo M. Bertoncelli1,2 Paola Altamura3 Edgar Ramos Vieira4 Domenico Bertoncelli5 Federico Solla1

1 Department of Pediatric Orthopaedic Surgery, Lenval University Address for correspondence Prof. Carlo M. Bertoncelli,, PhD, MSc
Pediatric Hospital of Nice, Nice, France Psych, PT, Hôpital pour Enfants–E.E.A.P. H. Germain., 337 Chemin
2 EEAP H. Germain, Department of Physical Therapy, Fondation Saint Antoine de Ginestiere. 06200 Nice, France
Lenval–Children Hospital, Nice, France (e-mail: bertoncelli@unice.fr).
3 Department of Medicinal Chemistry and Pharmaceutical
Technology, University of Chieti, Chieti, Italy
4 Department of Physical Therapy, Florida International University,
Miami, Florida, United States
5 Department of Information Engineering, Computer Science and
Mathematics, University of L’Aquila, L’Aquila, Italy

Neuropediatrics

Downloaded by: Université Paris Sud XI. Copyrighted material.

Abstract Autism spectrum disorder (ASD) is common in adolescents with cerebral palsy (CP) and
there is a lack of studies applying artificial intelligence to investigate this field and this
population in particular. The aim of this study is to develop and test a predictive
learning model to identify factors associated with ASD in adolescents with CP. This was
a multicenter controlled cohort study of 102 adolescents with CP (61 males, 41
females; mean age
SD [standard deviation] ¼ 16.6
1.2 years; range: 12–18
years). Data on etiology, diagnosis, spasticity, epilepsy, clinical history, communication
abilities, behaviors, intellectual disability, motor skills, and eating and drinking abilities
were collected between 2005 and 2015. Statistical analysis included Fisher’s exact test
and multiple logistic regressions to identify factors associated with ASD. A predictive
learning model was implemented to identify factors associated with ASD. The guide-
lines of the “transparent reporting of a multivariable prediction model for individual
prognosis or diagnosis” (TRIPOD) statement were followed. Type of spasticity (hemi-
plegia > diplegia > tri/quadriplegia; OR [odds ratio] ¼ 1.76, SE [standard error]
¼ 0.2785, p ¼ 0.04), communication disorders (OR ¼ 7.442, SE ¼ 0.59,
p < 0.001), intellectual disability (OR ¼ 2.27, SE ¼ 0.43, p ¼ 0.05), feeding abilities
(OR ¼ 0.35, SE ¼ 0.35, p ¼ 0.002), and motor function (OR ¼ 0.59, SE ¼ 0.22,
Keywords p ¼ 0.01) were significantly associated with ASD. The best average prediction model
► prediction model score for accuracy, specificity, and sensitivity was 75%. Motor skills, feeding abilities,
► cerebral palsy type of spasticity, intellectual disability, and communication disorders were associated
► statistics with ASD. The prediction model was able to adequately identify adolescents at risk of
► machine learning ASD.

received © Georg Thieme Verlag KG DOI https://doi.org/

October 5, 2018 Stuttgart · New York 10.1055/s-0039-1685525.
accepted after revision ISSN 0174-304X.
February 20, 2019
Predictors of Autism in Cerebral Palsy Bertoncelli et al.
Introduction
Cerebral palsy (CP) is a group of nonprogressive disorders of
motor and postural control that occur secondarily to brain
damage during early stages of development. Clinical mani-
festations include impaired movement, abnormal posture,
involuntary movements, and gait abnormalities.1 Children
with CP may also present associated Autism spectrum dis-
orders (ASD) characterized by impairments in communica-
tion, social interaction, and repetitive behavior.2,3 ASD is a
chronic and debilitating neuropsychiatric disorder that
includes autistic disorder, Asperger’s syndrome, and perva-
sive developmental disorder not otherwise specified.4
Cognitive impairment, speech difficulties, hearing and
vision disturbances, difficulty in learning, intellectual disabil-
ities (ID), and epilepsy are common in children with CP and
develop depending on the cause and topography of brain
damage.5 Early evaluation is strongly recommended for chil-
dren with CP.6 However, there is a lack of studies assessing
relationships between cognitive status, ASD, and motor func-
tioning.7 Statistical analysis of patients with CP and various
comorbidities is complex; just a few studies have done it to
better understand intellectual functioning, and they only pro-
vided percentages by ID level.6,7 Furthermore, statistical ana-
lysis was conducted on a limited number of participants, and no
information on the accuracy of predictions was provided.8
Neurological and psychiatric research has entered the age
of “Big Data”. Datasets now routinely involve thousands of
heterogeneous variables, including clinical, neuroimaging,
genomic measures. Statistical learning-based models are a
natural extension of classical statistical approaches and
provide more effective methods to analyze very large data-
sets.9 Artificial intelligence (AI) and the underlying methods
of machine learning (ML) and neuronal networks have made
dramatic progress in recent years and allow computers to
surpass human performance in domains that used to be
thought of as uniquely human as in neurology and psychia-
try.10 ML is a contemporary branch of statistics and AI
adapted for analysis of complex data to find patterns and
make predictions or infer new knowledge. ML algorithms
combine many features to construct an optimized equation
to predict outcomes with high accuracy. Supervised ML
models are suited to solve classification problems where
the classification of each patient is known a priori; samples
are used to train the model to classify other samples.11 The
nature of ML algorithms and AI can be fully harnessed for
predicting the onset of mental illness. Such applications will
benefit the society by serving as a monitoring tool for
individuals with altered behaviors.12 ML approaches for
clinical psychology and psychiatry focus on learning statis-
tical functions from multidimensional datasets to make
generalizable predictions. This approach is important for
practice given its potential to augment the accuracy of
diagnosis, prognosis, and treatment of people suffering
from mental illness using clinical and biological data.13
Neurologists and investigators now have an unprece-
dented opportunity to benefit from understanding complex
patterns in brain, behavior, and genes using methods from
Neuropediatrics
ML (e.g., support vector machines, modern neural-network
algorithms, cross-validation procedures).14 However, studies
concerning the use of ML in ASD diagnosis and treatment
suffer from conceptual, implementation, and data are sues,
such as the way diagnostic codes are used, the type of feature
selection employed, the evaluation measures chosen, and
class imbalances in data among others. Rigorous develop-
ment of a ML-based diagnosis for ASD is needed.15
The domain of our study is AI-based diagnostics. We aim
to introduce clinicians and researchers to the opportunities
in bringing machine intelligence into psychiatric practice
and in clinical studies. Recently, we developed and validated
a clinical prediction ML-model for neuromuscular scolio-
sis16,17 and factors associated with intellectual disability and
impaired adaptive functioning in children with CP.18 The
purpose of this study was to develop and test a supervised
predictive ML-model to identify factors associated with ASD
in adolescents with CP and assess potential associations
between ASD and physical impairments.
Downloaded by: Université Paris Sud XI. Copyrighted material.
Methods
This was a multicenter, double-blinded, controlled cohort
study.
Subjects
A total of 486 adolescents with CP in Nice region (France)
were screened based on the following inclusion criteria, age
between 12 and 18 years, being spastic, dystonic, having
mixed spastic/dystonic, or hypotonic CP classified using the
Surveillance of Cerebral Palsy in Europe system,19 and having
at least 3 years of follow-up. The exclusion criteria were
progressive encephalopathy or spinal neuropathology. One
hundred two (61 males, 41 females; 64% Whites, 30% Arabs,
4% Blacks, and 2% Asians) adolescents met the inclusion
criteria. Sixty inpatients from the Pediatric University Hos-
pital, Lenval, and 42 outpatients from the Day Hospital. The
mean age was 16.5
1.2 years (range: 12–18 years). Mean
follow-up time was 6.3
1.2 years (range: 3–12 years). The
initial diagnose of ASD was made by a child neurologist when
the patients were 5 years old.20
Ethics
All procedures were performed in accordance with the
ethical standards of the institutional research committee
and with the 1964 Helsinki Declaration and its later amend-
ments. All participants and parents consented to participate.
The data were anonymized and analyzed according to the
requirements of reference method 003, which number is
“2017728 v 0.” Ethics committee approval and informed
consent was registered with number “2017728 v 0–MR003
(reference method 003) “.
Measures
Data on etiology, diagnosis, functional assessments, type of
spasticity, intellectual disability, epilepsy, and clinical his-
tory were collected between 2005 and 2015 from the
 Predictors of Autism in Cerebral Palsy Bertoncelli et al.

medical records by members of the multidisciplinary team Intellectual Disability

which included neuropsychiatrists, pediatricians, neurolo-
gists, orthopedic surgeons, physiotherapists, child psychol-
ogists, and epidemiologists. Data analysis began in
December 2015. Narrative notes were coded and entered
into the electronic database “PredictMed”16,17.
Functional Assessments
The following functional assessments16,17 used are described
in ►Table 1: eating and drinking ability classification system
(EDACS,) communication function classification system
(CFCS), supports intensity scale for children (SIS-C): medical
and behavioral of individuals with intellectual and related
developmental disabilities’, manual ability classification sys-
tem (MACS), and the gross motor function classification
system (GMFCS).
Etiology
CP etiology was classified as antenatal (genetic, cerebral
malformation, infection, or vascular), perinatal (anoxic,
ID severity was assessed by a child psychiatrist based on the
DSM-5 (Diagnostic and Statistical Manual of Mental Disor-
ders) after 2013 and DSM-4 before 2013 (American Psychia-
tric Association, 2013), and the Wechsler’s intelligence scale
for Children (WISC).18,21 ID was classified as “mild,” “mod-
erate “, “severe,” or “profound.” Severity levels are based on
adaptive functioning, and not IQ (intelligence quotient)
scores because it is adaptive functioning that determines
the level of support required. Individuals with mild ID can
read with difficulty and are less likely to have associated
medical conditions than those with severe or profound ID.21
Individuals with moderate ID require moderate self-care
support, can travel to familiar places in their community,
and learn basic skills related to safety and health.21 Indivi-
duals with severe ID have major delays in development, often
have physical limitations and limited communication skills,
but often have the ability to understand speech. Despite
being able to learn simple daily routines and engage in
simple self-care, individuals with severe ID need supervision

Downloaded by: Université Paris Sud XI. Copyrighted material.

ischemic, or infectious), or postnatal (cranial trauma, infec- in social settings and often need family care. They cannot live
tious, epilepsy, or postnatal anoxic/ischemic injury).16,17 independently, and often live in supervised settings as a

Table 1 List of assessment tools, skills evaluated and function levels of ratings

Assessment tools Skills evaluated Levels of ratings

1 Gross motor Motor function on the I. Can walk, climb stairs, running and jumping. Has decreased speed,
function basis of self-initiated balance and coordination.
classification movement abilities II. Can climb stairs with a railing. Has only minimal ability to run or jump.
system (GMFCS) III. Walks with assistive mobility devices. May propel a manual wheelchair
and need assistance.
IV. Walking ability severely limited. Uses wheelchairs most of the time.
Need assistance.
V. Has physical impairments in all areas of motor function. Cannot sit or
stand independently.
2 Manual ability Manual ability with I. Handles objects easily and successful.
classification objects during II. Handles most objects but with somewhat reduced quality and/or speed
system (MACS) activities of of achievement.
daily living III. Handles objects with difficulty; needs help to prepare and/or modify
activities.
IV. Handles a limited selection of easily managed objects in adapted
situation.
V. Does not handle objects and has severely limited ability to perform even
simple action.
3 Eating and drinking Feeding ability using I. Eats and drinks safely and efficiently.
ability classification the key features of II. Eats and drinks safely but with some limitations to efficiency.
system (EDACS) safety and efficiency III. Eats and drinks with some limitations to safety; maybe limitations to
efficiency.
IV. Eats and drinks with significant limitations to safety.
V. Unable to eat or drink safely–tube feeding may be considered to provide
nutrition.
4 Communication Functional I. Effective sender and receiver with unfamiliar and familiar partners.
function performance II. Effective but slower paced sender and/or receiver with unfamiliar and/or
classification of communication familiar partners.
system (CFCS) in daily life III. Effective sender and receiver with familiar partners.
IV. Inconsistent sender and/or receiver with familiar partners.
V. Seldom effective sender and receiver even with familiar partners.
5 Supports intensity Medical an behavioral Medical (0/36): respiratory care, feeding assistance, skink care, other
scale (SIS) support needs exceptional medical care.
Behavioral (0/30): externally directed, self-directed, sexual behavior, other
exceptional behaviors

Neuropediatrics
Predictors of Autism in Cerebral Palsy Bertoncelli et al.

group homes.21 These individuals and they require close
supervision and help with self-care activities. Individuals
with profound ID require round-the-clock support and care.
They depend on others for all aspects of day-to-day life, have
major physical limitations, and extremely limited commu-
nication ability.
Neurologic Status
Neurologic status was classified according to the anatomy of
the spastic disorder (hemiplegia, diplegia, or tri/quadriplegia),
the presence of hypertonia in the upper or lower limbs, the
presence of dystonia, and the severity of epilepsy. Spasticity
was quantified using the Bohannon and Smith modified Ash-
worth’s scale and the modified Tardieu’s scale.22 Severity of
epilepsy was determined by the pediatric neurologists and
identified as “well controlled” or “intractable”23 accordingly
with the International League against Epilepsy, which defines
intractable epilepsy as continued seizures despite adequate
trials of at least two appropriate antiepileptic agents.24 Long-
term psychotropic medication (antipsychotics and/or antide-
have ASD if one of the diagnoses of category F84 of the
international classification of diseases, 10th revision (ICD-
10) was reported in medical records.25
Statistical Analysis
The guidelines of the “Transparent Reporting of a multi-
variable prediction model for Individual Prognosis or Diag-
nosis” (TRIPOD; ►Appendix 1) statement were followed.26
All data were analyzed in anonymous form and entered into a
database including demographics, functional diagnosis, neu-
rologic, and cognitive assessments (►Table 2).
First step: to generate the predictive model, we divided our
cohort into two groups: with and without ASD. The dependent
binary variable was ASD (yes or no), and the 10 independent
variables were etiology (ET), type of spasticity (SP), intellectual
disability (ID) dystonia (D), epilepsy (EP), sex (SE), EDACS,
CFCS, GMFCS, and psychotropic drugs (PS). Our aim was to find
which combinations (tuples) of independent variables best
predicted ASD. Initial analysis included using contingency
tables and Fisher’s exact test to identify associated factors,

Downloaded by: Université Paris Sud XI. Copyrighted material.

pressants) use was reported by the neuropsychiatrist.
Finally, ASD was determined by a child neurologist. In the
first stage, probable participants were determined by direct
observation, autism behavior checklist score, and medical
reports. In the second stage, those with “probable” symp-
toms underwent psychiatric examination (independent
reviewer) and their autistic symptoms were scored on the
childhood autism rating scale.20 Children were considered to
confidence intervals, and distribution frequencies for ADS
versus etiologic, functional, and neurologic features.27 The
univariate analysis was done using OpenEpi software28 and
the MedCalc statistical software.29
Second step: we generated all possible combinations
(tuples) from the 10 independent variables based on Com-
binatorics theorems,29 and got 2^10–1 ¼ 1,023 tuples (each
tuple having from 1 to 10 elements).

Table 2 Characteristics of participants in relation to autism spectrum disorder (ASD)

Patients’ Characteristics With ADS Without ADS Total

Patients, n (%) 331 (30) 71 (70) 102 (100)
Male 22 (22) 39 (78) 61 (60)
Female 9 (9) 32 (91) 41 (40)
Average age (y), mean, (SD) 15.4 (1.4) 15.7 (1.3) 16.6 (1.4)
Spasticity, n (%) 20 (20) 60 (80) 80 (78)
Hemiplegia, n (%) 2 (2) 5 (5) 7 (7)
Diplegia, n (%) 6 (6) 10 (10) 16 (16)
Tri/quadriplegia, n (%) 12 (12) 45 (44) 57 (55)
Dystonia, n. (%) 6 (6) 10 (10) 16 (16)
Well controlled epilepsy, n. (%) 15 (15) 39 (38) 54 (53)
Intractable epilepsy, n. (%) 8 (8) 13 (13) 21 (21)
No epilepsy, n. (%) 7 (7) 20 (20) 27 (26)
Psychotropic medication, n. (%) 9 (9) 14 (14) 23 (23)
Intellectual disability: mild to moderate 0 (0) 11 (11) 11 (11)
Intellectual disability: severe n. (%) 9 (9) 15 (15) 24 (24)
Intellectual disability: profound (%) 22 (22) 45 (44) 67 (65)
Antenatal causes, n. (%) 18 (18) 42 (41) 60 (59)
Perinatal causes, n. (%) 11 (11) 17 (17) 28 (27)
Postnatal causes, n. (%) 3 (3) 11 (11) 14 (14)

Abbreviations: ASD, autism spectrum disorder; SD, standard deviation.

Neuropediatrics

Third step: for each tuple, we performed a logistic regres-
sion and assessed its performance to predict each patient’s
probability of having ASD using open source software R with
the GLM (generalized linear model) function.30 Thus, we
trained the logistic regression algorithm on a “training set”
of 82 patients to predict the probability that a patient in the
“test set” (n ¼ 21) would have ASD. In accordance with the
statistical learning theory described by Vapnik,31 we used
cross validation by randomly generating 20 different couples
of training and test sets. We calculated the accuracy, sensitiv-
ity, and specificity of the predictions for each couple of training
and test sets and calculated their average. Sensitivity, specifi-
city, and accuracy were described in terms of true positives
(TP), true negatives (TN), false negatives (FN), and false posi-
tives (FP).32 Sensitivity was defined as the proportion of actual
positives and identified as such [TP/(TP þ FN)]. Specificity was
defined as the proportion of actual negatives and identified as
such [TN/(TN þ FP)]. Accuracy was defined as the proportion
of TP and TN in all assessments and identified as (TP þ TN)/
(TP þ TN þ FP þ FN).33 We had no missed data.
Fourth step: we found which tuple out of 1,023 was the
best in terms of predictive performance (accuracy, sensitiv-
ity, and specificity).
Fifth step: we used the tuple found in step 4 in a new
logistic regression to predict ASD in new incoming patients.
Therefore, we performed a feature reduction eliminating
redundant variables (e.g., eliminating some of the predic-
tors having more mutual interactions with the others)
leading to the maximization of the predictive performance
and reducing the number of independent variables from 10
to 8.
Results
Descriptive Analyses
ASD was observed in 30% of subjects; 71% of those with ASD
were males. A breakdown of the clinical presentation accord-
ing to the presence of ASD is shown on ►Table 2. The most
60
50
40
LEVELS
30
20
10
0
GMFCS CFCS
I 1 0 2
II 16 0 0
III 5 14
IV 24 40
V 56 48
Fig. 1 Distribution of patients according to the gross motor function classification system (GMFCS), manual ability classification system (MACS),
eating and drinking ability classification system (EDACS), communication function classification system (CFCS)
Predictors of Autism in Cerebral Palsy Bertoncelli et al.
common pattern of spastic disorder was tri/quadriplegia
(55%), followed by lower limbs diplegia (16%), and hemi-
plegia (7%). Epilepsy was heavily present (75%) in ASD
patients but not statistically linked with autistic features.
A third of ASD patients were taking long-term psychotropic
medications (antipsychotics 64%, antidepressants 66%).
Motor skills, communication abilities, eating and drinking
capacities evaluated with the EDACS, CFCS, MACS, and
GMFCS are summarized on ►Fig. 1. The support need scores
evaluated with the SIS indicated that subjects with ASD had
higher support need scores (mean score, 12; SD, 1.8) com-
pared with subjects without ASD (mean score, 5; SD, 1.4).
Subjects with ASD had higher behavioral scores (mean score,
8.1; SD, 1.4), when compared with subjects without ASD
(mean score, 6.6; SD, 1.4).
Logistic Regressions Analyses
Fisher’s exact test revealed that factors significantly asso-
ciated with ASD were: presence of spasticity (hemiplegia >
diplegia > tri/quadriplegia; p ¼ 0.03), communication dis-
Downloaded by: Université Paris Sud XI. Copyrighted material.
orders (p ¼ 0.03), and good feeding abilities
(p ¼ 0.05; ►Table 3). The factors that were significantly
associated with ASD in the multivariate analysis were: CFCS
> 4 (p < 0.001), EDACS score < 3 (p ¼ 0.002), GMFCS score
< 3 (p ¼ 0.01), spasticity (p ¼ 0.04), intellectual disability
(p ¼ 0.05; ►Table 4).
The best tuple to predict ASD in terms of accuracy,
sensitivity, and specificity was: ID þ SP þ EP þ SE þ
GMFCS þ EDACS þ CFCS þ PS. The logistic regression based
machine learning model using as independent variables the
tuple (ID þ SP þ EP þ SE þ GMFCS þ EDACS þ CFCS þ PS)
showed a good predictive performance with accuracy of 73%,
sensitivity of 79%, and specificity of 72% (►Table 4).
Discussion
Adolescents with CP had approximately twice the odds of
having ASD compared with adolescents without CP. Our
MACS EDACS
20
22
20 14
26 20
54 26
Neuropediatrics
Predictors of Autism in Cerebral Palsy Bertoncelli et al.

Table 3 Contingency table comparing subjects with and without autism spectrum disorder (ASD) using the Fisher’s exact test

Independent variables Autism Fisher’s exact Odds 95% confidence

spectrum test; p-value ratio intervals
disorder
Yes No
CFCS four-fifth vs. Yes 31 60 0.031 Infinity Infinity
CFCS  3
No 0 11
Presence of Yes 20 60 0.035 0.33 0.12–0.88
Spasticity
No 11 11
EDACS four-fifth vs. Yes 10 38 0.050 0.41 0.17–1.00
EDACS  3
No 21 33

Abbreviations: CFCS, communication function classification system; EDACS, eating and drinking ability classification system.

Table 4 List of the logistic regression coefficient (independent variables) associated with the presence of autism spectrum
disorder (ASD)

Downloaded by: Université Paris Sud XI. Copyrighted material.

Logistic regression
Independent variables
Odds ratio estimate Standard error Z ratio Prob (> |z|)
Logaritm Linear
Intercept 9.1966 16.338 2.7935 3.292 0.000994
Spasticity (SP) 0.5675 0.5669 0.2785 2.038 0.041595
Sex (SE) 0.8721 2.3919 0.5505 1.584 0.113136
EDACS 1.0439 0.3520 0.3500 2.983 0.002857
CFCS 2.0072 7.442 0.5908 3.397 0.000681
Psychotropic medication 0.1547 1.1673 0.5954 0.260 0.795068
Intercept 1.9133 0. 1475 1.4954 1.279 0.2008
Intellectual disability (ID) 0.8207 2.2720 0.4334 1.893 0.0500
Epilepsy (E) 0.3015 1.3518 0.3717 0.811 0.4173
GMFCS 0.5262 0.5908 0.2213 2.378 0.0174

Abbreviations: CFCS, communication function classification system; EDACS, eating and drinking ability classification system; GMFCS, gross motor
function classification system; SE, standard error.
Logistic regression: the increasing of CFCS and ID (positive values) as well as decreasing SP (tri/quadriplegia < diplegia< hemiplegia), EDACS,
GMFCS, (negative values) are associated factors with the presence of ASD (in the “estimate” column).
More precisely this means, that is, that for every unit increase in CFCS the log odds ¼ ln (p/1  p) increases 2.0072 times (where p ¼ probability to
have ASD), while for every unit decrease in spasticity the log odds ¼ ln (p/1  p) decreases 0.5675..
The “Prob (> |z|)” column indicates the significance strength of the respective parameter in terms of p-value as ASD predictor. This means that the
significance of SP, EDACS, CFCS, ID, and GMFCS in predicting ASD is very probable, with a p-value < 0.05.

ML model provides more accurate prediction than models
developed by previous studies7; the predictive accuracy, sen-
sitivity, and specificity33 average of the model we developed
was 75%.
Teenagers with CP and ASD differed significantly from
those without ASD in terms of type of CP, intellectual level,
and speech ability.
As mentioned in the introduction,3,4 previous stu-
dies3,20,34 found the prevalence of ASD in children with CP
to be 10 to 20%; approximately 40% of the children with CP
had moderate or severe intellectual impairment, 40% had
profound intellectual impairment, and 0% had no ID.34
Likewise, in the present study, ASD was observed in 30% of
the subjects, 35% of which presented moderate or severe ID,
the others profound ID, and 0% had no ID.
Feeding difficulties are relatively common in children with
ASD but current evidence for their treatment is limited.35 A
study36 showed significantly larger prevalence of autism
spectrum disorder in people with than without eating dis-
orders. Conversely, we found that adolescents with CP with
ASD showed better feeding abilities compared with the control
group. This is interpretable because ASD symptoms are more
evident and detectable in young people with CP than in
subjects with more developed motor and feeding skills.

Neuropediatrics

A recent study37 found three- to four-fold increase in
emotional and behavioral disorders among children with CP
compared with similar-aged peers. In our study, we also
found that subjects with ASD had higher behavioral support
need scores than subjects without ASD. Given that recent
population prevalence estimates of ASD are as high as 2% the
relative elevation in ASD rates in children with CP appears to
be in line with other emotional–behavioral comorbidities.33
A study38 highlighted a link between prenatal causes (25%
of very preterm infants) and ADS. We did not find any link
between the etiology (prenatal, perinatal, and postnatal
causes) and ASD. In line with Surén et al,39 we confirm that
CP boys had a slightly (p ¼ 0.11) increased risk to present ASD.
The clinical predictive learning model developed pre-
dicted ASD with a good precision (75%), and can be used to
recommend early intervention. Early identification of high
likelihood of ASD development will allow families and
therapists to anticipate patient support and intervention.
This is important because in early childhood there are more
chances of success in stimulating communication and intel-
lectual abilities. Also, communication with families is extre-
mely delicate, using this model, doctors will be supported by
accurate predictive data and objective information when
informing families about patient’s conditions and prognosis.
In addition, this predictive model could be used in various
fields of medicine because it is easily adaptable. In the future
“PredictMed” could be used for eating disorders, mood
disorders, and epilepsy. The results can also aid clinicians
in choosing assessment tools, in addition to the classic ASD
specific tools; this study has identified that functional
assessments (GMFCS, EDACS, CFCS, and SIS) should be prior-
itized in the predictive diagnosis of ADS.
The main limitation of this study is the relatively small
number of patients. The number of independent variables
was numerous, so a clinical choice was made on the most
representative/relevant ones to test. In future, advances in AI
will likely be used to remove potential researcher selection
biases. It is also possible that other variables that were not
available in the data are good or even better predictors.
Future research may want to assess other factors and their
association with the probability of having ASD. Finally, for
clinical applications, the prediction score should be
increased with a larger series of testing set data.
Prediction Model Identifying Codes
True positive (TP), true negative (TN), false negative (FN),
false positive (FP), etiology (ET), type of spasticity (SP),
dystonia (D), epilepsy (E), sex (SE), psychotropic drugs
(PS), autistic features (A), feeding abilities (EDACS), com-
munication (CFCS), and gross motor (GMFCS) function.
Disclosure Statement
This research did not receive any specific grant from
funding agencies in the public, commercial, or not-for-
profit sectors.
Financial Support
None.
Predictors of Autism in Cerebral Palsy Bertoncelli et al.
Conflicts of Interest
None. There are not potential conflicts of interest, real or
perceived; this includes a description of the role of the
study sponsor(s).
Acknowledgments
We would like to thank Dr. Elodie Zviadadzé, psychologist
(Établissement pour enfants et adolescents polyhandi-
capés H. Germain–Fondation Lenval, Nice) for her help in
preparing this manuscript.
References
1 Evans PM, Evans SJW, Alberman E. Cerebral palsy: why we must
plan for survival. Arch Dis Child 1990;65(12):1329–1333
2 Hirschberger RG, Kuban KCK, O’Shea TM, et al; ELGAN Study
Investigators. Co-occurrence and severity of neurodevelopmental
burden (cognitive impairment, cerebral palsy, autism spectrum
disorder, and epilepsy) at age ten years in children born extremely
preterm. Pediatr Neurol 2018;79:45–52
Downloaded by: Université Paris Sud XI. Copyrighted material.
3 Christensen D, Van Naarden Braun K, Doernberg NS, et al. Pre-
valence of cerebral palsy, co-occurring autism spectrum disor-
ders, and motor functioning - autism and developmental
disabilities monitoring network, U.S.A., 2008. Dev Med Child
Neurol 2014;56(01):59–65
4 Ung D, Wood JJ, Ehrenreich-May J, et al. Clinical characteristics of
high-functioning youth with autism spectrum disorder and anxi-
ety. Neuropsychiatry (London) 2013;3(2)
5 Bottcher L. Children with spastic cerebral palsy, their cognitive
functioning, and social participation: a review. Child Neuropsy-
chol 2010;16(03):209–228
6 Bjorgaas HM, Hysing M, Elgen I. Psychiatric disorders among
children with cerebral palsy at school starting age. Res Dev Disabil
2012;33(04):1287–1293
7 Stadskleiv K, Jahnsen R, Andersen GL, von Tetzchner S. Neuropsy-
chological profiles of children with cerebral palsy. Dev Neurorehabil
2018;21(02):108–120
8 Kim H, Carlson AG, Curby TW, Winsler A. Relations among motor,
social, and cognitive skills in pre-kindergarten children with
developmental disabilities. Res Dev Disabil 2016;53-54:43–60
9 Iniesta R, Stahl D, McGuffin P. Machine learning, statistical learning
and the future of biological research in psychiatry. Psychol Med
2016;46(12):2455–2465 10.1017/S0033291716001367
10 Meyer-Lindenberg A. [Artificial intelligence in psychiatry-an
overview]. Nervenarzt 2018;89(08):861–868
11 Mossotto E, Ashton JJ, Coelho T, Beattie RM, MacArthur BD, Ennis
S. Classification of paediatric inflammatory bowel disease using
machine learning. Sci Rep 2017;7(01):2427
12 Srividya M, Mohanavalli S, Bhalaji N. Behavioral modeling for
mental health using machine learning algorithms. J Med Syst
2018;42(05):88
13 Dwyer DB, Falkai P, Koutsouleris N. Machine learning approaches
for clinical psychology and psychiatry. Annu Rev Clin Psychol
2018;14:91–118
14 Bzdok D, Meyer-Lindenberg A. Machine Learning for Precision
Psychiatry: Opportunities and Challenges. Biol Psychiatry Cogn
Neurosci Neuroimaging 2018;3(03):223–230
15 Thabtah F. Machine learning in autistic spectrum disorder beha-
vioral research: A review and ways forward. Inform Health Soc
Care 2018;13:1–20; ( Epub ahead of print ) doi:10.1080/
17538157.2017.1399132
16 Bertoncelli CM, Solla F, Loughenbury PR, Tsirikos AI, Bertoncelli D,
Rampal V. Risk factors for developing scoliosis in cerebral palsy: a
cross-sectional descriptive study. J Child Neurol 2017;32(07):
657–662
Neuropediatrics
Predictors of Autism in Cerebral Palsy Bertoncelli et al.
17 Bertoncelli CM, Bertoncelli D, Elbaum L, et al. Validation of a
clinical prediction model for the development of neuromuscular
scoliosis: a multinational study. Pediatr Neurol 2018;79:14–20
18 Bertoncelli CM, Altamura P, Vieira ER, Bertoncelli D, Thummler S,
Solla F. Identifying factors associated with severe intellectual
disabilities in teenagers with cerebral palsy using a predictive
learning model. J Child Neurol 2019; ( Epub ahead of print). Doi:
10.1177/0883073818822358
19 Gainsborough M, Surman G, Maestri G, Colver A, Cans C. Validity
and reliability of the guidelines of the surveillance of cerebral
palsy in Europe for the classification of cerebral palsy. Dev Med
Child Neurol 2008;50(11):828–831
20 Kilincaslan A, Mukaddes NM. Pervasive developmental disorders
in individuals with cerebral palsy. Dev Med Child Neurol 2009;51
(04):289–294
21 Boat TF, Wu JT; National Academies of Sciences, Engineering, and
Medicine; Institute of Medicine; Board on the Health of Select
Populations; Board on Children, Youth, and Families; Committee
to Evaluate the Supplemental Security Income Disability Program
for Children with Mental Disorders. Mental Disorders and Dis-
abilities Among Low-Income Children. Washington (DC):
National Academies Press (US); 2015
22 Hareb F, Rampal V, Bertoncelli CM, Solla F. Botulinum toxin in
children with cerebral palsy: an update. Clin Cases Miner Bone
Metab 2019:16(1)
23 Sinha S, Siddiqui KA. Definition of intractable epilepsy. Neuros-
ciences (Riyadh) 2011;16(01):3–9
24 Berg AT. Identification of pharmacoresistant epilepsy. Neurol Clin
2009;27(04):1003–1013
25 Delobel-Ayoub M, Klapouszczak D, van Bakel MME, et al. Prevalence
and characteristics of autism spectrum disorders in children with
cerebral palsy. Dev Med Child Neurol 2017;59(07):738–742
26 Moons KG, Altman DG, Reitsma JB, Collins GS; Transparent
Reporting of a Multivariate Prediction Model for Individual
Prognosis or Development Initiative. New guideline for the
reporting of studies developing, validating, or updating a multi-
variable clinical prediction model: the TRIPOD statement. Adv
Anat Pathol 2015;22(05):303–305
27 Solla F, Tran A, Bertoncelli D, Musoff C, Bertoncelli CM. Why a p-value
is not enough. Clin Spine Surg 2018;31(09):385–388
Neuropediatrics
28 Sullivan KM, Dean A, Soe MM. OpenEpi: a web-based epidemio-
logic and statistical calculator for public health. Public Health Rep
2009;124(03):471–474
29 Altman DG, Deeks JJ, Sackett DL. Odds ratios should be avoided
when events are common. BMJ 1998;317(7168):1318
30 The R Core Team. R: A Language and Environment for Statistical
Computing. Vienna, Austria: R Foundation for Statistical Comput-
ing; 2012
31 Vapnik V. The Nature of Statistical Learning Theory. New York, NY:
Springer-Verlag; 2000
32 Wen Z, Zeng N, Wang N. Sensitivity, specificity, accuracy, asso-
ciated confidence interval and ROC analysis with practical SAS
implementations. 2010. Available from: https://pdfs.seman-
ticscholar.org/d1e5/c3097daf99db2c8dce3ac0edc3c5ade41460.
pdf?_ga¼2.97662365.32552516.1552555838-
600088543.1552555838. Accessed March 14, 2019
33 Thombs BD, Rice DB. Sample sizes and precision of estimates of
sensitivity and specificity from primary studies on the diagnostic
accuracy of depression screening tools: a survey of recently
published studies. Int J Methods Psychiatr Res 2016;25(02):
145–152
34 Kilincaslan A, Mukaddes NM. Pervasive developmental disorders
in individuals with cerebral palsy. Dev Med Child Neurol 2009;51
Downloaded by: Université Paris Sud XI. Copyrighted material.
(04):289–294
35 Marshall J, Ware R, Ziviani J, Hill RJ, Dodrill P. Efficacy of inter-
ventions to improve feeding difficulties in children with autism
spectrum disorders: a systematic review and meta-analysis. Child
Care Health Dev 2015;41(02):278–302
36 Huke V, Turk J, Saeidi S, Kent A, Morgan JF. Autism spectrum
disorders in eating disorder populations: a systematic review. Eur
Eat Disord Rev 2013;21(05):345–351
37 Bjorgaas HM, Hysing M, Elgen I. Psychiatric disorders among
children with cerebral palsy at school starting age. Res Dev Disabil
2012;33(04):1287–1293
38 Limperopoulos C, Bassan H, Sullivan NR, et al. Positive screening
for autism in ex-preterm infants: prevalence and risk factors.
Pediatrics 2008;121(04):758–765
39 Surén P, Bakken IJ, Aase H, et al. Autism spectrum disorder, ADHD,
epilepsy, and cerebral palsy in Norwegian children. Pediatrics
2012;130(01):e152–e158
 Predictors of Autism in Cerebral Palsy Bertoncelli et al.

Appendix 1 TRIPOD checklist: prediction model development and validation

Section/topic Item Checklist item Page

Title and abstract
Title 1 D; V Identify the study as developing and/or validating a multivari- 1
able prediction model, the target population, and the outcome
to be predicted.
Abstract 2 D; V Provide a summary of objectives, study design, setting, parti- 3
cipants, sample size, predictors, outcome, statistical analysis,
results, and conclusions.
Introduction
Background and 3a D; V Explain the medical context (including whether diagnostic or 5
objectives prognostic) and rationale for developing or validating the
multivariable prediction model, including references to existing
models.
3b D; V Specify the objectives, including whether the study describes 5
the development or validation of the model or both.
Methods
Source of data 4a D; V Describe the study design or source of data (e.g., randomized 6

Downloaded by: Université Paris Sud XI. Copyrighted material.

trial, cohort, or registry data), separately for the development
and validation datasets, if applicable.
4b D; V Specify the key study dates, including start of accrual; end of 6
accrual; and, if applicable, end of follow-up.
Participants 5a D; V Specify key elements of the study setting (e.g., primary 6
care, secondary care, general population) including number and
location of centres.
5b D; V Describe eligibility criteria for participants. 6
5c D; V Give details of treatments received, if relevant. None
Outcome 6a D; V Clearly define the outcome that is predicted by the prediction 6
model, including how and when assessed.
6b D; V Report any actions to blind assessment of the outcome to be 6
predicted.
Predictors 7a D; V Clearly define all predictors used in developing or validating the 8
multivariable prediction model, including how and when they
were measured.
7b D; V Report any actions to blind assessment of predictors for the 6
outcome and other predictors.
Sample size 8 D; V Explain how the study size was arrived at. 6
Missing data 9 D; V Describe how missing data were handled (e.g., complete-case 10
analysis, single imputation, multiple imputation) with details of
any imputation method.
Statistical 10a D Describe how predictors were handled in the analyses. 9
analysis methods
10b D Specify type of model, all model-building procedures (including 9, 10
any predictor selection), and method for internal validation.
10c V For validation, describe how the predictions were calculated. 10
10d D; V Specify all measures used to assess model performance and, if 9
relevant, to compare multiple models.
10e V Describe any model updating (e.g., recalibration) arising from None
the validation, if done.
Risk groups 11 D; V Provide details on how risk groups were created, if done. None
Development 12 V For validation, identify any differences from the development None
versus validation data in setting, eligibility criteria, outcome, and predictors.
(Continued)

Neuropediatrics
Predictors of Autism in Cerebral Palsy Bertoncelli et al.

Appendix 1 (Continued)

Section/topic Item Checklist item Page

Results
Participants 13a D; V Describe the flow of participants through the study, including 11
the number of participants with and without the outcome and, if
applicable, a summary of the follow-up time. A diagram may be
helpful.
13b D; V Describe the characteristics of the participants (basic demo- 11
graphics, clinical features, available predictors), including the
number of participants with missing data for predictors and
outcome.
13c V For validation, show a comparison with the development data of None
the distribution of important variables (demographics, predic-
tors and outcome).
Model 14a D Specify the number of participants and outcome events in each 11, 12
development analysis.
14b D If done, report the unadjusted association between each can- –
didate predictor and outcome.
Model 15a D Present the full prediction model to allow predictions for ►Table 3

Downloaded by: Université Paris Sud XI. Copyrighted material.

specification individuals (i.e., all regression coefficients, and model intercept
or baseline survival at a given time point).
15b D Explain how to the use the prediction model. 7, ►Table 3
Model 16 D; V Report performance measures (with CIs) for the prediction ►Table 2
performance model.
Model 17 V If done, report the results from any model updating (i.e., model None
updating specification, model performance).
Discussion
Limitations 18 D; V Discuss any limitations of the study (such as nonrepresentative 13
sample, few events per predictor, missing data).
Interpretation 19a V For validation, discuss the results with reference to performance None
in the development data, and any other validation data.
19b D; V Give an overall interpretation of the results, considering objec- 12, 13
tives, limitations, results from similar studies, and other relevant
evidence.
Implications 20 D; V Discuss the potential clinical use of the model and implications 13
for future research.
Other information
Supplementary 21 D; V Provide information about the availability of supplementary 9
information resources, such as study protocol, Web calculator, and datasets.
Funding 22 D; V Give the source of funding and the role of the funders for the None
present study.

Note: Items relevant only to the development of a prediction model are denoted by D, items relating solely to a validation of a prediction model are
denoted by V, and items relating to both are denoted D; V. We recommend using the TRIPOD checklist in conjunction with the TRIPOD explanation
and elaboration document. TRIPOD, transparent reporting of a multivariable prediction model for individual prognosis or diagnosis.

Neuropediatrics