Professional Documents
Culture Documents
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
Neuropediatrics
Predictors of Autism in Cerebral Palsy Bertoncelli et al.
Table 1 List of assessment tools, skills evaluated and function levels of ratings
Neuropediatrics
Predictors of Autism in Cerebral Palsy Bertoncelli et al.
group homes.21 These individuals and they require close have ASD if one of the diagnoses of category F84 of the
supervision and help with self-care activities. Individuals international classification of diseases, 10th revision (ICD-
with profound ID require round-the-clock support and care. 10) was reported in medical records.25
They depend on others for all aspects of day-to-day life, have
major physical limitations, and extremely limited commu- Statistical Analysis
nication ability. The guidelines of the “Transparent Reporting of a multi-
variable prediction model for Individual Prognosis or Diag-
Neurologic Status nosis” (TRIPOD; ►Appendix 1) statement were followed.26
Neurologic status was classified according to the anatomy of All data were analyzed in anonymous form and entered into a
the spastic disorder (hemiplegia, diplegia, or tri/quadriplegia), database including demographics, functional diagnosis, neu-
the presence of hypertonia in the upper or lower limbs, the rologic, and cognitive assessments (►Table 2).
presence of dystonia, and the severity of epilepsy. Spasticity First step: to generate the predictive model, we divided our
was quantified using the Bohannon and Smith modified Ash- cohort into two groups: with and without ASD. The dependent
worth’s scale and the modified Tardieu’s scale.22 Severity of binary variable was ASD (yes or no), and the 10 independent
epilepsy was determined by the pediatric neurologists and variables were etiology (ET), type of spasticity (SP), intellectual
identified as “well controlled” or “intractable”23 accordingly disability (ID) dystonia (D), epilepsy (EP), sex (SE), EDACS,
with the International League against Epilepsy, which defines CFCS, GMFCS, and psychotropic drugs (PS). Our aim was to find
intractable epilepsy as continued seizures despite adequate which combinations (tuples) of independent variables best
trials of at least two appropriate antiepileptic agents.24 Long- predicted ASD. Initial analysis included using contingency
term psychotropic medication (antipsychotics and/or antide- tables and Fisher’s exact test to identify associated factors,
Neuropediatrics
Predictors of Autism in Cerebral Palsy Bertoncelli et al.
Third step: for each tuple, we performed a logistic regres- common pattern of spastic disorder was tri/quadriplegia
sion and assessed its performance to predict each patient’s (55%), followed by lower limbs diplegia (16%), and hemi-
probability of having ASD using open source software R with plegia (7%). Epilepsy was heavily present (75%) in ASD
the GLM (generalized linear model) function.30 Thus, we patients but not statistically linked with autistic features.
trained the logistic regression algorithm on a “training set” A third of ASD patients were taking long-term psychotropic
of 82 patients to predict the probability that a patient in the medications (antipsychotics 64%, antidepressants 66%).
“test set” (n ¼ 21) would have ASD. In accordance with the Motor skills, communication abilities, eating and drinking
statistical learning theory described by Vapnik,31 we used capacities evaluated with the EDACS, CFCS, MACS, and
cross validation by randomly generating 20 different couples GMFCS are summarized on ►Fig. 1. The support need scores
of training and test sets. We calculated the accuracy, sensitiv- evaluated with the SIS indicated that subjects with ASD had
ity, and specificity of the predictions for each couple of training higher support need scores (mean score, 12; SD, 1.8) com-
and test sets and calculated their average. Sensitivity, specifi- pared with subjects without ASD (mean score, 5; SD, 1.4).
city, and accuracy were described in terms of true positives Subjects with ASD had higher behavioral scores (mean score,
(TP), true negatives (TN), false negatives (FN), and false posi- 8.1; SD, 1.4), when compared with subjects without ASD
tives (FP).32 Sensitivity was defined as the proportion of actual (mean score, 6.6; SD, 1.4).
positives and identified as such [TP/(TP þ FN)]. Specificity was
defined as the proportion of actual negatives and identified as Logistic Regressions Analyses
such [TN/(TN þ FP)]. Accuracy was defined as the proportion Fisher’s exact test revealed that factors significantly asso-
of TP and TN in all assessments and identified as (TP þ TN)/ ciated with ASD were: presence of spasticity (hemiplegia >
(TP þ TN þ FP þ FN).33 We had no missed data. diplegia > tri/quadriplegia; p ¼ 0.03), communication dis-
60
50
40
LEVELS
30
20
10
0
GMFCS CFCS MACS EDACS
I 1 0 2 20
II 16 0 0 22
III 5 14 20 14
IV 24 40 26 20
V 56 48 54 26
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)
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
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)
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 Likewise, in the present study, ASD was observed in 30% of
developed by previous studies7; the predictive accuracy, sen- the subjects, 35% of which presented moderate or severe ID,
sitivity, and specificity33 average of the model we developed the others profound ID, and 0% had no ID.
was 75%. Feeding difficulties are relatively common in children with
Teenagers with CP and ASD differed significantly from ASD but current evidence for their treatment is limited.35 A
those without ASD in terms of type of CP, intellectual level, study36 showed significantly larger prevalence of autism
and speech ability. spectrum disorder in people with than without eating dis-
As mentioned in the introduction,3,4 previous stu- orders. Conversely, we found that adolescents with CP with
dies3,20,34 found the prevalence of ASD in children with CP ASD showed better feeding abilities compared with the control
to be 10 to 20%; approximately 40% of the children with CP group. This is interpretable because ASD symptoms are more
had moderate or severe intellectual impairment, 40% had evident and detectable in young people with CP than in
profound intellectual impairment, and 0% had no ID.34 subjects with more developed motor and feeding skills.
Neuropediatrics
Predictors of Autism in Cerebral Palsy Bertoncelli et al.
Neuropediatrics
Predictors of Autism in Cerebral Palsy Bertoncelli et al.
17 Bertoncelli CM, Bertoncelli D, Elbaum L, et al. Validation of a 28 Sullivan KM, Dean A, Soe MM. OpenEpi: a web-based epidemio-
clinical prediction model for the development of neuromuscular logic and statistical calculator for public health. Public Health Rep
scoliosis: a multinational study. Pediatr Neurol 2018;79:14–20 2009;124(03):471–474
18 Bertoncelli CM, Altamura P, Vieira ER, Bertoncelli D, Thummler S, 29 Altman DG, Deeks JJ, Sackett DL. Odds ratios should be avoided
Solla F. Identifying factors associated with severe intellectual when events are common. BMJ 1998;317(7168):1318
disabilities in teenagers with cerebral palsy using a predictive 30 The R Core Team. R: A Language and Environment for Statistical
learning model. J Child Neurol 2019; ( Epub ahead of print). Doi: Computing. Vienna, Austria: R Foundation for Statistical Comput-
10.1177/0883073818822358 ing; 2012
19 Gainsborough M, Surman G, Maestri G, Colver A, Cans C. Validity 31 Vapnik V. The Nature of Statistical Learning Theory. New York, NY:
and reliability of the guidelines of the surveillance of cerebral Springer-Verlag; 2000
palsy in Europe for the classification of cerebral palsy. Dev Med 32 Wen Z, Zeng N, Wang N. Sensitivity, specificity, accuracy, asso-
Child Neurol 2008;50(11):828–831 ciated confidence interval and ROC analysis with practical SAS
20 Kilincaslan A, Mukaddes NM. Pervasive developmental disorders implementations. 2010. Available from: https://pdfs.seman-
in individuals with cerebral palsy. Dev Med Child Neurol 2009;51 ticscholar.org/d1e5/c3097daf99db2c8dce3ac0edc3c5ade41460.
(04):289–294 pdf?_ga¼2.97662365.32552516.1552555838-
21 Boat TF, Wu JT; National Academies of Sciences, Engineering, and 600088543.1552555838. Accessed March 14, 2019
Medicine; Institute of Medicine; Board on the Health of Select 33 Thombs BD, Rice DB. Sample sizes and precision of estimates of
Populations; Board on Children, Youth, and Families; Committee sensitivity and specificity from primary studies on the diagnostic
to Evaluate the Supplemental Security Income Disability Program accuracy of depression screening tools: a survey of recently
for Children with Mental Disorders. Mental Disorders and Dis- published studies. Int J Methods Psychiatr Res 2016;25(02):
abilities Among Low-Income Children. Washington (DC): 145–152
National Academies Press (US); 2015 34 Kilincaslan A, Mukaddes NM. Pervasive developmental disorders
22 Hareb F, Rampal V, Bertoncelli CM, Solla F. Botulinum toxin in in individuals with cerebral palsy. Dev Med Child Neurol 2009;51
Neuropediatrics
Predictors of Autism in Cerebral Palsy Bertoncelli et al.
Neuropediatrics
Predictors of Autism in Cerebral Palsy Bertoncelli et al.
Appendix 1 (Continued)
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