Stereotypies in Autism: An Innovative Mathematical Approach to Depict the Natural Association Scheme of Their Co-Occurrence

Background:

In autism, stereotypies (stereotypic movement disorders) are frequent and disabling and represent one of the most complex clinical pictures due to a broad spectrum of anomalies.Following the new wave of biology-based research in autism, motor anomalies and other repetitive behaviors are increasingly receiving attention. Indeed, the co-occurrence of many different stereotypies in the same subject theoretically offers the possibility to derive associative patterns useful in developing interpretative models.

Objectives:

The aim of this study is to analyze stereotypies patterns observed in a sample of children and adolescents residing at our Institute and subsequently classify them by means of video-recordings. By using advanced machine learning systems, we are able to develop a semantic connectivity map of the variables under investigation.

Methods:

To define the spectrum of expressions of stereotypies we studied 67 autistic individuals which, as a group, expressed 37 different types of stereotypies defined through standardized video-recording. All individuals but one presented a certain number of stereotypies: average = 11.5; range 0-27. The data were analyzed with a special kind of unsupervised artificial neural network ( Auto-CM). Auto-CM is able to a semantic connectivity map in which the matrix of connections, visualized through a minimum spanning tree filter, takes into account nonlinear associations among variables and captures connection schemes among clusters. In this way, the patient state can be viewed as an hyperpoint in a “multimorbidity space” in which each dimension corresponds to a quantitative phenotype.

Results:

The semantic connectivity map showed a meaningful scheme of connections among stereotypies.As far as motor abnormalities are concerned, mouth-trunk-arms movements constitute a central axis of the system from which all other type of movements involving head, legs, shoulder and feet take place. Toe walking is directly linked to other walking abnormalities. Licking, biting, smelling, rubbing and touching body parts form a unique cluster associated to medium ID severity and separate from licking, biting and smelling objects, which is associated to mild ID severity. Severe ID is associated to simple voicing and facial grimacing.

Conclusions:

Machine learning algorithms are able to depict the complex pattern of stereotypies commonly observed in autism, thus allowing for a better definition of major phenotypes that are amenable for future large epidemiological surveys.