Background: Research suggests that families' education and economic status together with cultural perceptions of autism, can affect parents' recognition of the first signs of autism in their children and influence their decision to seek medical help for ASD suspicion. No specific studies in this regard have been conducted in Italy.

Objectives: The focus of this pilot study is the association between age at first medical contact for diagnostic suspicion and parental characteristics in Lombardy Region, the most populous and rich Region in Italy.

Methods: Forty-nine ASD cases referred to our Institute for Rehabilitation from 2022 to 2024 coming from Lombardy municipalities were included in the study. The age at first medical contact for parental concern, determined from parents' reports, was the prediction target. Child gender and age at presentation in our Centre, mother, and father age at childbirth, their education achievement level, their occupation level, and their place of birth (North, Centre, South Italy, Foreign country) were used for input data.
The existence of a poor linear correlation between features and targets prompted us to use a machine-learning system approach. The evolutionary algorithm TWIST was used to subdivide the dataset into training and testing sets as well as to select features yielding the maximum amount of information. After this pre-processing, 11 out of 18 input variables were selected and different machine learning systems were used to develop a predictive model based on a training-testing crossover procedure able to distinguish subjects lying in the two classes of age at first medical contact (<36 months vs >36 months). A fourth- generation artificial neural network called Auto-CM coupled with a minimum spanning tree filter developed a semantic connectivity map of the 20 variables on study (figure1).

Results: The study group was composed of 33 males and 16 females (mean age at presentation 7.5 yrs; range 2-15 yrs) diagnosed with autism according to DSM V criteria. The age at first medical contact for parental concern ranged between 6 months and 8 years (mean 38 months). TWIST systems selected 11 variables (Child age at observation, parents' education and occupational level, parents place of birth). A machine learning model using these variables predicted class age at first contact with an overall accuracy of 88.65% (sensitivity 93.5% - specificity 83.7%) with a ROC/AUC = 0.85. Higher socio-economic status and north parental origin were associated to a later age at first contact.

Conclusions: The educational professional and cultural background of parents play a substantial role in influencing the decision to seek medical help for ASD suspicion in a complex way. A machine learning model built on study variables predicted class of age at first contact with high accuracy.

Notes:

Autism Research Unit, Villa Santa Maria SCS, Tavernerio, Como, Italy