Background: Feeding problems are commonly associated with autism spectrum disorder (ASD). Selective eating has been linked to sensory sensitivity and cognitive rigidity in ASD, ARFID, and in typically developing children.

Objectives: This study aims to define and investigate the overlap and differences in eating problems, sensory profiles, and restricted repetitive behaviours present in ASD and ARFID patients, using data mining techniques based on AI to unveil hidden associations between variables that are not visible with classical statistics. Eating difficulties are defined in terms of food selectivity, food refusal, mealtime rigidity, and disruptive mealtime behaviors.

Methods: This is a cross-sectional observational study of 44 patients (22 ASD and 22 ARFID), aged 2-14 years, assessed at the Regional Centre for Feeding and Eating Disorders at the IRCCS ISNB in Bologna, Italy. All parents completed the Brief Autism Mealtime Behavior Inventory-Revised (BAMBI-R), Short Sensory Profile-2, and Repetitive Behavior Scale-Revised. A retrospective analysis of the total and subscale scores and other clinical variables, using ANCOVA, t-test, and chi-square statistics, was done. To graphically show the most important connections among the variables, we used a fourth-generation artificial neural network (ANN) called Auto-CM, which develops weights that are proportional to the strength of the associations among all variables. After the training phase, the weights matrix of the Auto-CM represents the warped landscape of the dataset. Subsequently, a minimum spanning tree filter is applied to the weight matrix of the Auto-CM system to obtain a map of the main connections between the variables of the dataset and the basic semantics of their similarities, defined as a connectivity map.

Results: The ARFID group showed higher (p=0.01) BAMBI-R scores compared to ASD, when controlling for the effect of repetitive behaviour and alterations in the sensory profile. This difference is maintained when considering the Food selectivity subscale (p=0.02) but not with the other subscales. Figure 1 shows the semantic connectivity map produced by the Auto-CM software. The main associations shown through the map support an overlap in mealtime rigidity and food refusal between ARFID and ASD, a closer relationship of food selectivity to ARFID compared to ASD, and link ASD to disruptive mealtime behaviours.

Conclusions: To date, this is the first study to apply data mining techniques in the comparison of individuals with ASD and ARFID. There is a notable overlap in feeding issues between the populations. Such difficulties appear to be more relevant in ARFID, independently of the effects of restricted/repetitive behaviours and the sensory profile, whilst disruptive behaviours can occur in both populations, but this is more relevant in ASD.

Notes:

(1)IRCCS Istituto delle Scienze Neurologiche di Bologna, UOC Neuropsichiatria dell’Età Pediatrica, Centro Regionale per i Disturbi della Nutrizione e dell’Alimentazione in Età Evolutiva, Bologna, Emilia-Romagna, Italy, (2)Dipartimento di Scienze Mediche e Chirurgiche (DIMEC), Università di Bologna, Bologna, Emilia-Romagna, Italy, (3)Autism Research Unit, Villa Santa Maria SCS, Tavernerio, Como, Italy