The present study identified two subtypes of restrictive food intake disorder in children using LCA based on clinical information, growth chart analyses and sociodemographic parameters available at hospital admission. The clusters found with LCA based on the parameters at admission were consistent with the diagnosis of EOAN and ARFID made at the end of hospitalisation. Compared to previous LCA studies in children with restrictive eating disorders [16, 17], we did not include core symptoms of anorexia nervosa (i.e., fear of gaining weight and disturbed body image) in our variables because these symptoms are often difficult to identify in children at the beginning of treatment. We demonstrated that clinical and sociodemographic characteristics other than the core symptoms of EOAN available at admission to hospital can help obtain a highly accurate diagnosis of children with food intake disorder (only two ARFID patients were misclassified in EOAN group).

Cluster 1, representing 74% of the sample, was mostly associated with a diagnosis of EOAN at the end of hospitalisation. Patients in cluster 1 were more often girls, with a higher socioeconomic level, intensive sports activity, more frequent familial eating disorders other psychiatric histories and more frequently had an identified trigger for the onset of their eating disorder. Although these features are consistent with the classic features of EOAN [8], it should be noted that they are not always present in single patients. In our sample, none of the patients from cluster 1 had associated ADHD but all had a retarded growth curve. These features must be interpreted as a whole to determine the diagnosis. Cluster 2 represented 26% of the sample, had features suggesting a diagnosis of ARFID [7] and was associated with a diagnosis of ARFID at the end of hospitalisation.

Differences between ARFID and EOAN groups were consistent with previous studies and helped characterise patients with ARFID (see Supp Table S1) with the presence of earlier onset of symptoms and more comorbid psychiatric and neurodevelopmental disorders and a higher likelihood of a history of chronic disease [3, 6, 7, 14, 15]. Unlike one previous observational study, we found that children with EOAN more frequently had first-degree psychiatric histories (including a history of ED) than those with ARFID (Kurotori et al., 2019). This difference may be explained by different samples size in the two studies (13 ARFID patients in Kurotori study versus 27 in ours).

To our knowledge, this is the only study that has compared auxological parameters in children with ARFID and those with EOAN. We reported growth retardation in 34% of ARFID patients at hospitalisation, although this was less frequent than in children with EOAN (95%, p < 0.001), it is to be noted that undernutrition can also lead to growth impairment in ARFID patients. In ARFID, the health consequences of reduced food intake may be similar to those in EOAN (hypothermia, anaemia, bradycardia, tooth decay) [15] thus, early and intensive treatment to restore weight is indicated [3].

Accurate diagnosis of children with food intake disorder (ARFID or EOAN) at early time at hospitalisation admission is helpful to define treatment strategy and adequate psychoeducation for patients and families. ARFID is a heterogeneous group that includes various subtypes: chronic low appetite, or limited interest in feeding, fear of aversive consequences (fear of choking or vomiting), severe selective eating and/or food neophobia. A single patient may belong to more than one subtype due to frequent overlap of eviction drivers [4]. In our hospitalisation sample, reported primary drivers of food avoidance for ARFID patients were mainly fear of aversive consequences (48%), low appetite/food interest (30%) and sensory sensitivity (22%). It is particularly useful to identify additional factors that differentiate ARFID patients with acute food restriction from EOAN patients with similar symptoms. Refeeding strategies, dietary diversification targets or weight targets differ between ARFID and EOAN, early accurate diagnosis helps providing adequate and individualised treatment, especially relevant psychotherapy, and allows the healthcare team to better tailor their attitude and treatment goals to the patient.

Our results may be interpreted in the light of several limits. As for many retrospective studies, the reliability of the initial clinical data assessment by different psychiatrist can be questioned, to address this issue, initial inpatient clinical assessments was carried out according to a standardised clinical assessment framework for all patient and retrospective collection was carried out in a standardised manner by trained psychiatrists specialised in eating disorders. Additionally, the appropriate sample size in LCA is not clearly defined, large sample sizes above 250 participants are preferred, our sample remains limited (n = 97) but smaller sample sizes can be adapted in simpler models with larger class differences (Nylund-Gibson & Choi, 2018), our 2 cluster model with clear differences between the two group appear to be reliable. Finally, this study was not designed to explore the heterogeneity of ARFID, more specific question related to ARFID presentations has been recently addressed in larger sample size, confirming a 3 distinct clusters in LCA corresponding to the 3 ARFID subtypes described in the literature, and emphasizing the need for individualised treatment [9].