Classification of Children and Adolescents With Avoidant/Restrictive Food Intake Disorder Free

Cases for this study were ascertained through prospective active surveillance via a key informant design by the Canadian Pediatric Surveillance Program.¹⁵  The surveillance study period was from January 1, 2016, to December 31, 2017. Pediatricians from diverse clinical settings (community- and hospital-based) were surveyed monthly and asked to report on any new cases between 5 to 18 years of age who met the DSM-5 diagnostic criteria for ARFID and who presented for the first time in the previous month. Those who reported an incident case received a detailed questionnaire to establish the presence or absence of specific ARFID criteria; associated eating behaviors and symptoms; information on the medical, psychiatric, family, and social history; and the results of the physical examination and management of each case. A previously published study includes a description of the methodology and total sample, including the incidence and patient demographics and characteristics.²  The research ethics boards at SickKids, Toronto, Ontario, Canada, and the Children’s Hospital of Eastern Ontario, Ottawa, Ontario, Canada, approved this study.

An exploratory LCA approach was undertaken with latent class models, taking into consideration 1 to 5 classes. Because LCA is restricted to dichotomous variables, 19 initial differentiating indicators were identified from the available questionnaire data on the basis of expert opinions and previous literature descriptions. The indicators selected were 6 Boolean (yes/no) variables related to the diagnostic criteria¹  (weight loss, failure to achieve expected weight gain, faltering growth, nutritional deficiency, dependence on enteral feeding or oral nutritional supplements, or marked interference with psychosocial functioning), along with 13 additional symptoms (length of illness [LOI] <12 months, somatic complaints, medical hospitalization, heart rate [HR] <60 beats per minute, difficulties swallowing, eating but not eating enough, not initiating eating, diagnosis of autism spectrum disorder [ASD], preoccupation with food/eating, feeding-associated symptoms that preceded the onset of the feeding difficulties [ie, choking or gagging], sensitivities based upon food sensory characteristics, general food avoidance, and the avoidance of certain foods).

The consistent Akaike information criterion, Bayesian information criterion, sample-size adjusted Bayesian information criterion, and entropy were examined and used to determine the most parsimonious model. An iterative process was then used to remove indicators that showed no discriminatory value (largest probability differences across classes of <15%) to retain a minimal set of important indicators to the proposed classes. The theoretical response probabilities for each indicator variable and overall class probabilities were calculated and presented in the results.

Class assignment probabilities were calculated for each of the 207 children and adolescents using the final model from the LCA; these probabilities are referred to as posterior probabilities. Given the pattern of uncertainty observed in the posterior probabilities, children and adolescents were subdivided into 4 groups. For clarity, throughout the remainder of the article, the term “class” will be used when referring to the theoretical findings from the initial LCA, and the term “groups” when referring to the analysis based on posterior probability assignment. Demographic and additional clinical variables were calculated for the groups. Significance testing was performed on variables of interest between groups. The analysis was completed using R 4.0.3 statistical software (R Core Team, 2021) and the polytomous variable LCA package.¹⁶ 

A total of 207 children and adolescents aged 5 to 18 years (mean age = 13.1 years [SD = 3.2; range 5.0–17.8 years]) living in Canada were included.² 

Based on fit statistics and class interpretability, modeling that included 16 signs and symptoms was the most parsimonious model and a 3-class model had the best overall fit (Supplemental Table 2). The 3 classes were named Acute Medical (AM), Lack of Appetite (LOA), and Sensory (S). The LCA prediction class membership probabilities reveal 52.4% probability of belonging to the AM class, 40.7% probability for LOA class, and 6.9% for S class (Fig 1). The AM class was distinct for having increased likelihood of weight loss (92%), a shorter LOI (<12 months) to presentation (66%), medical hospitalization (56%), and HR <60 beats per minute (31%). The LOA class was distinct for failure to gain weight (97%) and faltering growth (68%). In addition, this class had a greater probability of eating, but not enough, and not initiating eating. The S class was distinct from the other classes in that these patients avoided certain foods (100%) and refused to eat because of sensory characteristics of the food (100%).

Posterior probability assignments resulted in 3 groups of participants where the posterior probabilities were strongly indicative of a single class (probability of 1 of the class assignments was >80%), and a fourth group that was composed of children and adolescents with a mixed-class assignment (class assignment probabilities between 20% and 80%). Of the 207 individuals with ARFID, 85.0% fit into 1 of 3 distinct posterior probability groups (Fig 2): 44.4% (n = 92) were assigned to the AM group, 33.8% (n = 70) to the LOA group, and 6.8% (n = 14) to the S group. There were 14.5% (n = 30) of individuals who presented with a mixture of both AM- and LOA (AM/LOA)-group characteristics and were assigned to a fourth group. There was 1 case (0.5%) that did not meet the 80% assignment threshold to be included in any group (indeterminate) and is not included in the remaining posterior probability analysis. Table 1 shows a descriptive analysis and comparison of these groups using additional variables. Observed differences across the AM, LOA, S, and AM/LOA groups were noted for the following variables: sex (P = .0001); age at diagnosis (P <.001), age at onset of symptoms (P <.001), LOI (P <.001), body mass index (BMI) z score (P <.001), HR (P <.001), and weight at presentation (P <.001).

The AM group was observed to have much shorter LOI to presentation (13.49 [± 19.16] months); lower HR (69.9 beats per minute ± 21.99 beats per minute); and higher incidence of choking (20.2%; 18 of 92), vomiting (23.6%; n = 21 of 92), and swallowing difficulties (23.6%; 21 of 92) preceding the onset of feeding difficulties. Anxiety (52%; 52 of 92) and depression (15.7%; 13 of 92) were also observed to be highest among the AM group compared with the other groups. The LOA group appeared to have the lowest BMI z score; lowest percentage treatment goal weight¹⁷ ; and highest proportion of individuals with attention-deficit/hyperactivity disorder (ADHD) (29.2%; 19 of 70) and “other psychiatric disorders” (Table 1). The highest proportion of males were observed in the S group (85.7%; 12 of 14). The S group also appeared to have the highest percentage of treatment goal weight at presentation (101.4 ± 12.8); a younger age at diagnosis (7.79 ± 3.10 years) and age at onset of symptoms (1.11 ± 1.02 years); higher proportion of comorbid medical problems (42.9%; 6 of 14); and gagging (50%; 7 of 14) that preceded the onset of feeding difficulties. The AM/LOA group was observed to have mixed symptoms, including having an average age at diagnosis (13.29 ± 2.09 years) and average duration of illness (31.7 ± 37.8 months) that fell between the 2 groups.

The current study used LCA to examine heterogeneity in the clinical presentation of children and adolescents with ARFID presenting to pediatricians from diverse clinical settings in Canada. Based on fit statistics and class interpretability, a 3-class model (AM, LOA, and S classes) yielded the best overall fit to the data. The 3-class solution bears important similarities to the accumulating work in this area.^3–11  Despite identifying 3 classes in the LCA, our posterior probability analysis suggested a fourth mixed group. This supports the literature that “mixed presentations” may be present among children and adolescents with ARFID and the 3 presentations described above are not mutually exclusive^8,12,18  (Fig 2).

The AM group identified in the posterior probability analysis is most similar to the fear of aversive consequences group described in the literature. Unlike individuals in the LOA or S group, those in the AM group included a higher proportion of individuals who were exposed to a reported potentially acute traumatic event, such as choking or vomiting, that preceded the feeding difficulties.¹⁹  Those in the AM group demonstrated a more rapid onset of symptoms, accompanied by distinct medical complications that required more immediate intervention, including hospitalization, compared with the other groups. In addition, this group had a higher proportion of anxiety. Evidence suggests that anxiety coupled with exposure to perceived acute traumatic events (eg, choking, vomiting) can increases one’s vulnerability to negative responses and may trigger the onset of feeding difficulties.^12,20–22  Although our questionnaire lacked sufficient ability to discriminate whether these potentially traumatic events correlated with the feeding disturbance, we hypothesized that the short duration of illness, higher weight loss, and medical hospitalizations supported the severity of presentation related to these events.

The LOA group identified in this study was most similar to those described by other recent studies as having lack of interest in food or eating, with features of failure to gain weight, faltering growth, eating but not enough, and not initiating eating. This group also had a younger age at diagnosis and longer duration of illness than the AM group, suggesting that feeding difficulties were already present at the onset of puberty, a time of increased energy requirements. Children and adolescents who have insufficient intake because of longstanding low appetite and indifference to food will not meet the increased energy needs of puberty, resulting in failure to gain weight and faltering growth. Another consideration for this group involves the association between early adolescence and increased autonomy and independence, which might predispose youth who resist supervised nutrition to lower intake on account of their lack of drive related to appetite. This group also exhibited high rates of anxiety, obsessive-compulsive disorder, and other mental health diagnoses. The lack of appetite and indifference to food may be compounded or exacerbated by the increased comorbid mental health issues. Further, almost a third of patients had comorbid ADHD (29.2%), raising the possibility that distractibility around meals or stimulant-induced appetite reduction may have further exacerbated their feeding difficulties.²³  A recent study of food preferences, food neophobia, and chemosensation (smell and taste) in children and youth with ADHD found that those with ADHD were more likely to exhibit impaired chemosensation than the non-ADHD group, suggesting impaired chemosensory function might be another factor contributing to diminished eating pleasure and appetite among these youth.²⁴  In addition, reduction in HR during periods of malnutrition is typically the result of increased vagal tone as a method of energy conservation in response to starvation. It has been shown that acute reductions in nutritional intake and rapid weight loss have a more profound effect on metabolic regulation of HR than chronic malnourishment.²⁵  Compared with the AM group, the mean HR in the LOA group was 85.2 ± 20.3 beats per minute, suggesting a more chronic adaptation to inadequate nutrition.

The S group identified in this study was most similar to sensory sensitivity group described in the literature, with a high probability of avoidance of certain foods and refusal to eat because of dislike of certain sensory characteristics of foods. Our data suggest that the S group is a distinct, mutually exclusive group and does not seem to have mixed presentation with the AM or LOA group. Unlike those groups, there were few medical concerns identified in the S group. This group had a male predominance, younger age at presentation and diagnosis, and a longer duration of illness. Approximately half of this group was noted to have a history of gagging predating the onset of their feeding difficulties. Consistent with earlier studies in children with ARFID, sensitivity and fear of aversive reactions to sensations from various parts of the gastrointestinal tract (eg, esophageal sensitivity to feelings of gagging) have been noted to be a prevalent driver of food avoidance.^6,26,27 This may explain why the aversive somatic experience of gagging may be overrepresented in this clinical sample who are presenting to pediatricians for assessment. In addition, this group had a higher probability of a diagnosis of ASD (29%). Previous studies have acknowledged the association between feeding disturbances among children with ASD.^28–34  A recent prospective study of children with ASD demonstrated feeding difficulties in 76% of cases, and that 28% met threshold criteria for ARFID.³⁴ 

The fourth group of individuals presented with mixed AM/LOA characteristics that spanned the continuum between the AM and LOA groups. This group may represent a group of young people with >1 driver of food avoidance, and/or children in the LOA group may migrate to the AM group when they enter puberty and are no longer able to meet the increasing energy needs. In addition, >50% of the AM/LOA group was noted to have comorbid anxiety (58.6%), and a high proportion of cases reported events that might have served as potential triggers for the feeding difficulty (17.2% and 13.8% had incidents of choking and gagging preceding the onset of the feeding difficulty, respectively), suggesting that any of these motivating drivers for feeding disturbance could be exerting impact separately.

To date, limited evidence-based approaches have shown that treatment variations have been suggested to address the needs of children and adolescents with different clinical presentations of ARFID.³⁵  Further systematic treatment studies are needed to investigate the effectiveness of various treatments on these different clinical presentations of children and adolescents with ARFID.

The strengths of this study include the application of the highly rigorous Canadian Pediatric Surveillance Program methodology that enabled national, prospective, and active surveillance by pediatricians from diverse clinical settings in Canada. It represents one of the largest samples of children with ARFID assessed by pediatricians. To our knowledge, this is the first study to use LCA methodology to assess the heterogeneity of patients presenting with ARFID. A few limitations warrant consideration. Cases were identified on the basis of presentation to a pediatrician, which may have increased the likelihood of a medically-based presentation. This may explain a higher probability of cases within the AM group, and further explains the overall medical issues observed in this study. This Canadian pediatric study may not be generalizable to adult populations or other countries with different health care systems. Additionally, although the indicators used were based on the ARFID literature and expert opinion, additional indicators could reveal other sources of heterogeneity among patients with ARFID. Finally, the study questionnaire was developed before empirical data proposed various drivers for feeding disturbances observed in patients with ARFID; specifically, there were few questions that focused on correlates and timing of conditioned negative responses or phobias associated with feeding, types and course of enteral feeds required, and other feeding-specific information (eg, nature and course of swallowing difficulties).

In summary, this study of children and adolescents with ARFID presenting to pediatricians in Canada adds to the growing body of literature suggesting that ARFID is a heterogeneous diagnosis. Although our LCA suggests the presence of 3 distinct classes, corresponding to the 3 clinical subtypes described in the literature, further analysis revealed a mixed group with an overlapping symptom presentation. Clinicians need to be aware of these different ARFID presentations because clinical and treatment needs will vary. Future long-term follow-up studies will provide a better understanding of the overall clinical course of individuals with different ARFID presentations. This information will contribute to a greater understanding of the etiology, comorbidities, and serious risks associated with these various presentations, and will further the development of treatments that target the underlying drivers of feeding disturbance.

We thank Ross D. Crosby, PhD, Vice President for Research, Director of Biomedical Statistics at the Neuropsychiatric Research Institute, and Professor in the Department of Psychiatry and Behavioral Science at the University of North Dakota School of Medicine and Health Sciences, for his advice and assistance with the statistical analyses.

ADHD

attention-deficit/hyperactivity disorder

AM

Acute Medical

AM/LOA

Acute Medical and Lack of Appetite

AN

anorexia nervosa

ARFID

avoidant/restrictive food intake disorder

ASD

autism spectrum disorder

DSM-5

Diagnostic and Statistics Manual, Fifth Edition

HR

heart rate

LCA

latent class analysis

LOA

Lack of Appetite

LOI

length of illness

S

Sensory