Participants

We recruited participants (N = 35) from an undergraduate research participant pool and from the general community using flyers, postings to online social media platforms, and advertisements sent to professionals who treat eating disorders. Inclusion criteria were: (a) at least 18 years of age; (b) current or lifetime anorexia nervosa, bulimia nervosa, binge-eating disorder, or other-specified feeding or eating disorder; and (c) lifetime ARFID preceding the development of a subsequent eating disorder (i.e., anorexia nervosa, bulimia nervosa, binge-eating disorder, or other-specified feeding or eating disorder). In sum, eligible participants were comprised of individuals who retrospectively met criteria for ARFID and later developed another eating disorder following their course of ARFID.

MeasuresNine Item ARFID Screen – Lifetime Adaptation

The Nine Item ARFID Screen (NIAS; [23]) is a self-report questionnaire assessing avoidant/restrictive eating patterns characteristic of the three ARFID profiles (i.e., sensory sensitivity [ω = 0.80], fear of aversive consequences [ω = 0.91], lack of interest in food/eating [ω = .84Footnote 1), which each represent a subscale. To capture lifetime history of ARFID, we modified NIAS items to query about lifetime avoidant/restrictive eating (e.g., instead of “I am a picky eater,” we asked respondents to rate the statement “In my lifetime, I was a picky eater”). Each item on the NIAS is rated on a scale ranging from 0 (Strongly disagree) to 6 (Strongly agree). Subscale scores range from 0 to 15, with higher scores indicative of greater levels of avoidant/restrictive eating within that profile. We used Burton Murray and colleagues’ [26] cutoff scores of ≥ 10 for sensory sensitivity and fear of aversive consequences, and ≥ 9 for lack of interest in food/eating to screen for potential ARFID symptoms.

Structured Clinical Interview for DSM−5 – Research Version (SCID−5−RV) - Feeding and Eating Disorders Module

The Structured Clinical Interview for DSM-5 – Research Version (SCID-5-RV; [27]) is a semi-structured interview used to confer DSM-5 diagnoses. Participants completed the SCID-5-RV Feeding and Eating Disorders Module after a positive screen on the NIAS to confirm lifetime ARFID diagnosis and assess subsequent eating disorder diagnosis. Rather than assessing for current ARFID, we keyed questions to assess lifetime symptoms (e.g., rather than “In the past month, have you been uninterested in food in general or have you kept forgetting to eat?”, we asked participants “Were you ever uninterested in food in general or did you ever keep forgetting to eat?”), consistent with the SCID-5-RV’s assessment of past disorders in other diagnostic categories (e.g., past major depressive disorder). We obtained age of onset for ARFID and the subsequent eating disorder using the SCID-5-RV and queried participants about the month and year of the onset of each disorder (utilizing the techniques we describe below to mitigate memory errors). We used age, month, and year of ARFID and other eating disorder onset when setting up the Longitudinal Interval Follow-Up Evaluation – Eating Disorders Module (LIFE-EAT-3; 28).

The Longitudinal Interval Follow-Up Evaluation – Eating Disorders Module

The Longitudinal Interval Follow-Up Evaluation – Eating Disorders Module (LIFE-EAT-3; [28]) is a semi-structured interview that assesses the presence/absence and relative severity of diagnostic features of feeding/eating disorders over a pre-specified length of time determined by the study purpose. The pre-specified length of time for the LIFE-EAT-3 for this study varied depending on ARFID and subsequent eating disorder ages of onset derived from the SCID-5-RV. For instance, if an individual reported that their ARFID onset at age eight years in January 2000 and their anorexia nervosa onset at age 13 years in June 2005, the interview would span the inclusive 5-year period ranging from ages 8–13 years. Using the LIFE-EAT-3, we dichotomously assessed the following cognitive and behavioral symptoms of eating disorders by rating them as present or absent: body image disturbance, overvaluation of shape/weight, fear of gaining weight or becoming fat, lack of recognition of seriousness of low weight, food avoidance for reasons related to shape/weight, fasting, excessive exercise, objective binge-eating episodes, subjective binge-eating episodes, self-induced vomiting, laxative use, and diuretic use. Specific details pertaining to the administration of the LIFE-EAT-3 are outlined below (see “The Current Study”).

Procedure

We directed interested individuals to fill out a brief, online screening survey in which they completed the NIAS and reported on their eating disorder history using a single question. We asked participants “Has a medical professional ever diagnosed you with an eating disorder such as anorexia nervosa, bulimia nervosa, binge-eating disorder, or other-specified feeding or eating disorder?” to distinguish individuals who had been formally diagnosed with an eating disorder from those who suspected they had an eating disorder without professional confirmation. We left “medical professional” purposefully vague to encompass whomever the participant considered appropriate, whether it be their primary care physician, a medical doctor, or a mental health professional. If individuals appeared eligible based on their responses to the online screener, we invited them to take part in an online study visit conducted via Health Insurance Portability and Accountability (HIPAA) compliant videoconferencing technology. After providing informed consent, author PEK conducted the SCID-5-RV Feeding and Eating Disorders Module to confirm eligibility, establish ARFID and subsequent eating disorder diagnosis, and ascertain ages of onset of ARFID and subsequent eating disorder diagnoses to facilitate LIFE-EAT-3 set-up. Author PEK then conducted the LIFE-EAT-3. We gave participants the option of receiving research participation credit (if applicable) or $20 for their participation. The University of Wyoming Institutional Review Board approved all study procedures.

Retrospective assessment of eating disorder symptoms

Considering that individuals with eating disorders often experience diagnostic shift [12,13,14,15], reliable retrospective assessment to ascertain diagnosis is critical. Fortunately, retrospective assessment of eating disorder symptoms is common [1, 29]. A major pitfall of retrospective assessment, however, concerns the extent to which respondents accurately recall and report events that occurred in the past. When individuals try to recall past events, memory errors may occur: events may be completely forgotten, events may be remembered as occurring farther back in time than they actually did (forward telescoping), and/or events may be erroneously remembered as having occurred more recently than they did [30, 31]. Fortunately, there are techniques that can be implemented to help ameliorate memory errors.

Three such techniques are bounding, the use of landmark events, and the timeline follow-back approach. Bounding is a technique that helps reduce forward telescoping errors [32]. Rather than specifying the length of time within a reference period (e.g., “Have you experienced a binge-eating episode within the last six months?”), the assessor provides the respondent with specific dates pertaining to that reference period (e.g., “Have you experienced a binge-eating episode since March 15th?”). Landmark events similarly help bound a reference period by providing respondents with salient context for that reference period [30]. Landmark events can be elicited by both the assessor (e.g., public events, such as 9/11, the summer Olympics, the Covid-19 global pandemic) and the respondent (e.g., private events, such as the start of the school year or a relationship breakup). Using landmark events to assess a reference period of interest provides a salient context for relevant symptoms and behaviors, resulting in more accurate retrieval of memories [33]. Finally, the timeline follow-back approach is commonly implemented to assess alcohol use disorders [34]. The assessor uses a calendar to help orient the respondent to the assessment period and asks the respondent to report on their daily alcohol consumption forward to the present, one day at a time. Although ratings are made on a week-by-week basis, the respondent is not queried about each week. When stability in a symptom is detected, the assessor inquiries about “change points” (e.g., “When did that symptom change?, Did that occur before or after Christmas?, How long has this been true?”) and makes ratings based on those change points. The timeline follow-back approach demonstrates good test-retest reliability in clinical and community samples [35]. Together, the combination of bounding, landmark events, and the timeline follow-back approach improves the accuracy of retrospective reporting over approaches that do not implement these techniques. Retrospective eating disorder assessments utilize some of the aforementioned techniques to mitigate erroneous memory reporting [1, 36, 37]. To that end, we implemented these gold-standard interviewing techniques in the current study to aid with retrospective recall.

The Current Study

We began by setting up a calendar with the date of the assessment inputted. The rest of the calendar populated the dates going back as long as indicated by the participant; that is, from the age of ARFID onset to the age of subsequent, other eating disorder onset. We then bounded the reference period by providing the month and year of age of onset for ARFID and the subsequent, other eating disorder diagnosis (e.g., September 2001 to October 2012). Next, we identified any public landmark events and yearly recurring events that occurred during those years and noted them on the calendar. We then inquired about any private events that would help orient the participant to the period under assessment [1]. Once the calendar was complete, we used the LIFE-EAT-3, tracing eating disorder symptom forward to the present, eliciting change points in the participant’s eating disorder trajectory. We coded each symptom dichotomously for each month, with 0 = Absence and 1 = Presence.

Statistical analysesSample characterization and clinical characteristics for those presenting with restricting versus binge-spectrum eating disorders following ARFID history

We computed descriptive statistics to characterize the sample sex, race, age at study presentation, ARFID age of onset, ARFID NIAS profile scores , subsequent eating disorder age of onset, subsequent eating disorder diagnosis, and duration (years) between ARFID and the subsequent eating disorder diagnosis. We conducted independent samples t-tests to compare individuals with restricting eating disorders (n = 25) and binge-spectrum eating disorders (n = 10). The independent variable for each analysis was diagnostic group (i.e., restricting eating disorder or binge-spectrum eating disorder) and the dependent variables were ARFID age of onset, ARFID NIAS profile scores (i.e., sensory sensitivity, fear of aversive consequences, lack of interest in food/eating), subsequent eating disorder age of onset, and duration between ARFID and subsequent eating disorder.

We conducted a binary logistic regression to examine whether ARFID age of onset, the three ARFID NIAS profiles, and duration between ARFID and subsequent eating disorder were associated with a greater likelihood of a restricting eating disorder (compared to a binge-spectrum eating disorder).Footnote 2The five predictor variables were simultaneously entered as covariates. The binary criterion variable for the logistic regression was restricting ED = 0, binge−spectrum ED = 1.

What eating disorder symptoms commonly emerge in the trajectory from ARFID to a subsequent eating disorder?

To elucidate which eating disorder symptoms occurred most frequently, we computed descriptive statistics to assess the presence of each eating disorder symptom on the LIFE-EAT-3, grouping symptoms into two clusters of cognitive and behavioral symptomsFootnote 3. We present results for the overall sample and by restricting and binge-spectrum eating disorders separately.

In what order do eating disorder symptoms emerge in the trajectory from ARFID to a subsequent eating disorder?

We computed descriptive statistics to ascertain order (e.g., for each participant, which symptom onset first, second, third, etc.) and time (years) to each eating disorder symptom onset for the overall sample and by restricting and binge-spectrum eating disorders. For each participant, the month and year of ARFID age of onset was coded as Week 0. Each symptom was coded as present on the first week it onset. We took the average of each symptom onset (e.g., fasting onset at Week 52 for one participant, Week 104 for another, etc.) to compute time to the onset of each symptom following ARFID age of onset. This approach elucidated in what order eating disorder symptoms typically occurred (i.e., common pathways for the development of symptoms), and how long it took for each eating disorder symptom to occur following ARFID onset.

When Do Eating Disorder Symptoms Emerge in the Trajectory from ARFID to a Subsequent Eating Disorder?

Next, we utilized a series of paired samples t-tests to compare time (years) to symptom emergence between eating disorder symptoms. This approach allowed us to map the average trajectory and timeline from ARFID age of onset to subsequent eating disorder age of onset. We included censored data at the earliest possible unobserved symptom onset [39]. This approach increased statistical power because it ensured that every participant contributed to each comparison. Further, censoring data at the earliest possible unobserved symptom onset is a conservative approach because it assumes that although the symptom has not yet occurred, it will do so at the soonest possible opportunity (i.e., the assumption that the symptom of interest will develop immediately after the assessment period). When a participant did not experience a symptom by the last possible observation point, that symptom was coded as present just after the last possible observation point; in other words, the data point was imputed, a common method for addressing missing data in paired samples t-tests. Of note, we only censored data for the purpose of increasing power to detect differences in paired samples t-tests; all other data presented (e.g., descriptive statistics in tables) are uncensored.