Participants and Procedure
Data for the present study came from 38 young adults who were admitted to the University of California San Diego (UCSD), Young Adult Intensive Family Treatment Program (YA-IFT) between October 2017 and June 2019. Patients who sought out the program were self-referred or referred by other healthcare providers. Criteria for admission to the program included a primary ED diagnosis, attendance with at least one designated primary support person, and medical stability assessed through review of relevant medical information by UCSD program physicians. Participants met criteria for an ED based on the Diagnostic and Statistical Manual of Mental Disorders- Fifth Edition.(42) Diagnoses were made by trained raters administering the Structured Clinical Interview (SCID) for DSM-5.(43) For patients who did not complete the SCID (n =8), diagnoses were ascertained by staff psychiatrists using a semi-structured interview. For the purposes of this study, patients classified with an AN diagnosis included AN-spectrum disorders: AN-R (n = 12), AN-BP (n = 7), atypical AN (n =4), and subthreshold AN (n= 1). Patients classified with BN (n = 8) included one patient with subthreshold BN. Six patients diagnosed with ARFID participated.
The study used a naturalistic design and participants who consented to participate in the study completed a structured clinical interview assessing for DSM-5 diagnoses upon admission and online self-report questionnaires at admission, discharge, and at 12-month follow-up. Participants were compensated with a $40 gift card for survey completion at 12-month follow-up. All study procedures were approved by the UCSD Institutional Review Board.
TBT-S Program Description
YA-TBT-S is a short-term, intensive treatment specifically designed for YA and their parent(s). YA-TBT-S was delivered in a five-day, intensive, multi-family format. The treatment introduces a model of parent involvement that is tailored to developmental stage and addresses eating behavior from a framework of temperament and neurobiology. Additionally, participants received: psychoeducation on temperament and neurobiology; experiential activities focused on explaining neurobiology and building effective YA-parent relationships; skills training; dietary support; and in-vivo, therapist-assisted coaching during 21 therapeutic meals and snacks.
Outcome Measures
Study Acceptability
An acceptability measure was designed to assess satisfaction. The measure consisted of 22 items assessing degree of satisfaction with the overall program, specific components of the treatment including neurobiology exercises and education, parent involvement, participants’ perceptions of feeling equipped to recover, and multi-family group support. Participants rated satisfaction on a 5-point Likert scale (1= Strongly Disagree to 5 = Strongly Agree) and completed the survey at post-treatment. Cronbach’s alpha was 0.92.
Study Adherence
Questions were devised to assess adherence to specific treatment recommendations that are considered unique and fundamental to the treatment program, including neurobiology activities and education, parent involvement, and behavioral contracting. Participants self-reported adherence to specific treatment components at 12-month follow-up by reporting on the frequency of use based on a 5-point scale ranging from Never to Always.
Client Satisfaction Questionnaire (CSQ-8)
The present study used an 8-item version of the CSQ(72) to assess satisfaction with treatment length, quality, and delivery. Higher scores indicate greater satisfaction, with scores above 26 indicating high levels of satisfaction. Patients completed the CSQ at post-treatment. Cronbach alpha was 0.74.
Body Mass Index
Height and weight were measured at admission and discharge (height at admission only) by a staff nurse and were used to calculate BMI (kg/m2). Self-reported weight was used to calculate BMI at follow-up. Measured weight and self-reported weight at baseline were strongly positively correlated, r (25) = .98, p = <.001, providing support for the use of self-reported data at follow-up.
Eating Disorder Examination Questionnaire
The Eating Disorder Examination Questionnaire (EDE-Q; Fairburn & Beglin, 1994) is a 31-item self-report questionnaire used to evaluate the presence and severity of eating pathology during the previous 28 days. The EDE-Q is a widely used measure with good validity and reliability. The scale renders a composite score, the Global EDE-Q score, which was used in the present study as the measure of general ED pathology.(45) The measure was modified at discharge to assess for eating pathology over the past seven days due to the short-term nature of the program. Response items were re-scaled to converge with the seven day time course (0 = no days to 6= every day). The EDE-QS, a short-form version of the EDE-Q with a modified response scale to assess change over a shorter time course, has demonstrated good reliability and validity, and performs similarly to the EDE-Q. (46) Cronbach’s Alpha for Global EDE-Q was 0.95. The EDE-Q also assesses counts of eating disorder behaviors, which were used in the present study to assess bingeing and purging frequencies. In order to compare across all time points, discharge scores for items measuring frequency of binge and purge behaviors were then multiplied by four to ensure the time period assessed was equivalent to that of pre-treatment and follow-up.
Clinical Impairment Assessment (CIA)
The CIA is a 16-item self-report measure of the severity of psychosocial impairment due to ED features in the past 28 days.(47) Like the EDEQ, the CIA was modified at discharge to reflect the past seven days. The measure assesses impairment in primary domains based on eating pathology. Cronbach’s alpha was 0.98.
State-Trait Anxiety Inventory- Trait subscale
The State-Trait Anxiety Inventory- Trait subscale (STAI-T)(48) is a 20-item self-report measure assessing the presence of trait anxiety. Items were modified at discharge to reflect the past seven days. Cronbach’s Alpha was 0.98.
Remission Criteria
The recovery criteria established by Bardone-Cone and colleagues were used to classify remission status among participants with AN and BN.(49) Specifically, full remission was rigorously defined as having a BMI greater than 18.5 kg/m2, no binge eating or purging for the past 28 days, and an EDE-Q Global score less than 1 SD above the mean for young adults ages 18 to 22 (i.e., 2.91).(50) The partial remission criteria were the same as the full remission criteria with the exception of an elevated EDE-Q Global score. Because ARFID-specific measures were not included in the study, participants with ARFID were classified dichotomously (i.e., remitted or not remitted) based on their BMI and CIA scores. ARFID remission was defined as having a BMI above 18.5 kg/m2 and a CIA score below the clinical cut-point.(74) Remissions rates were reported as a percentage of the total sample, as remission status of participants who did not complete surveys could not always be discerned. Participants who did not complete admission or discharge surveys but were objectively underweight at these time points were classified as not remitted.
Statistical Analyses
Independent samples t-tests and chi square tests were used to examine patterns of missing data and predictors of loss to follow-up. McNemar exact tests examined differences in both the proportion of participants who were hospitalized pre- and post-IFT and the proportion of participants reporting a good or poor outcome at admission, discharge, and follow-up.
Linear mixed models were used to evaluate changes in continuous outcomes variables over time (admission to follow-up). All participants, including those who dropped out of IFT, were included in analyses. To account for missing data, full information maximum likelihood estimation was used. Repeated measurements of dependent variables (EDE-Q Global, CIA, bingeing, purging, and STAI) nested within participants were modeled at level 1. Given variability in the timing of follow-up assessments and the aim of examining symptomatology over time, time was modeled continuously and centered at IFT admission (admission = day 0). ED diagnosis and the interaction between ED diagnosis and time were modeled at level 2. To account for participants being nested within treatment weeks, IFT week and the interaction between IFT week and time were also modeled at level 2. Age and length of illness were included as covariates in all models.
Model fitting was conducted in stages, and subsequent models were evaluated using log likelihood, Akaike’s information criterion, and Bayesian information criteria. Quadratic terms (i.e., time squared and time cubed) did not improve model fit and were removed from final models. Final EDE-Q Global, and STAI trait models represent random intercept, fixed slope models, while the final CIA model was modeled as a random intercept, random slope model. All participants, regardless of diagnosis, were included in the STAI and CIA models. BMI and EDE-Q Global models examined subsets of participants based on the relevance of these outcome variables to their ED diagnosis. Specifically, low weight participants (i.e., BMI less than 18.5 kg/m2) with AN and ARFID were included in the BMI model, and participants with AN and BN were included in the EDE-Q Global model. Correlation coefficients describing the magnitude of relationships between predictors and dependent variables in the multilevel models were calculated by taking the square root of the following equation: t2/(t2+df). To interpret interaction effects, Cohen’s d effect sizes were calculated for pairwise comparisons.
To examine changes in binge eating and purging frequencies over time, generalized linear mixed models fit with a Poisson distribution and log link function were attempted. Despite attempts to revise the fit of the models, model convergence issues (i.e., the final Hessian matrix not being positive definite) precluded the use of multilevel modeling to measure changes in these behaviors over time. As such, mean bingeing and purging frequencies at admission, discharge, and follow-up were reported among participants with AN-BP and BN.
To approximate the number of subjects needed to adequately power our multilevel models, a power analysis was conducted for a mixed repeated measures Anova with six groups (i.e., the number of separate treatment groups) and three repeated measurements (i.e., admission, discharge, and follow-up). To detect a medium size effect with 80% power, 54 subjects were needed, suggesting the present study was likely underpowered to examine nuanced interaction effects.