Towards an understanding of help‐seeking behaviour for disordered eating: Refinement of a barriers to help‐seeking measure

Early intervention in eating disorders (EDs) is hampered by a lack of validated measures of barriers to treatment seeking. The present study examined the factor structure of the Perceived Barriers to Psychological Treatment scale (PBPT) and a combination of PBPT and Barriers to Seeking Help for ED items (BATSH‐ED) with respect to treatment‐seeking for an ED.

treatment.Facilitators included the presence of emotional distress and concerns about health (Ali et al., 2017).A notable limitation of the current literature is a paucity of quantitative studies, together with the lack of validated measures.Innes et al. (2018) addressed this limitation by examining the factor structure of the Perceived Barriers to Psychological Treatment scale (PBPT; Mohr et al., 2010) in a sample with disordered eating.
The PBPT is a 27-item scale designed for individuals with mood disorders, and comprises of eight-factors: stigma, lack of motivation, emotional concerns, negative evaluations of therapy, misfit of therapy to needs, time constraints, participation restriction and access to services.Confirmatory factor analysis (CFA) supported a 25-item 7-factor solution in the disordered eating sample (TLI = 0.94, RMSEA = .05[95% CI: .04,.06]).
The PBPT does not include subscales relevant to EDs, namely denial or the failure to perceive the severity of illness and ambivalence.
Denial of the symptoms or their significance has been consistently identified as a barrier to help-seeking for EDs (Akey et al., 2013;Ali et al., 2020;Becker et al. 2004;Cachelin & Striegel-Moore, 2006;Griffiths et al., 2018).Failure to identify an ED as problematic is associated with lower intent to seek help (Fatt et al., 2021).Similarly, ambivalence (a strong fear of change, including gaining weight or losing the perceived positive aspects of the ED) has been identified as preventing people from seeking help (Gulliksen et al., 2015;Hepworth & Paxton, 2007).
While Ali et al. (2020) identified 40 items representing 15 different barriers for EDs in the Barriers to Seeking Help for EDs (BATSH-ED), there is no report on the psychometric properties of this measure.
The present study examines the factor structure of the PBPT in a female university sample who reported a wide range of disordered eating severity, representing an age group in which emergence of disordered eating is common and EDs are elevated compared to the general population (Fitzsimmons-Craft et al., 2019).We then tested two further factor structures; one with PBPT subscales supplemented with a selection of BATSH-ED items (Ali et al., 2020), and one that only retained BATSH-ED items and PBPT subscales that predicted treatment seeking in women who had disordered eating in our sample.We examine the validity of the best fitting structure against other variables, including attitudes to treatment seeking, eating disorder psychopathology, mood (depression, anxiety, and stress), socioeconomic-status (SES) and body-mass-index (BMI).

| Sociodemographic variables
BMI (kg/m 2 ) and SES were assessed, the latter by self-reported postcode used to generate a Socio-Economic Index for Area (SEIFA, 2016) mean score where a quintile score of 1 and 5 represents the most disadvantaged and advantaged areas respectively.

| Eating disorder risk
The five-item Weight Concern Scale (WCS; Killen et al., 1994) yields a score from 0 to 100, with scores greater than 47 demonstrating good predictive validity for the development of an ED (Jacobi et al., 2011;Killen et al., 1994Killen et al., , 1996)).In a previous study of female university students (Zhou et al., 2020), 94% of participants meeting the cut-off score reported engaging in disordered eating behaviours in the previous month; 73.8% received an EDE-Q Global score that was higher than the clinical cut-off (i.e., ≥2.77) for young adult women (Mond et al., 2006).In the current study, a dichotomised low ED risk (≤47) and high ED risk (>47) score was used to examine mean differences and invariance testing across the two eating disorder risk groups.

| Eating disorder symptomology
The 22-item Eating Disorder Examination Questionnaire (EDE-Q 6.0; Fairburn & Beglin, 2008) was used to assess global eating disorder symptomology over the last 28 days on a 7-point Likert scale (ranging from 0 to 6).A higher score indicates either a greater frequency or severity.The EDE-Q has been validated in clinical ED populations and the general population (Berg et al., 2012).In the present study, Omega total of the subscale was .96.

| Psychological distress
The Depression Anxiety and Stress Scales (DASS-21; Lovibond & Lovibond, 1995) consists of 21-items rated on a 4-point Likert scale ranging from 'did not apply to me at all' (0) to 'nearly every day' (3), with higher scores reflecting greater negative affect.Use of the three subscales have been validated (Henry & Crawford, 2005) and Omega totals in the present study were .92,.86 and .85respectively.

| Help-seeking attitudes
The 10-item Attitudes Towards Seeking Professional Psychological Help Scale-Short Form (ATSPPH-SF; Fischer & Farina, 1995) has items rated on a 4-point Likert scale ranging from 'disagree' (0) to 'agree' (4), with higher scores reflecting more positive attitudes towards seeking professional help, for example, 'If I believed I was having a mental breakdown, my first inclination would be to get professional attention'.The ATSPPH-SF has been validated in university students and clinical populations (Elhai et al., 2008).Omega total in the present study was .77.

| Treatment seeking status
To determine treatment seeking status, participants were asked: 'Have you previously sought treatment for eating or body image concerns?'.Participants responded with 'yes' (coded as 1) or 'no' (coded as 0).

| Barriers to help-seeking
The 27-item PBPT (Mohr et al., 2010) investigates barriers to individuals attending weekly therapy appointments with items rated on a 5-point Likert scale, ranging from 'not difficult at all' (1) to 'impossible' (5).Higher scores reflect a higher level of difficulty.A total score is derived from all 27 items.A previous CFS (Mohr et al., 2010) supported an eight-factor structure, with four items being repeated in more than one factor (items 13, 20, 25 and 26) and three items being excluded altogether (items 2, 3 and 15), thus resulting in a 28-item 8-factor solution.In the present study, participants were asked 'We would like you to rate the degree to which different kinds of problems might get in the way of seeing a therapist for eating or body image concerns'.Omega total for the total PBPT scale in the present study was .92.
Additionally, 19 items were selected from the BATSH-ED that complemented items not captured by the PBPT (Table 1).The authors collaboratively worked on selecting items from the BATSH-ED based on previous key barrier themes relevant to the EDs (e.g., denial or failure to perceive severity of illness and ambivalence), as well as barriers around lack of mental health literacy around EDs and social support.
Items were rated on a 5-point Likert scale, ranging from 'strongly disagree' (1) to 'strongly agree' (5), with higher scores reflecting greater personal resistance to help-seeking.

| Statistical analyses
CFA was performed with MPlus software version 7.31, using weighted least squares with mean and variance adjustment (WLSMV), for categorical data (Brown, 2015).The following models were tested: Model 1, replication of the original factor structure of the PBPT (Mohr et al., 2010); Model 2, addition of ED relevant items using a selection of BATSH-ED items (Ali et al., 2020); Model 3 retained only subscales shown to be significant in predicting differences between treatment seekers and non-treatment seekers for those women who met the WCS cut-off score.Chi-square values are sensitive to large samples, and nearly always significant (Byrne, 2012), so each model was judged on the following indices: Root-mean-square Error of Approximation (RMSEA), Comparative Fit Index (CFI), and Tucker Lewis Index (TLI) with the following a priori benchmarks: good fit RMSEA <0.10; CFI/TLI ≥0.9 and excellent fit RMSEA <0.06; CFI/TLI ≥0.95 (Schreiber et al., 2006).
Factor invariance between participants displaying low/high ED risk was evaluated by testing three nested models: Configural invariance, metric invariance, and full invariance.The configural invariance model estimates separate factor loadings and item threshold values (cut points between the ordinal responses) between the two risk T A B L E 1 Selected items from the Barriers Towards Seeking Help for Eating Disorders (BATSH-ED; Ali et al., 2020)  3 | RESULTS

| Description of participants
Participants (n = 456) included 261 (57.2%) high ED risk where 25.7% (n = 67) reported previously seeking help for eating or body image concerns compared to 9.8% of the low ED risk group.When comparing high risk treatment seekers, high risk non-treatment seekers, and low risk, both high ED risk groups had a significantly higher BMI than those in the low ED risk group (Table 2).Those in the high ED risk with previous treatment seeking displayed higher ED psychopathology than the other two groups.

| Preliminary analyses
Data were checked for normality (Tabachnick & Fidell, 2012).Across all measured variables there was less than 5% missing data.In addition, analyses were conducted to examine differences between the  Significantly differs from high ED risk with no previous Tx seeking.

| Confirmatory factor analyses
Model 1 demonstrated a good fit and replicated the original 8-factor structure of the PBPT (Table 4).Model 2, which added the two subscales (Denial and Ambivalence), also had a good fit.Model 3 retained only subscales demonstrating a significant difference between treatment seekers and non-treatment seekers for those at high risk for disordered eating maintained a good overall model fit.In a fourth model, items with weak factor loadings (<.4) were removed, leaving a total of 15 items.Using the Bayesian Information Criterion (BIC) which can inform comparative fit of non-nested models that have the same set of observed variables, Model 4 was preferred over Model 3 and was thus examined in the remaining analyses.

| Invariance testing
Invariance testing was conducted by ED risk group (Table 5) showing the fit of the configural models to be acceptable (RMSEA = .06;CFI = .98;TLI = .97).Analyses revealed metric non-invariance, that is, factor loadings could not be constrained to be equal.Therefore, further testing of invariance was discontinued.
T A B L E 3 Means (standard deviations) and correlation coefficients (95% confidence interval) for barriers to help-seeking measures by participant treatment seeking status for high ED risk group Table 6 shows moderate to strong significant positive correlation between the four PBPT subscales for the high ED risk group.
Higher levels of denial were associated with lower levels of Lack of Motivation and higher levels of Ambivalence; higher levels of denial were associated with lower levels of depression and eating disorder psychopathology.On the other hand, the Ambivalence subscale displayed moderate correlations with all four PBPT subscales.All relationships between barriers factors and attitudes to help-seeking were in the expected direction, with the strongest relationships being found with Negative Evaluations of Therapy and Ambivalence.

| Concurrent validity
In the high ED risk group concurrent validity was examined using a logistic regression with treatment seeking status as the categorical dependent variable and BMI, global ED psychopathology, depression, anxiety entered in Step 1, followed by the six barrier factors in Step 2, explaining 22.3% of the variance in treatment seeking status (Table 7).Global ED psychopathology and denial were the only variables uniquely associated with treatment seeking, where higher levels of disordered eating and lower levels of denial were associated with higher likelihood of treatment seeking.

| DISCUSSION
The present study aimed to further develop a measure of barriers for treatment seeking in people with disordered eating (Ali et al., 2020;Innes et al., 2018).Overall, findings replicated the original 8-factor structure of the PBPT in our sample, but the best-fitting model retained only significant subscales predicting treatment seeking and when items with weak loadings were removed.This included four PBPT subscales (Lack of Motivation, Negative Evaluations of Therapy, Participation Restriction, Time Constraints), and two ED related subscales: Denial and Ambivalence.Internal consistency for the Ambivalence subscale remained borderline.Despite stigma being the most cited barrier in the literature (Ali et al., 2017;Innes et al., 2017;Regan et al., 2017) it did not predict treatment seeking.However, most of the previous research in this area is qualitative.These findings highlight how the use of a standardized and validated measure of help-seeking in the EDs will allow for more robust investigation of barriers and their predictive relationship to treatment seeking.
The present study established convergent and divergent validity between the developed barriers questionnaire and a range of psychosocial measures.Of note, more denial was associated with lower levels of ED psychopathology, depression, anxiety, and stress.This finding is not surprising as one would expect that the more denial/ failure to perceive the severity of illness endorsed, the lower the selfreported psychopathology.This is consistent with by Couturier and Lock (2006), who grouped a sample of 86 adolescents with anorexia nervosa into 'deniers', 'minimizers' and 'admitters', with the former group to having the lowest ED psychopathology.
The Denial subscale strongest had the strongest negative relationship with treatment seeking and was the only unique association found with this outcome along with ED psychopathology.The initial recognition of eating behaviours as problematic has been argued to be one of the major triggers for help-seeking for an ED (Hepworth & Paxton, 2007), with self-recognition of an ED leading to greater likelihood of seeking treatment (Fatt et al., 2021).
In This study addresses a significant gap in the literature by developing a shorter measure of barriers to treatment seeking that was more applicable for use with disordered eating populations However, further testing to improve this questionnaire is warranted.Future studies should consider the involvement of individuals with lived experience when further investigating this questionnaire, to help generate further items for both the Denial and Ambivalence subscales.
Inclusion criteria were: (1) female, (2) aged 17-25; (3) have eating or body image concerns.All participants volunteered via the Flinders University School of Psychology research pool between July 2020 and December 2021 and received course credit for their participation in the 20-minute online survey.The project was approved by the Flinders University Social and Behavioural Research Ethics Committee (Project Number:1953).In total, 555 survey responses were completed, 99 responses were removed (n = 18 did not meet inclusion criteria; n = 81 duplicate responses), resulting in a final sample of 456 participants.
high ED risk group (n = 261) across treatment seekers and nontreatment seekers for PBPT subscales and BATSH-ED items.This was conducted by contrasting the mean and standard deviations of the two groups on each of the continuous PBPT and BATSH-ED variables using an online effect size calculator (Campbell Collaboration tool: https://www.campbellcollaboration.org/research-resources/ effect-size-calculator.html), which generated r coefficients.The strength of group differences was determined based on regular correlation benchmarks (.10-.30= small association; .30-.50 = medium association; .50-1.00 = large association).Five BATSH-ED items significantly differentiated between treatment seekers and nontreatment seekers ( addition, the six factors identified together with psychopathology measures only explained 22% of the variance in treatment seeking, which indicates other factors may play a role in predicting treatment seeking.These may include individual and demographic characteristics such as duration of illness, ethnicity, gender, as well as impairment caused by the ED.There are limitations of the present study.First, 'treatment-seeking' was measured using a single item.Treatment for ED encompasses a multitude of different providers and settings, and future studies should use standardized measures of treatment/help-seeking or use a more nuanced measure, for example, McLean et al. (2019), as well as investigate the ways in which help-seeking may vary across different diagnostic groups.Second, the present sample was restricted demographically; only females aged 17-25 were included, and COVID-19 pandemic effects may have changed the nature of barriers to treatment seeking during this time.Third, while our mixed sample was similar to that used in the work of Innes et al., 2018, our sample had a small percentage of participants who were in a low ED risk group (with 10% seeking previous help for an ED), future research should further validate this questionnaire in a high-risk sample only.Fourth, the PBPT refers to seeking treatment, while the BATSH-ED refers to seeking help, and future investigations of help-seeking measures should focus on the latter, given the difficulties accessing treatment for EDs post-pandemic (Nuffield Trust, 2022) and the variety of other help available, such as non-government ED helplines, which may encourage a journey to treatment-seeking.Lastly, future studies should validate the PBPT and BATSH-ED in separate ED samples, and further test the factor structure supported in the present study as replication across different sample and diagnoses is warranted.

ACKNOWLEDGEMENT
The authors would like to thank David Mohr and Kathina Ali for their assistance in making their measures available for use in the present study.Open access publishing facilitated by Flinders University, as part of the Wiley -Flinders University agreement via the Council of Australian University Librarians.
, representing the 'baseline' model against which the subsequent two models are compared.The Metric Invariance Model fixes the factor loadings for each item to be equivalent across the two risk groups but allows the item thresholds to differ.The Full Invariance Model fixes both the factor loadings and item threshold values between low ED risk and high ED risk. groups

Table 3
Demographic characteristics, eating disorder symptomology, negative affect, and attitudes to help-seeking for with low ED risk, and high ED risk by treatment seeking status ).This included items 1, 3, 8, 27 and 39 (see Table1).Items 1, 3 and 39 can be conceptualized as 'denial'or 'failure to perceive illness severity', whereas item 27 taps into ambivalence.Three items(27, 17 and 29)were selected to be included as an ambivalence subscale.Item 8 was not selected, as the original PBPT scale already taps into 'not knowing where to find counsellor/therapist'.T A B L E 2Note: Superscripts denote which groups differ significantly from each other in post-hoc comparisons.Abbreviations: ATSPPH, Attitudes Towards Seeking Professional Psychological Help-Short Form; EDE-Q, Eating Disorder Examination Questionnaire; M, mean; n, number of participants; SD, standard deviation; Tx, treatment; WCS, Weight Concerns Scale; %, percentage.a Significantly differs from low ED risk.b Significantly differs from high ED risk with previous Tx seeking.c Bolded subscales and items represent significant group differences between treatment seekers and non-treatment seekers based on correlation coefficients >.10.Abbreviations: BATSH-ED, Barriers Towards Seeking Help for Eating Disorders; CI, confidence interval; PBPT, Perceived Barriers to Psychological Treatment Scale; r, correlation coefficient; Tx, treatment.
T A B L E 4 Confirmatory factor analyses: Model fit indices and internal consistency comparisons Summary of logistic regression analyses for the high ED risk group with treatment seeking status as the categorical dependent variable and the six barrier factors, BMI, ED psychopathology, depression and anxiety as predictors 2 (10) = 41.48,p < .001.Nagelkerke R 2 = .223Note: Significant subscales bolded.Abbreviations: B, unstandardized beta; BMI, body mass index; df, degrees of freedom; EDE-Q, Eating Disorder Examination-Questionnaire; SE, standard error; Tx, treatment; Wald, Wald test; χ 2 , chi-square.