A Scale to Assess Factors Inuencing Treatment Initiation in Patients With Anorexia Nervosa. Development and Evaluation of the FABIANA-Checklist.

A long duration of untreated illness (DUI) is an unfavorable prognostic factor in anorexia nervosa and often associated with chronic illness progression. The FABIANA- checklist, developed using a qualitative multi-informant approach, includes factors inuencing treatment initiation which are potentially modiable and are mentioned as being relevant by patients with AN. The study focusses on the development and evaluation of the FABIANA-Checklist and aims at providing descriptive data on DUI in a German sample. primary caregivers addition

eating disorders, showed shorter DUIs, faced shorter waiting times and had higher rates of treatment uptake in comparison to regular care. The effectiveness of such programs in routine health care remains, however, questionable, as the effects were mainly found in a subgroup with optimal conditions. In addition, a substantial proportion of potential patients were not reached by the program, underscoring that prevention programs that primarily target the service-related part of DUI and neglect personal factors such as treatment ambivalence may fall short.
A shared shortcoming of previous studies on factors involved in AN treatment initiation is their inclusion of individuals with various eating disorders and their lack of focus on AN-speci c factors [18]. Further, previous studies almost exclusively include adult patients, even though the onset of eating disorders is often in early to mid-adolescence [17]. This is of particular importance given that Neubauer et al. [4] observed a higher DUI in patients with an early onset (< 14 years) in comparison to patients with an onset in adulthood whereas McClelland et al. [23,24] reported higher DUI in young adults (18-25 years) compared to adolescents (<18). These ndings suggest the presence of sub-groups particularly at risk for delayed help-seeking behavior which might at least partially be related to the age of onset.
Moreover, there might be differences in the pathways to care in these sub-groups, with the early onset group being more often externally motivated and more frequently informed about treatment options by social networks than the group with an intermediate or late onset [4]. Against the background of these results, it becomes clear that a more differentiated analysis of factors involved in treatment initiation is indispensable.
Overall, there are few studies on treatment initiation in AN so far. Beside a missing speci c AN focus in the studies that have analyzed potential predictors of treatment initiation in eating disorders, existing studies do neglect the period before the rst contact with the health care system or do not consider the perspective of relatives and primary caregivers. Moreover, little attention has been paid to factors that can be in uenced by preventive measures.
Another issue that arises when attempting to assess those factors is the lack of valid and reliable instruments that would allow quanti cation of their impact and, in the long term, help primary care practitioners to identify AN patients at risk for an unfavorable illness course [17,25]. Validated questionnaires exist for particular barriers as e.g. the Universal Stigma Scale [26] or the Perceived Barriers to Psychological Treatment Scale [27,28] Based on a systematic review, Ali et al. [29] developed a scale for assessing hindering factors of treatment initiation. However, psychometric parameters for this scale are not yet available [17].
To our knowledge, there is currently no validated instrument that combines the assessment of potential predictors of treatment initiation and focusses on patients with AN. Against this background, the primary aim of the present study was to develop an instrument for the assessment of factors involved in AN treatment initiation.
The present study is part of the FABIANA-study, funded by the German Research Foundation (DFG) and is divided into three sub-studies [30]). In sub-study 1, we qualitatively identi ed the factors using a multi-informant approach (patients, relatives, primary care physicians). We focused at factors that can be potentially in uenced in order to guide the conception of effective secondary prevention approaches. The present article relates to sub-study 2 and has a threefold objective: 1) It reports on the development and evaluation of the FABIANA checklist in a large sample, 2) allows to quantify factors involved in AN treatment initiation as experienced by patients in the period from onset of illness to start of treatment and 3) describes DUI and age of onset (AOO) for a nationwide German sample.

Development of the FABIANA-Checklist
The study has been registered (NCT03713541) and ethical approval was obtained prior to recruitment (PV5108). Instrument and item generation were based on a recommended procedure for mixed-methods studies [31]. Key facilitators and barriers of treatment initiation were identi ed based on 22 qualitative interviews with AN-patients, their relatives and referring health care professionals (sub-study 1 [32]). From this pool, items for the FABIANA-Checklist were derived based on prototypical illustrating quotes from the interviews. Items were considered when they had been 1) mentioned in at least ten interviews, 2) rated as being signi cant factors for treatment initiation by the patients and 3) were considered as potentially modi able by preventive interventions. The latter rating was provided by four researchers, consisting of a professor of psychosomatic and psychotherapy, two post-doctoral clinical psychologists and one clinical psychologist.
For the cognitive pretests, a preliminary sample of 30 items was presented to nine female AN-patients (mean age= 22.8 years, SD= 5.6). Items were rated on a 5-point Likert rating scale, ranging from 1 =does not apply to 5 = does fully apply, according to the patient's experiences during the period between the onset of the AN to the start of the rst psychotherapy.
We used a comprehension probing and information retrieval probing [33] to evaluate the items comprehensibility and recallability. After the cognitive pretests, ve items were excluded and another seven items were excluded after statistical item and reliability analysis. Exclusion was based both on statistical parameters, such as low item-total correctedcorrelation, low variance, skewness or kurtosis of the item, redundancy and parameters of relevance and modi ability. A total of 18 items were included in the further analyses.

Dimensionality
On one hand, in creating the checklist, we aimed to capture factors in uencing treatment initiation as differentiated and as speci c as possible using a multiinformant bottom-up approach. On the other hand, the use of a differentiated checklist in further statistical analysis is associated with the risk of α-error accumulation. For the latter reason (e.g. the planned analysis of the predictive effect of FABIANA-Checklist items on the DUI), a reduction of dimensionality is desirable. For this reason, we decided to subject the checklist to an exploratory factor analysis before proceeding to investigate potential predictors of the DUI (sub-study 3).

Hypotheses on validity
We expected correlations of the FABIANA-Checklist items with perceived support from health care providers, the social environment and general societal factors. Assumptions for construct validation were made for each item separately. For items related to experienced support for treatment initiation by the health care system (items 8-16), we hypothesized correlations with subjectively perceived collaborative care. Correlations with perceived social support were expected for items which include concern or concrete help from relatives or the social environment (items 1-2 and 4-7). For the FABIANA-Checklist items relating to societal factors (items 3, 17-18), as the in uence of printed or social media or stigmatization, we assumed relations to factors of illness perception, mainly illness representations and the perceived control. Correlations were assumed to be positive, for items that were negatively poled, polarity was reversed. A single negative correlation was expected in relation to item 18 and the mention of media in uence as a cause of illness.
The instruments used to operationalize the constructs are listed under measures.

Data collection
Data were collected between July 2018 and June 2019 in 11 cooperating in-and outpatient centers who provide specialized psychotherapeutic treatment for eating disorders. Inclusion criteria were an age at or above 14 years, female gender and typical or atypical anorexia nervosa diagnosis. We included patients who were either currently in their rst AN treatment or who sought their rst psychotherapeutic AN treatment within the last 12 months. Psychotherapeutic treatments were de ned by a minimum duration of seven days in inpatient care or ve consecutive sessions in an outpatient setting.
After obtaining written informed consent from the patients or their legal guardians, eligible patients received the assessment battery with the indication to refer to their rst psychotherapeutic AN treatment. Concerning memory effects (e.g. recall biases) especially in self-reported utilization of health care services, literature provides no su cient evidence on the optimal recall period [34]. However, it is recommended to use periods of three or six months periods when frequently used services are surveyed while salient visits and rarely used medical care services seem to be accurately reported over a longer period [34]. We assume the commencement of a psychotherapeutic treatment to be a salient and rarely event, justifying the use of a 12-months-period for our study purposes.

Measures
We assessed sociodemographic and clinical data as AN subtypes, comorbid diagnosis and Body Mass Index (BMI, kg/m²), date and age of onset and treatment data as setting, date of treatment initiation. The date and age of onset were assessed in a semi-structured clinical interview (SCID-5-CV [35]). The aim was to explore when the criteria for AN according to DSM-V were fully met for the rst time. With the help of anchor examples and a timeline the patients were supported in giving as precise information as possible about their weight history and the other symptoms of AN.
We operationalized our hypothesis on construct validity of the FABIANA-Checklist by using the following measures.
The PatientAssessment of Chronic Illness Care questionnaire (PACIC-5A [36] is a brief self-administrated instrument to assess whether the patients were provided with patient-centered collaborative care prior to their psychotherapy. The PACIC-5A relates to the chronic care model [37] and measures the extent to which professionals tried to induce behavioral changes in patients [36]. The 5A approach is evidence-based, has achieved widespread acceptance and is considered the most appropriate and psychometrically robust instrument assessing patient experience with chronic disease care [38]. The global sore includes the assessment of present behavior (Assess), patient counselling (Advise), collaborative agreement with the patient about realistic goals (Agree), assisting the patient during her lifestyle changes (Assist), and frequent follow-ups (Arrange).
The short version of the Social Support Questionnaire [39] measures patients perceived and anticipated social support. It assesses the social support that patients experienced in the period between diagnosis and the start of the AN-treatment on a scale from 0 = did not apply to 4 = did fully apply. The unidimensional short version with 14 items shows good psychometric properties and a good internal consistency (Cronbach's α = .94).
The Brief Illness Perception Questionnaire (BIPQ [40]) assesses the patients' illness representations, including cognitive dimensions such as the degree of understanding of the illness, the perceived personal control and treatment control, the experience of symptoms as well as emotional aspects such as e.g.
concerns about and emotional affection by the illness. Items of the short Version of the IPQ can assume values from 0-10. For item validation, we considered personal control, treatment control, illness comprehensibility and subjective illness causes. In addition, we included the open-ended B-IPQ item which records the three major causes of illness (in our case for AN) perceived by the patient. Since we are particularly interested in the in uence of media, we included weather respective answers were given in the open-ended item of the B-IPQ (Yes/No).

Data analysis
To analyze the dimensionality of the FABIANA-Checklist we performed Principal Component Analysis (PCA). Polarity of negatively poled items (1,11,16,17,18) was reversed. Data suitability test included the Kaiser-Meyer-Olkin (KMO) criterion and Bartlett' test for sphericity. We considered components with eigenvalues ≥ 1 [41] and tested for varimax, quartimax and equamax rotation on data. We considered and reported factor loadings of >.30. In the case of cross-loadings on multiple components, we considered the loading that showed the best interpretability.
For item analysis, we considered descriptive data, graphical distributions of the raw values, di culty, and discriminatory power analysis, skewness, kurtosis and reliability analysis. Internal consistency was evaluated by Cronbach's α and values >0.70 were considered satisfactory. For Item-total correlation we used Pearsons's product-moment correlation coe cient (r it ), i.e. the correlation between each item and the respective principal component. The internal consistency of the remaining items (α) indicates the value if the selected items are deleted from the total score.
Construct validity was tested with bivariate correlations. Polarity of negatively poled items (1,11,16,17,18) was reversed. Correlation coe cients were interpreted based on Cohen's d with d= < .30 as a small, d=.30-50 medium and d>.50 as large [42]. Given that correlations coe cients represent effect sizes, we focused on the magnitude of the correlations in the validity analyses. We additionally report p values (one-tailed testing, α < .05). DUI is calculated as the difference between the date of illness onset and the date of rst treatment initiation. DUI and Age of Onset (AOO) are reported in years. Comparisons between adults and adolescents regarding DUI and AOO are calculated using simple t-tests with F-value and p-value, given. All calculations were performed with SPSS 27.

Sample characteristics, DUI and AOO
We recruited 75 female patients with AN. Of those 54 were adult and 21 adolescent patients. Mean age was 21.4 (SD=7.35) ranging from 14 to 61 years. Most patients (89%) were diagnosed with a typical AN and 77% presented the restrictive AN-subtype. The mean BMI was 15.5 kg/m 2 (range=10.6 -23.0 kg/m²; SD=1.96). Most patients (77%) had at least one comorbid mental disorder, diagnosed by SCID-5-CV interview [35]. Table 1 shows the sample characteristics. Data collection resulted in only a few individual missing values. For the FABIANA checklist FSozU and B-IPQ, n=72-75 data sets could be evaluated. A slightly higher number of missing values was registered for the PACIC-5A, with n=62 records included.

Duration of untreated illness and Age of Onset
In our sample we observed an average DUI of 2.25 years (SD=4.33, Mdn=.54). DUI in adolescent patients (M=.49 years, SD=.59, Mdn=.28) was signi cantly shorter compared to adults (M=2.93 years, SD=4.94, Mdn=2.69) with p=.027. In one patient, the diagnostic criteria were not met before treatment, resulting in a negative DUI. A treatment initiation within the rst year after illness onset could be observed in 64% of patients, predominantly in the rst three quartiles (Q1: 17.3%, Q2: 25.3%, Q3: 17.3%, and Q4: 4%). A DUI between one and three years was found in 17.4% of the patients, 13.3% had a DUI between 3 and 7 years and 5.3% had a DUI between 18 and 20 years. Figure 1 shows the distribution of DUIs. The average age of onset (AOO) was 19.15 years (SD=5.18, Mdn=18.73). Adolescents had signi cantly lower AOO than adults (with M=15.03; SD=1.07 and M=20.75, SD =5.26; p=.000).

Dimensionality
We performed PCA on the 18 items of the FABIANA-Checklist. Sample size was adequate (KMO=0.72) and Bartlett's test of sphericity indicated that data structure was appropriate for running PCA (χ 2 =307.26; p<.001). A scree-plot yielded empirical justi cation for retaining six factors with eigenvalues > 1, which accounted for 62.46% of the total variance. Among the tested rotations, the varimax-rotated solution was the most interpretable. The internal consistency of the FABIANA-Checklist was acceptable (Cronbach's α = .76). Table 2 indicates item characteristics and results of the PCA. At the component level, the internal consistency for the rst two factors was in an acceptable range (Chronbach's α= .67-.79). The rst component explained 23.33% of the variance The eight items included aspects of the health care system that patients experienced as supportive and helpful (e.g., trust in the treating physician, the treating physician's competence in the area of eating disorders or good cooperation between different physicians). The second component, explained 9.98% of the variance and included three items related to emotional and practical support from relatives. On the third component, inconsistencies in terms of factor loadings manifested in negative internal consistency. On the one hand, this component re ected the disorientation of the patients and their relatives, not knowing whom to consult for help, and on the other hand it included the patient's attempt to gather information about helpful treatment processes from books or social media. The fourth component (7.71%) included items showing that AN was not trivialized by the patients or their social environment (e.g. AN was not addressed too late or at least one person of the environment understood the need for professional help). Internal consistency of this factor was however insu cient (Cronbach's α=.42). Component four and ve showed single-item loadings, relating on relatives who informed themselves (6.76%) and the absence of negative media in uence (5.82%).

Validity
An overview on all assumptions on validity and corresponding correlations is given in table 2. As hypothesized, patients who reported supportive experiences relating to the health care system perceived more patient-centered, collaborative care prior to treatment initiation, which was re ected by signi cant correlations with the PACIC-5A overall score with 8 out of 9 items of the FABIANA-checklist (with r varying from .24 to .59). The strongest correlations were found for the feeling of trust in the practitioner (item 13, r=.59, p<.001), perceived competence of the practitioner and regular contact after diagnosis (items 12 and 14, both r=.42, p<.001). Moderate correlations were found for the items describing early recognition of AN by the practitioner (item 9, r=.32, p<.005) and directly addressing the AN (item 10, r=.34, p<005). Dealing well with patient's di culties (not trivializing complaints) was, as expected related to perceived care (item 11, r=.35, p< .005). The FABIANA-Checklist item 16, indicating di culties to nd out whom to consider for appropriate help, did not correlate as expected with the PACIC-5A score.
Half of expected relations between items of the FABIANA-Checklist that have proximity to the concept of social support and the total score of the F-SozU could be con rmed. The strongest correlation was found for the concerns expressed by the social environment (item 4, r=.43, p<.001). Early recognition of the need for treatment by signi cant others (item 2, r=.22; p< .01) and encouragement of the patient to seek help (r=.21, p<.01) were signi cant at a lower level. The items regarding whether the relatives directly addressed AN (item 1), informed themselves about the AN in the media (item 7) or provided concrete practical support (item 5), e.g. arranged doctor's appointments or accompanied to doctor's appointments, were not signi cantly related to experienced social support.
The expected correlation between consuming media about successful treatments (item 3) and treatment control (BIPQ), i.e. the belief that treatment might be helpful could be con rmed (r=.23, p<.05). However, the consumption of media about successful treatments was not associated with more personal control (r=.-26, p<.05) or illness comprehensibility. Patients who did not perceive undergoing a psychotherapy as a weakness (item 17) had, as expected, a signi cantly better understanding of their illness (r=.20, p<.05, BIPQ), but we found no positive relation to personal control or treatment control. Patients who reported a lower in uence of their weight perception by media (item 18) had, as expected, a better understanding of illness (r =.20, p<.05) and named less often media in uence as one of the causes of their illness (r=-.31, p<.01).

Items of the FABIANA-Checklist -characteristics and frequencies
For each item, the total range between the minimum and maximum value was used. For ve items (items 2,3-6, and 25), the distribution was left-skewed, i.e., with a stronger tendency to agree, and for two items (items 9 and 15), the distribution was right-skewed. The items with the highest scores all focused on practical and emotional support from the social environment, e.g. if relatives expressed concern about AN (M=4.2, SD=1.02), perceived the need for help (M=3.6, SD=1.52), informed themselves about AN (M=3.5, SD=1.38) encouraged the patient to seek treatment (M=4.5, SD=0.98), arranged medical appointments or accompanied the patients to medical consultations (M=4.1, SD=1.41). The average scores indicate that the primary care provider was more likely to not recognize AN at an early stage (M=2.3, SD=1.40), and cooperation with other providers was more likely to not be as well organized (M=2.5, SD=1.45). In uenced by the media, patients tended to assume that low weight was "normal" (M=3.8, SD=1.28) or that they did not feel the need for help. They were more likely to agree that, prior to treatment initiation they felt, that that seeking therapy was a sign of weakness (M=3.4, SD=1.43). Figure 2 provides an overview on the mean values of all included items.

Discussion
In the present study, we presented the FABIANA-checklist, a questionnaire designed to assess factors that are involved in the process of treatment initiation in individuals with AN. The checklist was tested for the rst time in a large sample (n=75) of patients with AN. Overall, the checklist was well accepted by the patients, as evidenced by few missing values.

Dimensionality
PCA yielded six factors explaining 62.64% of the total variance. Overall internal consistency was acceptable (Cronbach's α= .76). The primary care perceived as supportive and competent (23.33%) and emotional and practical support from relatives (2, 9.98%) were internally su ciently consistent factors. The other components did not reach enough internal consistency or had only one item loading. Results pointed out the heterogeneity of the items and suggested that a differentiated consideration of factors, as initially intended by using a bottom up approach, is useful.

Validity
Hypotheses on construct validity were con rmed for 14 items. The checklist showed expected proximity with measures of social support, health care system support, the understanding of the illness, personal and treatment control and subjective illness causes. At an item level, most of the assumed relationships to the constructs of perceived support from the health care system, social support and illness perception could be con rmed. While the aspects of the FABIANAchecklist that depict emotional support from the environment could be validated with the global score of the FSozU, the expected correlations with the total score of the FSozU could not be con rmed for three items. This may be explained by the fact that these items refer to very disease-speci c aspects of help, i.e. practical help with treatment admission (item 5) or an explicit addressing of AN. Item 7 of the FABIANA-Checklist, on the other hand, is referring more to the support that relatives seek for themselves in literature or counselling in order to be able to support patients in turn. Looking at the correlations of these three FABIANA items (1,5 and 7) with individual items of the FSozU, we nd signi cant correlations with one item each. Thus, there is e.g. a positive correlation between the feeling that the environment addressed AN (item 1) and the FSozU item assessing the presence of people with whom the patient could share joys and sorrows. This indicates that the global social support construct may have been too nonspeci c for validating these items. The PACIC-5A was well suited to validate the items related to experienced assistance on the path to treatment admission. The only non-validated item (item 16) measures the di culty of relatives in order to nd out whom they could consult best for help. This item possibly depicts more of a general lack of orientation in the process of help seeking.

Items of the FABIANA-Checklist
Taken together, the rst results obtained with the FABIANA-Checklist showed, that on average, some factors were experienced more often than others or occurred more frequently in the period before treatment initiation. Patients frequently described emotional and practical support from their relatives or their social environment as "expressed concern from relatives" or "encouragement to seek treatment". Patients reported practical support, including relatives making appointments or accompanying patients to medical appointments. Adolescents in particular, reported that their relatives informed themselves about AN, either from books or in counseling centers. Information about successful treatments was sought by some patients and not by others. Both adults and adolescents con rmed that media had an impact on the fact that they considered their low weight as normal and that they did not felt in need to seek treatment.
The item on early detection of AN by relatives was on average more likely to be approved by the patients than the item referring to early detection of AN in primary care. Patients were more likely to report that they had a physician they could trust but were less likely to agree with items related to their doctor's competence in dealing with eating disorders, organizing cooperation with other treatment providers, or addressing their AN directly. The question of how reliably AN is diagnosed and addressed in primary care needs to be further explored. A recent study [43] indicates that only 61.3% of patients with AN were diagnosed by primary physicians and only 40% were referred to specialized treatment. The authors hypothesized that primary care physicians may be adopting a "watchful waiting" approach here, which is frequent when discussing about diagnosing mental illness in primary care [44]. However, this approach may fall short in patients with AN who require more active support from a multidisciplinary team [45]. Problems across care pathways and the relevance of physicians in early detection of AN has been highlighted in literature [46,47] and preventive measures have been proposed [45]. Other potential correlates of DUI to investigate include physician expertise, empathy, or gender [43], or variables that address the physician-patient interaction (e.g., quality of alliance, compliance).
These initial ndings are rst indications for predictors of DUI and treatment initiation. To quantify the relative importance of the factors assessed by the FABIANA-Checklist and to be able to derive recommendations on early-intervention approaches, the effect of those factors on the DUI should be determined.
This will be the aim of a subsequent study on the basis of a new sample of AN-patients, which will further include the perspective of relatives and primary caregivers in addition to the patient perspective.

Duration of untreated illness and age of onset
Our sample can be considered representative of female patients with AN undergoing specialized inpatient treatment for eating disorders. With an average BMI of 15.5 and a comorbidity of 77.4%, the analyzed population is comparable to other studies [48,49]. We found a mean DUI of 2.25 years (or 27 months) for female patients with AN. The DUI is slightly below the mean DUI for patients with AN of seven countries described by Austin et al. [49] in their recent systematic review (29.2 months or 2.42 years), and similar to another German sample from Hamburg metropolitan area [4]. In the subgroup of adolescent patients, DUI was a little less than 6 months and thus signi cantly shorter than in adults. Mean age of onset was around 19 years, with 15.5 years for adolescent patients and 20.7 years for adults. The shorter DUI (and younger AOO) found in adolescent patients has been demonstrated in other studies [44,49,50,51] and can be attributed to signi cantly lower dispersion in the subsample of adolescents. A closer look at the distribution shows that in both groups the earliest AOO is around 12-13 years. In the adult group, the latest AOO was 41 years.
The nding that slightly more than one-third entered treatment more than one year after the onset of AN is concerning, as a longer DUI is associated with unfavorable outcomes, lower remission rates and higher mortality [9,10,52]. Particular attention should be paid to subgroups for whom the path to treatment is protracted. Our data reveal a small subgroup (5%) with patients seeking treatment for the rst time around the age of 40, or after a DUI of approximately 20 years. Ackard et al. [53], found similar rates for patients aged >40 in inpatient treatment. This subgroup differed from younger adults by a later AOO and a longer duration of illness. Duration of untreated illness (DUI) for this subgroup is often not reported, so that it remains unclear, if patients were particularly ambivalent or reluctant to seek treatment or if previous treatments failed.

Strengths and Limitations
The strengths of the checklist developed within the FABIANA study are the explicit reference to patients with AN, the inclusion of adolescent patients, the continuous consideration of the modi ability of the single items and the item construction. In contrast to a classic questionnaire construction based on theoretical concepts, the FABIANA -Checklist was created using a mixed-method and multi-informant approach, and is thus summarizing aspects that were considered relevant to the patients, their relatives and practitioners. Although this approach can achieve a higher clinical relevance through proximity to the patients' experience, it could be the reason why, we could not nd a completely consistent factorial structure.
One strength of such a bottom-up approach is, that it directly refers to the experience of the people concerned and thus and is associated with greater clinical and practical signi cance. Statistically, however, this type of approach has the disadvantage that the analysis of a large number of factors is associated with the risk of α-error accumulation. As we plan to quantify the in uence of facilitating and hindering factors on the DUI in sub-study III, it will be necessary to control this error statistically.
As the FABIANA-Checklist explicitly refers to the time that preceded the rst psychotherapeutic treatment, it will not allow conclusions on correlates of treatment-seeking in general. The FABIANA-Checklist is based on a large sample compared to qualitative studies [32], which should guarantee that many signi cant factors were included. It is, however, conceivable that less frequently mentioned aspects might have a particularly large impact on the DUI. In addition, emphasis was placed on including items that could be modi ed. While this has the great advantage that concrete measures can be derived from the use of the checklist, the exclusion of non-modi able factors might lead to lower variance explanation when trying to explain the DUI. An advantage over other questionnaires that have been used to assess facilitators and barriers in AN [26,27,28,29] is that, to the best of our knowledge, it is the only list of items explicitly designed for patients with AN that has been psychometrically validated.

Conclusions
Overall, sub-study II of the FABIANA project not only described the psychometric properties of the checklist on factors related to treatment admission of patients with AN, but also provided rst data on the expression of these factors in the population of patients who were currently or recently undergoing their rst specialized AN treatment. Although it is very encouraging that two-thirds of the patients in our sample had a DUI of less than one year, we also know that the chances for a positive treatment outcome decrease with increasing DUI. Thus it is of great importance to further quantify the effect of the FABIANA-factors on the DUI, especially in those patients with a DUI >1 year. In addition to the patients' perspective we plan to include the perspective of primary care physicians and relatives in sub-study III. An additional focus of the subsequent study will be to derive intervention measures to support primary care physicians and relatives in addressing AN at an early stage and supporting the patient to undergo specialized treatment.

Declarations
Ethics approval and consent to participate: Ethical approval for the FABIANA-study was obtained from the Ethic Committee of the Medical Association Hamburg (PV5108). All participants provided written informed consent prior to study inclusion. In case of minor participants, an additional consent from legal guardians was provided.   1 The polarity of the items nr. 1, 11, 16, 17 and 18 was reversed for PCA and construct validity tests; 2 e.g. general practitioner, gynecologist, psychiatrist; r it = item-total correlation, α= Cronbach's alpha if the item is deleted, SK= skewness, KT= kurtosis, (+) positive correlation; (-) negative correlation; 3 IPQ item 3 with 0=absolutely no personal control and 10= extreme personal control; 4 IPQ item 4 with 0=perception that treatment is not helpful at all and 10= perception of treatment as extremely helpful; 5 IPQ item 7 with 0= no illness perception and 10= very clear illness perception; 6 IPQ Item 9 -open item asking for the tree major subjective causes for the illness. We considered all illness causes related to in uences of media; (-) negative correlation expected; * p<.01, ** p>.00; *** p<.001 (one-sided) Figure 1 Duration of untreated illness (DUI). Each bar corresponds to the frequency per three months (annual quarter) Duration of untreated illness (DUI). Each bar corresponds to the frequency per three months (annual quarter). Negative values mean that the diagnostic criteria for AN (according to DSM-V) were met after the start of treatment.