Introduction

Anorexia nervosa (AN) is an illness that affects 0.5–2.2% [22, 24] of the general female population. It can cause somatic complications that are linked to undernourishment and/or vomiting and has a very high death rate, about 6.2 times higher than in the general population [25]. In addition, psychiatric comorbidities, especially major depressive disorders (MDD) and anxiety disorders, are particularly common [13, 15].

Outcome for the patients that are treated remains precarious: only 57% of the patients having undergone treatment are cured 4–10 years after the start of their follow-up, making this illness a public health issue that justifies the need for early instatement of optimal care [38].

In addition, the most seriously ill patients have to be hospitalised because of life-threatening states or the chronicity of the disorder. They may also present numerous psychiatric disorders [15]. Thus, these different characteristics tend to select the thinnest subjects with the poorest prognosis for hospitalisation [3, 16, 19].

The main prognostic factors for the evolution of AN identified in studies on outcome, mixing adolescent and adult populations can be divided into three groups [34]: Positive prognosis (short-term symptom evolution, good parent/child relationships, hysterical personality disorder); Negative prognosis (bulimia and use of purgatives, premorbid developmental or clinical anomalies, obsessive–compulsive personality disorder); No significant effect on prognosis (early onset, short-term hospitalisation, significant weight loss, hyperactivity and restriction, vomiting and high socio-economic status). Anxious and depressive comorbidities were not included in the prognostic factors identified in this review. On the other hand, if we consider only subjects that were hospitalised for AN during adolescence, there are six studies in the literature on outcome [18–21, 35, 39], that evidenced a link between psychiatric comorbidities and the persistence of AN. There have been three studies focusing on the impact of anxiety or depression comorbidity on AN outcome, including the chronology of onset [20, 21, 39]; two studies have described the frequency of anxiety or depression comorbidities in relation to outcome parameters [18, 34].

In a multicentric study involving 241 adolescents (aged 13.9 ± 1.7) with eating disorders (ED) (of whom 90% suffered from AN) and followed for 6.4 ± 3 years on average, Steinhauser et al. [34] revealed that the known outcome prognostic factors were rare: initial BMI, adherence to treatment, and in particular, the existence of a psychiatric comorbidity. Thus, 40% of the teenagers for whom evolution was characterised by the persistence of ED, presented a psychiatric comorbidity (in descending order of frequency: mood disorder, somatoform disorders, OCD, other anxiety disorders, substance abuse, schizophrenia). Wentz et al. [39] then showed that, among 51 patients with AN taken from the general population and aged 16 at the start of follow-up, there was a tendency towards the existence of an OCD preceding AN, with a poorer outcome after 18 years of follow-up (p = 0.079).

In 2001, in a study on 39 patients with AN, hospitalised at 16.2 ± 2 years, Herpertz-Dahlmann et al. [20], found a similar relationship between a persistent eating disorder at 10 years and lifetime psychiatric comorbidity (p = 0.01).

In 2004, Halvorsen et al. (51 out- or in-patients, initially aged 14.9 ± 1.8 years, assessed 3–14 years later) revealed that ED persisted in a significantly larger proportion of patients (67% against 13% of patients no longer presenting ED) when the current depressive disorder was concomitant with AN.

A study on outcome carried out on 748 adolescents hospitalised in Sweden and evaluated 9 to 14 years after onset showed that a longer hospitalisation as well as psychiatric comorbidities worsened the prognosis, with the persistence of psychiatric comorbidities and difficulties in social adaptation [21].

Finally, Keski-Rahkonen et al. [24] have recently shown that, in a study involving 55 women having suffered from AN, those for whom the disorder persisted 6–12 years after the first hospitalisation were more likely to have had a major depressive disorder before AN onset (18.5% versus 2.6%, p = 0.04).

In general, the patients hospitalised for AN during adolescence represent a severe subsample where the prognosis is particularly poor. The determining factors for outcome in the long-term after hospitalisation for AN during adolescence remain uncertain or little known. More particularly, the prognostic value of anxiety or depression comorbidity (very common in these subjects) and the impact of its appearance (concomitant with or preceding AN) have not been studied, or only very little (for OCD). However, in cross-sectional studies (but not in follow-up studies), it has been demonstrated that depression and anxiety were associated with significant difficulties in social integration [14], which is a negative factor in AN outcome. Consequently, most studies evaluating the prognostic factors for AN outcome during adolescence seem to be contradictory, depending on whether or not the chronology of appearance of AN and psychiatric comorbidities is taken into account.

Therefore, our study aimed to investigate the link between lifetime comorbidity of anxiety or depression and the outcome 6–12 years after hospitalisation for AN. We hypothesised that anxiety or depression comorbidities preceding AN were linked to a more negative outcome, even if the known prognostic factors are taken into consideration, while lifetime anxiety or depression comorbidity does not influence outcome. Consideration of these elements as therapeutic targets could help improve a rather dark prognosis for these subjects.

Method

Sample (see Fig. 1)

200 female patients consecutively hospitalised for AN (DSM-IV Criteria) during adolescence between 1996 and 2002 in the Institut Mutualiste Montsouris psychiatric unit run by Pr. Jeammet (Paris) were included. Six patients did not meet the criteria and were excluded, thus making a total of 194 patients. 13 men were excluded from the study because of the small number in the male group. Among the 181 women in our sample (sample concerned by this study), 4 died (2.2%) and 143/177 among alive patients were re-contacted (80.8%): 116/177 (65.5%) women agreed to participate in the study. 17/177 (9.6%) refused and 10 (5.6%) of them never responded; 34/177 (19.2%) participants were lost to follow-up (contact details not up to date). Overall, 97/181 (53.6%) of the total sample were face-to-face evaluated and if we considerer people who died during the study’s period we had information for 100/181 (55.2%) of the initial sample.

Fig. 1
figure 1

Flow chart of sample selection

Full size image

Procedure

The protocol was approved by the Comité de Protection des Personnes (CPP), the Comité Consultatif sur le Traitement en matière de Recherche dans le domaine de la Santé (CCTIRS) and Commission Nationale de l’Informatique et des Libertés (CNIL).

Whenever possible, interviews were conducted face-to-face, except for 22 interviews that were conducted by telephone. Among the 116 women who had agreed to take part in this study, 97 were actually evaluated and 78 filled in the self-administered questionnaires as well which means that 19 patients did not response to the self-reports questionnaires.

Assessment

Data concerning initial hospitalisation

For all the patients hospitalised during the period of recruitment, we had a database of initial information that was made anonymous, gathering the socio-demographic and clinical characteristics from the start of their hospitalisation to their discharge. This database was developed from a systematic questionnaire, set up in 1996, recording personal and family, medical and psychiatric histories, the history of ED and their treatment, hospitalisation conditions (dates, weight contract, re-feeding data), socio-demographic data and treatment plan. Initial diagnoses were established according DSM-IV criteria (collected for research purpose), and ICD-10 criteria (collected for administrative purpose).

Data from the assessment 6–12 years later

The assessment was a questionnaire that took about 1 h to fill in and included:

  • Socio-demographic and clinical data [1];

  • ED symptoms on using the EDI-2 Eating Disorder Inventory [6, 9];

  • Social adaptation using the Social Adaptation Scale—Weissman’s Self-Report [40];

  • Quality of life using the WHO Quality of Life-BREF [41] including 26 questions rated from 0 (perceived bad quality of life) to 100;

  • Somatic history and somatic complications of AN.

In addition, the patients were assessed through a 1-h interview for the following:

  • Clinical data (age, minimum and maximum weight and the corresponding age and stature, hospitalisation history and follow-up treatment after hospitalisation, medical, surgical and dental care history) including BMI or Quetelet’s index (weight in kilograms/height in m2) was calculated to describe the subjects’ nutritional state [31];

  • Current and past diagnoses of eating disorders (AN and BN), major depressive disorder, anxiety disorders (panic attacks, agoraphobia, social phobia, OCD, post-traumatic stress and general anxiety) using the Mini-Neuropsychiatric Interview (MINI) [26]. The diagnosis for major depressive episode based on the MINI was modified by the exclusion from the diagnostic criteria of the question on weight loss, in order to avoid over-diagnosis.

  • Global clinical assessment: the Morgan-Russell outcome assessment score [23, 28, 29], assessing functioning over the past 6 months on five subscales exploring eating habits, menstruation, mental state, psycho-sexual functioning, socio-economic status; finally the Morgan-Russell [29] general outcome score was calculated on this data.

Judgement criteria

Outcome was assessed by a main criterion, i.e. the Morgan and Russell general score [28, 29]:

The score was adjusted in order to take the patients’ age into consideration, using the INSERM weight/height growth curve [31], in which the 10th percentile represents the anorexic threshold. This led to a three-category classification, as follows:

  1. 1.

    Good outcome: the patient’s weight is above the 10th percentile and there is menstruation;

  2. 2.

    Intermediate outcome: the patient’s weight is above the 10th percentile but there are menstruation anomalies;

  3. 3.

    Poor outcome: the patient’s weight is below the 10th percentile, there is no menstruation and/or there are bulimic symptoms.

We grouped the scores into two categories by aggregating the good and intermediate outcome categories, as it is normally done [8].

The secondary criteria for outcome were: the ability to reach and maintain a BMI above the 10th percentile (AN diagnostic threshold), the existence of ED symptoms (EDI score), the return of menstruation, the number of re-hospitalisations, social adaptation (SAS-SR score), Quality of Life (WHOQOL-BREF score), mortality rate [standardised mortality ratio (SMR)].

Prognostic factors for outcome

Among those cited by Steinhausen et al. [34], we retained the following factors: early age at onset, duration of the evolution, duration of hospitalisation, initial BMI, weight loss, bulimia, premorbid clinical or developmental anomalies, chronicity, socio-economic status, number of previous hospitalisations.

Regarding comorbidities, we examined lifetime panic attacks, agoraphobia, OCD, social phobia, post-traumatic stress, general anxiety and major depressive disorders (MDD). The post-traumatic stress syndrome was not included in the later analyses due to a low prevalence in our sample, too low to form subgroups (according to occurrence premorbid, simultaneous or posterior to AN).

Analyses

For each quantitative variable, we calculated the means and standard deviation.

Spearman’s non-parametric correlations were used for the measurement of association between quantitative factors and quantitative variables. Concerning qualitative variables, we determined the distribution frequencies. We then tested the representativeness of our re-contacted sample against the initial sample. For the same variables, we carried out chi-square tests or Fisher’s exact tests in case of a theoretical number below 5 in the groups. For the 97 female participants re-contacted, we considered the first hospitalisation in the study period in order to examine the prognostic elements before assessing the association between the prognostic factors already mentioned by Steinhausen [34] and outcome, according to the above-mentioned criteria. For this purpose, univariate tests assessing the association of each factor with each outcome criterion were carried out. For the comparison of the means, we used the Student t test. Because of the size of the subgroups (<30 patients), we resorted to Mann–Whitney’s non-parametric tests for the univariate analyses.

We then developed multivariate models in order to test the association between the judgement criteria for outcome and comorbidity for each anxiety or depression disorder. For this purpose, we adjusted on the prognostic factors recovered in the literature that were linked to the outcome criterion using univariate analyses, and on the duration of time from hospitalisation discharge to the outcome assessment. Anxiety and depression disorders considered were put into the model (OCD, social phobia, panic attacks, agoraphobia, general anxiety and major depressive disorders). For each outcome variable to be explained, three regression models were devised: the first model for premorbid anxiety and depression disorders comorbidites, the second model for those present over the whole lifespan, and the third one only for depressive episodes current to AN. Data were collected and analysed using the SPSS 20 program.

The chosen significance threshold was 0.05.

Results

Sample representativeness: comparison of the re-contacted subjects with other subjects

We compared the 97 re-contacted and assessed young women with the other 84 female patients (that were lost to follow-up or not yet assessed), by looking at their initial socio-demographic (including age at admission (i.e. inclusion), social class, type of family) and clinical characteristics (including minimal BMI, BMI at admission, BMI at discharge, duration of amenorrhea, duration of AN, number of past hospitalisations, length of stay, dropout rate, comorbidities, age at onset of disorder) in order to make sure that the sample of re-contacted subjects was representative and that there was no bias. No significant statistical difference was highlighted for any of the studied variables (results on request). Consequently, we consider this sample to be representative of the initial sample (Table 1).

Table 1 Description of the initial sample
Full size table

The 97 young women assessed during follow-up were on average 15.2 ± 2.2 years old at the onset of AN. Their father was from an upper social-professional category in 71.1% of cases, from a middle-class category in 18.6% of cases and working-class in 3.1% of the cases. There was missing data for 7.2%. The women were hospitalised on average at 16.3 ± 2.2 years for 4.2 ± 3.6 months.

At admission to the unit, they had already been ill for 19.6 ± 17.4 months and 28% had been previously hospitalised; their clinical state was worrying, as their BMI was 13.3 ± 1.3 kg/m2.

Concerning their vital status, in our sample we observed a mortality rate of 2.2% (N = 4) for a follow-up duration of 9.05 ± 1.93 years after hospital discharge.

Subject outcome: description (see Table 2)

At the time of the follow-up assessment, the patients’ global outcome was mainly good or intermediate for a little less than two-thirds of the subjects (62.9%, N = 6). BMI was above the AN threshold in 85.3% (N = 72) of cases and menstruation was present in nearly the same percentage of cases (73.2%, N = 71). Only 15.5% still presented AN or BN according to the DSM-IV-TR criteria. Their social adaption was good on the whole (mean score of 1.7 ± 0.43 on Weissman’s SAS-SR scale), as well as their Quality of Life (mean score of 67.05 ± 18.37 on the WHOQOL-BREF scale).

Table 2 Outcome at follow-up 7 to 12 years after hospitalisation
Full size table

Description of the comorbidities (see Table 3)

Among the patients assessed, 83.5% had suffered from at least one lifetime anxiety disorder; for 46.4% of them, the illness started before the onset of AN. In addition, most participants (85.6%) presented an MDE in the course of their lives, half of them before the onset of AN.

Table 3 Lifetime anxiety and depression comorbidities and chronological occurrence in relation to AN onset
Full size table

The most common anxiety disorders were: OCD (54.6%, N = 51), general anxiety (44.3%, N = 43). These disorders were the most frequently ongoing, with 22.7% of cases for OCD, 17.5% for agoraphobia, 16.5% for social anxiety, 6.2% for panic attacks, while no subject presented current post-traumatic stress syndrome. 7.2% of the subjects currently had an MDE.

The chronological occurrence of these disorders is described in Table 3.

Univariate comparisons

We tested the association between each outcome judgement criterion and the prognostic factors recognised by the literature (age at onset, duration of the evolution, duration of hospitalisation, weight loss, bulimia, premorbid clinical or developmental anomalies, socio-economic status), and premorbid and lifetime anxiety and depression disorders. Here is a summary of the most significant trends (p < 0.06).

The patients with a good or intermediate outcome on the Morgan-Russell scale had a BMI at hospital admission higher than those with a poor outcome (13.57 ± 1.13 kg/m2 versus 12.96 ± 1.43; p = 0.005) and had a premenarchal onset of AN much less often (4.9 versus 16.6%; p = 0.054).

For subjects with a low current BMI (below the 10th percentile for age), the duration of their AN evolution before hospitalisation was longer than for those who currently had a good BMI (24.2 ± 16.6 months versus 18.2 ± 17.6 months; p = 0.038), and their BMI at admission was lower on average (12.8 ± 1.45 versus 13.5 ± 1.17 kg/m2; p = 0.015). They had a premenarchal onset of AN more frequently (5/24: 20.8% versus 4/73: 5.5%; p = 0.025) and they were hospitalised more often (1.1 ± 1.5 times for those with a low BMI versus 0.41 ± 0.9 times; p = 0.011).

The patients who were currently menstruating had an average BMI at admission that was higher than those with amenorrhoea (13.6 ± 1.2 versus 12.7 ± 1.30 kg/m2; p = 0.002). They had premenarchal AN for 4/71 (5.6%) cases versus 5/26 (19.2%) for those who currently presented amenorrhoea (p = 0.041) and they presented panic disorder prior to AN for 11/71 (15.5%) versus 0% (p = 0.033) for those still presenting amenorrhoea.

The intensity of current ED symptoms was positively correlated to the intensity of under-nutrition before hospitalisation, based on BMI differential calculations (r = 0.243; p = 0.046). The EDI2 score was higher in case of social phobia before AN onset (75.1 ± 44.4 versus 50.1 ± 37.036), and lifetime comorbidity with social phobia (75.7 ± 38.6 versus 44.5 ± 36.9; p = 0.001) and depression (58.9 ± 40.96 versus 30.8 ± 23.2; p = 0.035).

Concerning the presence of a current ED, no significant link was found with the prognostic factors under study.

Social adaptation was on average not as good for patients with premenarchal onset of AN as it was for those with postmenarchal onset (2.02 ± 0.45 versus 1.7 ± 0.42; p = 0.044). The same goes for patients with comorbid social phobia (1.9 ± 0.46 versus 1.7 ± 0.4; p = 0.009, agoraphobia (score of 1.9 ± 0.42 versus 1.7 ± 0.42; p = 0.012) or comorbid major depressive episode (1.8 ± 0.43 versus 1.5 ± 0.31; p = 0.033), over the whole lifespan.

Quality of life was not as good in case of history of general anxiety or social phobia before AN onset (58.5 ± 17.8 versus 69.3 ± 17.9; p = 0.031; 60.2 ± 14.4 versus 68.7 ± 18.9; p = 0.058, respectively), and in patients suffering from lifetime social phobia (average score 57.5 ± 16.7 versus 71.8 ± 17.4 without social phobia; p = 0.001), agoraphobia (score of 59.3 ± 17.15 versus 70.99 ± 18; p = 0.009) and general anxiety (score of 61.6 ± 19.6 versus 71.5 ± 16.1; p = 0.021).

Patients that were re-hospitalised were younger at the onset of AN (14.1 ± 1.7 years versus 15.5 ± 2.2 years for those who had not been re-hospitalised).

Multivariate models

We then carried out multivariate models to assess the links between the outcome judgement criteria and comorbidity (preceding AN, lifetime and concomitant with AN) and each anxiety or depression disorder, by adjusting on the prognostic factors from the literature that were significant or that tended to be associated (p < 0.06), as described above in the univariate analyses, and also adjusting on the lapse of time between hospital discharge and assessment. (Tables 4 and 5 give a summary of the models identifying the factors, the other results >N are available on request).

Table 4 Results of the multivariate analyses, highlighting the prognostic factors for outcome, anxiety and depression comorbidities before AN onset
Full size table
Table 5 Results of the multivariate analyses highlighting the prognostic factors for outcome, lifetime anxiety and depression comorbidities
Full size table

Anxiety and depression comorbidities before AN onset (Table 4)

Global outcome (Morgan and Russell general score) was 3.7 times more often worse in case of an OCD preceding AN (this model accounts for 16% of the variance).

A pathological BMI (below the 10th percentile) at follow-up assessment was 11 times more frequent in case of a premorbid history of OCD, and BMI at admission was significantly lower. This model accounts for 35.1% for the variance.

If the patient had menstruation at follow-up, BMI at admission was significantly higher and the time since hospital discharge was significantly shorter. In case of premorbid social phobia preceding AN, menstruation was 5 times more likely to have returned. This model accounts for 24.3% of the variance.

The models assessing outcome based on other criteria did not reveal any significant links with the prognostic factors under study.

Lifetime anxiety and depression comorbidities (Table 5)

A current BMI below the 10th percentile was 14.1 times more frequent in case of premenarchal AN (p = 0.019). This model accounts for 24.9% of the variance.

If the patient had menstruation at follow-up, BMI at admission was significantly higher than for the reverse situation. This model accounts for 16.9% of the variance.

Eating symptoms were more disrupted (EDI-2) when associated with lifetime social phobia comorbidity. This model accounts for 18.4% of the variance.

Re-hospitalisation was more frequent when the age at AN onset was younger. This model accounts for 16.8% of the variance.

Finally, lifetime social phobia and agoraphobia were associated with a lower Quality of Life score (WHOQOL-BREF). This model accounts for 18.3% of the variance.

Mood disorder comorbidities concomitant with AN

No outcome judgement criterion was significantly linked to the existence of major depressives episode that occurred exclusively during AN.

Discussion

The majority of participants from our sample (84.5%), assessed 6–12 years after hospitalisation discharge, no longer presented AN; 4 patients still suffered from AN, 11 presented BN and four patients had died (2.1%). Outcome was good or intermediate for 62.9% of cases, according to the Morgan and Russell outcome assessment score and 85.3% had a BMI above the pathological threshold (10th percentile expected for their age). In terms of psychiatric comorbidities, more than 80% of participants presented at least one lifetime anxiety disorder or major depressive episode, and for half of the cases, at least one of these disorders occurred before AN onset.

Among the anxiety and depression disorders under study, the presence of a comorbid disorder has a significant impact on the prognosis only if it was an OCD, linked to a much poorer global outcome—as assessed on the Morgan and Russell scale—and a much poorer nutritional state—BMI below the AN diagnostic criteria—at the follow-up assessment (independent from BMI at admission). On the other hand, most lifetime anxiety and depression comorbidities did not have any impact on global outcome or on the subjects’ nutritional state at follow-up. Social phobia is the only symptom that changes the subjects’ quality of life. It was also linked to more intense ED symptoms.

Among the factors that are usually described in the literature, we found in the different models encountered that re-hospitalisation was more frequent when age at AN onset was younger. Participants presented a BMI below the AN diagnostic criterion more frequently at follow-up in case of premenarchal onset and lower BMI at admission.

Outcome and cure of eating disorders

It appeared that our sample, concerning subjects suffering from very severe forms of AN if we consider their BMI at admission, had one of the least unfavourable evolutions. In our sample, 4 patients died. This percentage is comparable to the mortality rate observed in the literature for adolescents or young adults suffering from AN, and requiring hospitalisation. The mortality rate was 2.9% in Steinhausen’s 2003 study and 1.2% in Halvorsen’s study [18], significantly more than in the general population (0.4%).

However, the good or intermediate outcome in our sample (62.9%) seemed to be less favourable than that reported in Steinhausen et al., their [34] review, where the cure rate was 47.1%, the improvement rate 32.4% and chronicity 19.7% for a follow-up duration of 4–10 years. Nevertheless, the categories of outcome they used are definitely not comparable to ours, as they are defined as: global, normalisation of the core symptom characteristics of anorexia nervosa, i.e., involving weight, menstruation, and eating behaviour, and psychiatric diagnoses other than eating disorders. Besides, the majority of participants (84.5%) in our study do not currently present ED, which is better than the 70% in Steinhausen’s study [35]. The frequency of the return of menstruation (73.2% of patients in our study against 81% in the study by Steinhausen), and the BMI levels were comparable (BMI above the 10th percentile for age) (75.3%) versus (83%).

Psychiatric comorbidities and the chronology of their occurrence

In comparison with studies on outcome published up to now, the lifetime prevalence rate of 80% for anxiety disorders was higher, as were the prevalence rates for each of these disorders (32 vs 14.5% [12] for agoraphobia, 40 vs 8.1% for panic attacks, 45 vs 33.9% for social anxiety [17], and 85.6 vs 67.7% [12] for major depressive disorder. However, it is difficult to compare our results with those from the literature, as no study has focused on the same disorders, or the same number of disorders, nor on the same samples or the same tools. For instance, Halmi et al. [17] found a lifetime prevalence of 62.9% of at least one anxiety disorder for six different disorders, but they did not consider general anxiety, or post-traumatic stress, and their study was based on a population of female patients, only 79% of whom had been hospitalised, and for whom mean duration of AN was 10 years. The clinical severity of the patients considered in our sample led to hospitalisation, and this is probably responsible for a selection bias [3], which could explain the frequency of the comorbidities observed. In addition, comparisons with the literature should be cautious, as there is only little data and few samples, which was for instance the case with panic attacks, investigated in only three studies on outcome that included 34–62 subjects.

The chronology of occurrence of the disorders according to AN onset has seldom been described in the literature and we have not found anything relating to this topic in studies on outcome. To our knowledge, only one study on anxiety [13] and another on depression [33]—on a small number of subjects (14 subjects per group for anxiety and 23 subjects for depression)—has taken an interest in the chronology of occurrence. At least one anxiety disorder before AN onset was observed in 75% of cases and 65% for social phobia, which was fairly similar to our findings. However, concerning other types of anxiety, the number of participants—varying from 1 to 10—makes any comparisons difficult. This study is therefore the largest to this day on the question of chronology in a study on outcome.

Regarding the impact on outcome of comorbidities before AN onset, our results confirmed the exploratory result obtained by Wentz et al. [39], demonstrating that outcome tended to be not as good after 18 years of follow-up in case of premorbid OCD (p = 0.079). This partly validates our first hypothesis: the existence of an OCD (but not of any other anxiety or depression disorder) preceding AN is associated with a worse outcome when follow-up duration and other factors influencing outcome are taken into consideration. On the other hand, the existence of lifetime OCD does not have the same impact.

In the light of these elements and the literature data, we can see that comorbidity of AN and OCD is a heterogeneous phenomenon that can be broken down into several entities that differ with respect to their inherent impact on the prognosis or the moment their association is observed. Thus, the existence of an OCD during AN does not have any impact on outcome [37]. It is perhaps because these OCD are in reality artefacts and could be the reflection of the increasing intensity of obsessive symptoms in undernourished AN patients. We also know that obsessive symptoms (not relating to eating behaviours) observed during under-nutrition phases decrease with re-feeding [27, 30]; the same observation was made for social phobia [5]. Pollice et al. [30] hypothesised that obsessive symptoms persisting after re-feeding, rather than a state linked to under-nutrition, could be a trait, unlike the symptoms that disappear during re-feeding. Depressive comorbidity before AN onset or lifetime did not evidence any links with outcome despite the fact that we were careful to remove criteria of weight loss and loss of appetite from the diagnostic criteria for depressive episodes [7]. The supposed link between MDD and AN has been studied previously more in depth than the link with anxiety disorders. But the chronology of recurrence has not been examined in studies on outcome.

Finally, the fact that the return of menstruation was five times more frequent in case of premorbid social anxiety could be explained by the fact that this subgroup of patients had a BMI 1 kg/m2 above the others, allowing the menstruation cycle to resume. The existence of a higher BMI for subjects with social phobia can be linked to the positive association of ED such as hyperphagia with excess weight and social phobia as mentioned by Gendall et al. [11].

Limitations of the study

The main limitation of our study concerned the retrospective mode of data collection, in particular for the diagnosis of comorbidities. Because we based ourselves on this data to determine the age at onset and comorbid anxiety and depression, there was a risk of recall bias. However, the principle is the same for both disorders, and because of the one-year discrepancy between the start of the two disorders, we thus avoided confusion over the respective order of their onset. The small size of our sample given the number of prognostic factors considered limited the power of statistical analyses.

Reconstitution of the cohort confronted us with the difficulty in re-contacting all the participants after an average of 9 years, especially as on account of their age, they tended to be geographically mobile; as a consequence, only half of the subjects were contacted and assessed (54%). However, the existence of an initial database with characteristics of our initial sample enabled us to guarantee the representativeness of the subjects contacted and assessed.

Conclusion

In conclusion, OCD observed during the acute phase of AN could be, at least for some, a form of artefact linked to a state of under-nutrition [33]. On the other hand, OCD observed before the onset of AN could be different, they could actually be real OCD with an early onset before 15 years of age, as described by Geller et al. [10], and with a much worse prognosis. This concurs with the hypothesis according to which OCD preceding AN could be the reflection of pre-existing vulnerability traits, suggesting biological or developmental elements preceding AN and could be an endophenotype characterising a subtype of AN with a particularly serious evolution profile [2, 4, 32].