Introduction

Anorexia nervosa (AN) is a psychiatric disorder characterized by low weight, a distorted body image, and an ego-syntonic nature that makes treatment challenging [1,2,3,4]. Outcome is often poor and treatment avoidance [5], treatment dropout [6, 7], and post-treatment relapse are common [8]. Furthermore, dissent to undergo treatment is frequent resulting in involuntary treatment (IT), e.g., involuntary admission, detention, nasogastric tube feeding, involuntary medication, or restraint, for between 13 and 44% of patients on inpatient wards [9]. Involuntary admission and detention are similar, because they uphold that the patient has to be admitted or remain in hospital, whereas the other IT measures are more limited interventions that logically also result in the patient not being free to leave, but themselves represent more time-limited tangible interactions with treatment staff.

Several factors have been associated with IT in AN including female gender, both lower and higher socioeconomic status, younger age at first admission, early and late onset age, longer duration of illness or treatment, AN symptom severity in the form of a lower body mass index (BMI) and purging behavior, previous hospital admissions, psychiatric comorbidity (depression, schizophrenia spectrum disorders, autism spectrum disorders, and personality disorders), previous IT, self-harm, a history of abuse, lower psychosocial functioning, and lower IQ [9,10,11,12,13,14,15,16,17]. These studies primarily concern involuntary admission, which to some extent can be considered a proxy for other IT measures [12, 18]. However, the clinical relevance of the factors in identifying patients at risk of IT is limited, as many patients with AN are female and young, and have comorbid psychiatric diagnoses, severe underweight, and purging behavior—and many do not require IT [9, 19, 20]. Moreover, legal routes to IT vary internationally [21], and accordingly, legal factors may moderate associations between clinical factors and IT, thereby complicating cross-country comparisons. Criteria for initiation of IT in Scandinavia are comparable [22]. Initiation of IT in Denmark requires the patient with AN to be in a state equivalent to psychosis and either in need of treatment to aid recovery or to pose a danger to self or others [23].

The literature on psychopathology symptoms associated with IT is sparse [15]. Ayton et al. [16] reported that IT in AN is associated with higher levels of depression (measured using the Beck Depression Inventory-II), whereas they found no association with eating disorder psychopathology (as observed on the Morgan–Russell scale) [16]. Watson et al. [24] found no differences in the scores on the Eating Disorder Inventory, Eating Attitudes Test-26, or Minnesota Multiphasic Personality Inventory-2 for patients with and without involuntary admission. Hence, symptom-level markers of IT still require exploration.

Although several clinical factors have been identified as severity markers increasing the risk of a severe illness course and treatment challenges in AN, they have not been tested in relation to IT. These factors include both eating disorder psychopathology and more general psychopathology. In addition to severe eating disorder symptoms such as severe underweight and purging [25,26,27,28,29], some eating disorder psychopathology symptoms (as measured by the Eating Disorder Inventory-2 (EDI-2)) [30] have been associated with poorer outcome. Specifically, the subscale "maturity fears" has been associated with poorer outcome [25] and inpatient treatment dropout [31], "perfectionism" has been associated with poorer outcome [25], and "asceticism" has been associated with longer hospital stays [32]. A lower level of "body dissatisfaction" has been found to predict better treatment outcome [33]. Furthermore, cognitive features characteristic of AN have been suggested to affect patients' competence to consent to treatment [34], thereby complicating treatment and illness course.

With reference to general psychopathology, psychiatric and somatic comorbidity [35], including obsessive–compulsive personality [26] and depression [28], have been associated with a poorer outcome in AN. "Paranoid ideation" (as measured by the Symptom Checklist-90-Revised (SCL-90-R)) [36] has been found to predict treatment dropout [37]. Patients who are more motivated than others for recovery and display better interpersonal functioning have been shown to have better outcomes [29, 38], whereas family problems (measured as parental expressed emotion) have been associated with poorer outcome [29]. Moreover, therapeutic alliance may be important for treatment and post-treatment outcomes [39].

IT can be difficult for patients, staff, and relatives [40,41,42,43,44] and poses ethical questions [e.g., 45, 46]. Therefore, reducing IT is a work in progress [47,48,49,50,51] and improving identification and prevention of IT a necessity. Specifically, improving the ability of clinicians to identify patients at risk of IT as early as possible is important to intervene and prevent future IT [52,53,54]. In light of this and the limited evidence and research on associations between IT and specific clinical symptoms, we explore associations between IT and clinical factors including symptom-level eating disorder psychopathology (EDI-2 [30]) and general psychopathology (SCL-90-R [36]). We did not generate a priori hypotheses due to the exploratory nature of the study.

Materials and methods

Study samples

This was an exploratory, retrospective cohort study combining clinical data from the Eating Disorder Centre (EDC) at Aarhus University Hospital in the Central Denmark Region (CDR) with nationwide Danish register-based data. Patients were included if they a) had at least one outpatient contact or inpatient admission at the EDC between 2000 and the end of 2016 and b) had at least one inpatient admission with an AN diagnosis (aged at least 6 years at the time of diagnosis) at the EDC or another psychiatric hospital unit in the CDR in the same period. For patients with more than one admission with AN in the CDR the first one was used as the index admission. Four ICD-10 diagnostic codes—F50.0 (AN), F50.1 (atypical AN), F50.8 (other eating disorders), and F50.9 (unspecified eating disorders) [4]—were included using both primary and secondary diagnoses, because they are sometimes used for patients admitted without a prior thorough eating disorder assessment, but with suspected AN [12]. Because this was an exploratory study, we used sequential methodology, with two sets of inclusion and exclusion criteria, yielding two study samples. The first sample includes patients with a follow-up time of 2 years to explore factors associated with IT (yes/no), as a recent study shows that most IT happens within the first 2 years [12]. The second sample encompassed the first sample and extended it by allowing a varying (i.e., shorter) instead of a fixed follow-up time beginning at the index admission, thereby increasing the sample size and statistical power in the analyses of time to event. The difference between the two samples was predicated on this change in the follow-up time frame. Figure 1 illustrates the selection process. As we wanted to examine IT in connection to admissions for patients as young as age 6 and as the Danish Psychiatric Central Research Register started in 1969, we excluded patients born before 1963 (to ensure that patients were not older than 6 years in 1969). In this way, we avoided patients having an index admission before 1969 that would be unknown to us. Only patients who had at least one clinical eating disorder assessment at the EDC were included. For patients with more than one clinical eating disorder assessment within the study period, the first one was chosen. We excluded patients with an admission with AN before 2000, because the Registry of Coercive Measures in Psychiatric Treatment was either non-existent or incomplete in earlier years [55]. In the first study sample, patients with their first inpatient admission with an AN diagnosis in the CDR (the index admission) after 2014 were excluded to obtain a fixed 2-year follow-up period starting at the index admission. The second study sample differed only in that we retained patients with an index admission after 2014, thereby allowing for varying follow-up periods up to 2 years. In both samples, we further excluded patients who had their clinical eating disorder assessment later than 2 months into, or after, their index admission, expecting the scores to be stable within the first months. Moreover, patients who had their clinical eating disorder assessment more than 2 years before their index admission were excluded, thereby establishing a maximum time period of 2 years from assessment to hospitalization because of our interest in those factors associated with the risk of IT at treatment onset and to avoid a longer time span that might dilute any possible associations. Patients who emigrated, were lost to follow-up, or died were censored.

Fig. 1
figure 1

Flowchart of the selection processes

Full size image

The study did not require ethical approval in Denmark. The study was registered with the Danish Patient Safety Authority, Danish Health Data Authority, and Danish Data Protection Agency. It conformed to the Helsinki Declaration [56] except regarding informed consent from patients. In accordance with Danish law, permission to access health record data was obtained from the Danish Patient Safety Authority and approval for using register-based data was obtained from the Danish Health Data Authority.

Data sources

We used the unique national Central Person Register number [57] to link the data. Health record data were collected from either the physical records at the Danish National Archives or the CDR's electronic health record systems and entered into an electronic database, REDCap (Research Electronic Data Capture) [58, 59], hosted at Aarhus University and the CDR's electronic research database, MidtX. Health record data from the different sources were then merged and kept in MidtX before being transferred to Statistics Denmark [60] and linked with data from the national Danish registers. The latter contained demographic data from the Danish Civil Registration System [57], treatment data from the Danish National Patient Register [61] and Danish Psychiatric Central Research Register [62], and IT data from the Registry of Coercive Measures in Psychiatric Treatment [12, 55].

Variables

Clinical data

Patients were identified by the Danish Health Data Authority. Clinical data from the health records of identified patients were retrieved by three clinical assistants and included data on the patients' BMI at index admission, residence (with family, alone, or at a specialized residence), and assessment of patient cooperation (good, neutral, or poor) as generally described at the pre-admission meeting and during the first week of admission. The latter categorical rating was made retrospectively based on a review of text in health records and entered into REDCap by the research assistant, and in case of doubt, group consensus was reached between the research assistants. Data from the clinical eating disorder assessments included self-reported age at onset, BMI at the time of the clinical eating disorder assessment, EDI-2 [30], and SCL-90-R [36] and were retrieved from health records by a clinical psychologist. The questionnaires were routinely conducted as part of a clinical eating disorder assessment at the beginning of the outpatient contact at the EDC. The EDI-2 measures eating disorder psychopathology, consists of 91 questions distributed across 11 scales [30], and has been validated in Denmark [63]. The SCL-90-R measures general psychopathology and consists of 90 questions distributed across nine scales [36]; it has also been validated in Denmark [64]. EDI-2 and SCL-90-R subscales were calculated according to their respective manuals [30, 36]. This included calculating subscales only if a required minimum of items were answered. If one or more subscales were not calculated due to too many missing items, the whole questionnaire was counted as missing. Hence, questionnaires were considered completed when results for all the subscales were available.

Register-based data

All admissions within the follow-up period were examined. We categorized IT events in the same way as Mac Donald et al. [12, p. 2], grouping IT measures into: "involuntary admission; detention; medication and electroconvulsive therapy; nasogastric tube feeding; IT for somatic illness; mechanical restraint including belt, straps and gloves; physical restraint; locked wards; constant observation; and sedative medication." These treatment types are all registered as IT by health care professionals in Denmark if used without consent. For instance, IT for somatic illness refers to treatment of physical conditions against a patient's wishes, such as administration of phosphate and potassium to avoid refeeding syndrome.

We also used the registers to extract information on sex, age at first AN diagnosis, age at index admission, previous admissions (yes/no) with other psychiatric disorders than AN, comorbidity at index admission (yes/no), and previous admissions (yes/no) with IT within 3 years before index admission. We limited the latter variable to 3 years to reduce the number of years with inconsistent or no collection of IT data.

Data analysis

For the first study sample, descriptive statistics, two-tailed t tests, Wilcoxon–Mann–Whitney U tests, and chi-squared tests were used to compare patients with and without IT. For analyses with a fixed 2-year follow-up period, we used multivariate logistic regression to estimate the odds ratios for the outcome (IT) associated with an increase or decrease in our exposures, namely, eating disorder psychopathology (EDI-2 scales) and general psychopathology (SCL-90-R scales). Covariates other than the EDI-2 or SCL-90-R subscales were not used in the multivariate analyses to avoid reducing statistical power. The second study sample provided an increase in sample size and statistical power in the inferential statistical analyses. We again used descriptive statistics and tests to compare patients with and without IT. However, we did not conduct tests of significance for the variables that could be affected by the varying follow-up time such as the total number of admissions. Due to the varying follow-up periods, we used Cox regression analyses to estimate the associations, presented as hazard ratios, between our exposures and outcome. Our exposures were factors from the national Danish registers and health records including the scales of the EDI-2 and SCL-90-R, and our outcome was the time to the first IT event measured from 1 day before the index admission. We marked 1 day before the index admission as the starting date to include cases, where the index admission was involuntary (i.e., the starting date of the index admission and of the first IT event is the same). We conducted uni- and multivariate Cox regression analyses for EDI-2 and SCL-90-R scales, respectively. Again, the only covariates were the EDI-2 or the SCL-90-R subscales. For both study samples, the multivariate analyses of the EDI-2 and SCL-90-R were underpowered; however, we justified conducting them given the importance of identifying potential robust associations. Stata 16 software was used to manage and analyze the data [65]. Results for fewer than five individuals were omitted to ensure patient anonymity.

Results

Patient characteristics

The differences between the two samples were as follows. Two hundred and twelve patients were included in the first sample (see Table 1), whereas the second sample included an additional 66 patients (see Table 2). In the first sample, 56 had been treated involuntarily, 179 completed the full EDI-2, and 165 completed the full SCL-90-R with some subscales of both questionnaires having been answered by more patients. In the second sample, 72 had been treated involuntarily, 229 completed the full EDI-2, and 214 completed the full SCL-90-R, again with some patients answering only some subscales of the questionnaires. In the first sample, there were 186 patients with index admissions at the EDC and 26 patients with index admissions at other psychiatric units, whereas in the second sample, there were 252 patients with index admissions at the EDC and still 26 patients with index admissions at other psychiatric units. 55.4% of patients with IT in the first sample and 56.9% in the second sample had involuntary admission and/or detention as well as other IT measures. The outcome variables in the two samples were similar with patients being somewhat younger in the second sample.

Table 1 Patient characteristics for sample 1
Full size table
Table 2 Patient characteristics for sample 2
Full size table

Comparing patients in the first sample with (n = 56) and without IT (n = 156), the total number of admissions was the only variable that differed significantly, with the IT group having more admissions (p < 0.001) (see Table 1). Comparing patients with (n = 72) and without IT (n = 206) in the second sample, the level of patient cooperation differed significantly (see Table 2). Furthermore, the proportion of patients with IT within 3 years before index admission was higher among patients with IT than those without IT.

Clinical factors associated with IT

In the first sample, using multivariate logistic regression, increased scores on the subscale somatization were associated with increased odds of IT, whereas increased scores on the subscale phobic anxiety were associated with decreased odds of IT (see Table 3).

Table 3 Multivariate logistic regression analyses for involuntary treatment (IT) for sample 1
Full size table

In the second sample, univariate Cox regression did not reveal any separate scale on the EDI-2 or SCL-90-R that was significantly associated with the time to IT (see Table 4). However, when the subscales were analyzed together within each questionnaire using multivariate Cox regression, increased somatization measured by the SCL-90-R was associated with decreased time to the first IT event (see Table 4).

Table 4 Uni- and multivariate Cox regression analyses for time to first involuntary treatment (IT) (in the up to 2-year follow-up period) for sample 2
Full size table

In the second sample, univariate Cox regression revealed that previous admissions with IT within 3 years before index admission and neutral or poor patient cooperation were associated with decreased time to IT.

Discussion

Clinical factors associated with IT

To the best of our knowledge, the present study is one of the first to explore associations between either symptom-level eating disorder psychopathology or general psychopathology and IT. Broadly, we were unable to identify any eating disorder-specific variables associated with increased likelihood of IT or decreased time to IT. By contrast, elevated somatization general psychopathology scores were associated with increased likelihood of IT and decreased time to the first IT event at or after the index admission. Elevated phobic anxiety general psychopathology scores were associated with decreased likelihood of IT. Moreover, patient cooperation (as rated by clinical assistants) and previous admissions with IT were also associated with decreased time to IT.

We had hoped to find specific clinical symptom profiles that could alert clinicians to higher likelihood of requiring IT. To some extent we succeeded by identifying somatization and phobic anxiety. Somatization refers to complaints about somatic symptoms and could reflect subjective distress from physical discomfort or pain [36]. It has been associated with irritable bowel syndrome severity scores in patients with AN [66] and it improves with nutritional rehabilitation [67]. Patients with AN and high somatization have also been reported to show illness denial, demoralization, and alexithymia syndrome [68]. Overall, somatization seems to be a significant severity and distress marker in AN, also in connection to IT, that could alert clinicians of a greater risk of IT. With future research it may prove relevant to consider further interventions to prevent IT such as help with relieving or managing somatic symptoms. Phobic anxiety refers to specific fears such as the fear of crowds, buses, and places [36]. The association between phobic anxiety and the decreased likelihood of IT might be spurious, as it was only significant in the first sample. Alternatively, it may point to the possibility of feeling less exposed and vulnerable when admitted to a specialist ward and, therefore, safer and more willing and cooperative to work toward recovery.

We found that neutral patient cooperation and, especially, poor cooperation were associated with decreased time to the first IT event. This is not an unexpected result, because IT is commonly a consequence of dissent to treatment and it points to the importance of establishing patient–staff cooperation. However, overall, cooperation was low among patients with AN in this study with the majority of patients in both groups (with IT (86%) and without IT (64.1%)) judged to have neutral or poor cooperation with treatment by clinical assistants. With these high base rates of low cooperation, the clinical utility of recognizing poor cooperation as a harbinger of IT is low. Less cooperation has previously been found in involuntarily admitted patients with AN than in voluntarily admitted patients [69].

We also found that previous IT at admissions before the index admission decreased the time to the first IT event at or after the index admission. This confirms our previous result that "an IT event is often followed by other events" [12, p. 7]. General psychiatric research has shown that previous involuntary admissions are associated with increased risk of subsequent involuntary readmissions [70]. IT could breed IT, as suggested by Seed et al. [71], and we speculate that patient–staff dynamics such as staff expectations based on patient history could also play a role in recommending IT measures [72]. Furthermore, patients with previous IT may be a particularly unwell subgroup, who are more likely to require IT to ensure safety. However, few patients (n = 17) had a history of IT before their first admission with AN in the CDR and thus they only explain a minority of the IT cases.

Regarding eating disorder psychopathology, none of the odds ratios measured by the EDI-2 differed significantly between the groups. Although our comparisons were underpowered, the estimates were near unity with fairly small 95% confidence bands, suggesting that eating disorder psychopathology symptoms (as measured by the EDI-2) were not reliable factors associated with the risk of IT. Outliers of very high or very low scores were not analyzed separately, but may be of interest, as some patients with AN report low scores possibly due to illness denial [73] or because they feel better at low weights. Our result corresponds to Watson et al.'s [24] result of no difference between patients with and without involuntary admission when examining their EDI scores. BMI at clinical eating disorder assessment and BMI at index admission were also not associated with the time to IT. Patients with AN and IT have been shown to have a lower BMI at admission than voluntarily treated patients [13, 74], but the difference is generally small and not clinically meaningful [15]. Furthermore, neither previous admissions with psychiatric disorders other than AN nor comorbidity at index admission differed significantly between the IT and no IT groups. We cannot rule out the importance of any specific previous psychiatric disorders affecting the time to IT, because there were too few patients to conduct the analyses. However, previous admissions with diagnoses other than AN, including schizophrenia spectrum disorders, autism spectrum disorders, and personality disorders, have been found to be associated with subsequent IT in register-based studies [11, 12]. A study showed more personality disorders in involuntarily admitted patients than voluntarily admitted patients [69]. Moreover, depression has been associated with IT in AN [10, 15].

We found that more than half of the patients with IT had involuntary admission and/or detention as well as other IT measures. This supports that involuntary admission can to some extent be considered a proxy for other IT measures [12, 18]. In fact, Danish law requires that the conditions for involuntary admission are met for other IT measures to be initiated [23].

Strengths and limitations

The present study has the strength of combining health record data with data from the national Danish registers, making it possible to examine all inpatient admissions in Denmark following a clinical eating disorder assessment at the EDC and index admission in the CDR. Moreover, using the Danish registers to identify patients, as done by the Danish Health Data Authority, and admissions maximized the detection of IT, while minimizing recall and selection bias.

There are limitations to our study. Despite employing a sample of 278 patients, the study was underpowered and we did not correct for multiple comparisons; therefore, the risk of type I or II errors is increased and the results should be interpreted with caution. For patients with an index admission in 2000, the variable previous admissions with IT within 3 years before index admission covered the years in which the Registry of Coercive Measures in Psychiatric Treatment [55] either did not yet exist or was inconsistently used. We included the diagnoses of other eating disorders (F50.8) and unspecified eating disorders (F50.9), thereby making the patient group studied more diverse. However, these diagnoses are sometimes used for patients admitted without a prior thorough eating disorder assessment, but with suspected AN [12], and post hoc analysis showed that only 5.4% of the patients in the second sample did not have a lifetime F50.0 or F50.1 diagnosis in relation to a hospital contact (outpatient or inpatient) before the end of follow-up.

We conducted our time-to-event analyses under the assumption that the clinical symptoms measured by the questionnaires were stable up to 2 months into the index admission. Although studies have found high test–retest reliability over a week for both the EDI-2 and the SCL-90-R [75, 76], this may not be the case over a longer period. However, post hoc analyses showed that less than 3% in both samples had their clinical eating disorder assessment more than a week into the index admission. Furthermore, the varying time between clinical eating disorder assessments and index admissions may have influenced the non-predictive results for the EDI-2 and SCL-90-R questionnaires. Moreover, the quality of clinical assistants' assessments of patient cooperation may have varied. We ensured that clinical assistants had bachelor's degrees in psychology, had an overall introduction to the content of the health records, and focused on patient cooperation, as described in the pre-admission meeting and within the first week of admission. Assistants collaborated and discussed unclear cases when needed.

The Danish registers do not provide information on which diagnosis is associated with an IT event. In cases with comorbid disorders, whether IT was related to AN (with the exception of nasogastric tube feeding) is unknown. Furthermore, the present study was based on a sample of patients with a clinical eating disorder assessment at the EDC and index admission in the CDR, which may limit its external validity. This should be considered when generalizing to other treatment facilities or countries [77] with different practices. For example, a physician in Denmark who has been alerted can propose IT, after which it is validated by a chief physician; a judge is only involved if a patient appeals a complaint [22, 23]. In addition, registration of IT is not systematic in many countries outside of Scandinavia. Thus, it is difficult to compare the use of IT across European and other non-European countries, where variation in legislation [22] likely also contributes to differences in the use of IT. The use of IT is generally comparable in Scandinavian countries, with some variation in the use of specific IT measures [77]. Finally, a number of patients were excluded due to having no clinical eating disorder assessment at the EDC. Patients directly admitted to inpatient treatment without a prior outpatient contact may not have participated in a clinical eating disorder assessment, including the EDI-2 and the SCL-90-R.

Conclusions

In this study, we used sequential methodology yielding two study samples (one encompassing and extending the other), because the literature on IT in AN is still sparse and we hoped to maximize the information gleaned by taking this exploratory approach. Overall, we did not identify a particular eating disorder profile from the EDI-2 that captured the risk of receiving IT. Notably, this suggests that the EDI-2 dimensions associated with illness course and treatment challenges [25, 31, 32] differ from those that influenced the likelihood of receiving IT or time to the first IT event. In terms of general psychopathology, only somatization on the SCL-90-R, which has been associated with more severe symptoms in AN, was positively associated with both likelihood of IT and time to IT, whereas phobic anxiety was negatively associated with the likelihood of IT.

Our findings of poor patient cooperation and previous IT being associated with shorter time until IT raise important clinical issues that need to be addressed using a more in-depth methodology. Patients' non-cooperation is subjective and decisions to use IT may in part be influenced by clinicians’ awareness of the historical use of IT with any given patient. Knowing that a patient underwent IT in the past may increase the likelihood of clinical staff recommending its use again and may also influence ratings of cooperation. Although historical clinical information is extremely valuable for treatment planning, clinicians should be open to the possibility of patients changing over time in both their level of cooperation and their need for IT. Ensuring unbiased evaluation upon readmission could help break the cycle of repeat readmissions marked by IT. Novel approaches to acute AN care such as self-admission [78, 79] and home-based care [80, 81] that have the potential to increase patient and family agency in recovery could mitigate the cycle of repeat admissions with IT that may contribute to demoralization of patients, family members, and clinicians. The fact that the number of associations between clinical factors and IT found in our study was limited points to the importance of extending future research to examine treatment contexts more broadly.

What is already known on this subject?

Improving the ability of clinicians to identify patients at risk of involuntary treatment (IT) is important to intervene and prevent future IT. Clinical factors including eating disorder psychopathology and more general psychopathology have been identified as severity markers increasing the risk of a severe illness course and treatment challenges in anorexia nervosa (AN), but have not been tested in relation to IT. The literature on psychopathology symptoms associated with IT is sparse and symptom-level markers of IT still require exploration.

What does this study add?

We explored associations between clinical factors and IT. This is one of the first studies to explore associations between either symptom-level eating disorder psychopathology or general psychopathology and IT. Somatization and phobic anxiety as measured by the SCL-90-R were positively and negatively, respectively, associated with the likelihood of IT. Somatization, previous admissions with IT, and neutral or poor patient cooperation were associated with decreased time to the first IT event. Eating disorder psychopathology measured by the EDI-2 was not significantly associated with likelihood of IT or time to the first IT event.