Introduction

Bulimia nervosa

Bulimia nervosa (BN) is an eating disorder (ED) that was included in the 3rd edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-III) in 1980 [1] after its description by Gerald Russell the year before [2]. The current main criteria for the diagnosis of BN according to DSM-5 are recurrent binges, i.e. consumption of an unusually large amount of food within a short time interval, compensatory behaviours such as vomiting, excessive physical activity or fasting at least once a week for 3 months, and excessive preoccupation with shape and weight [3]. Even though the current version of DSM, the DSM-5 is the most contemporary classification system (issued in 2013 [3]), its use is restricted to research and psychiatric practice in the United States of America (USA). Most countries such as the United Kingdom (UK) use chapter V of the10th revision of the International Statistical Classification of Diseases and Related Health Problems (ICD-10; issued in 1992 [4]) to diagnose psychiatric disorders in clinical practice. Diagnostic criteria for BN according to ICD-10, however, are similar to DSM-5 criteria and include recurrent episodes of overeating in which large amounts of food are consumed in short periods of time, a persistent preoccupation with eating and a strong desire to eat, attempts to counteract the fattening effects of food and a self-perception of being too fat [4].

BN occurs in 1–2% of women with a sex ratio of men to women of about 1:10 [5, 6]. It is often associated with affective disorders (AD) [7], substance use disorders (SUD) [8,9,10,11] and personality disorders (PD) [12,13,14]. It has been hypothesized that these comorbidities—specifically SUD and PD—develop on the basis of an increased impulsivity and affective instability in patients with BN [15,16,17,18]. This view is supported by findings that borderline PD traits are particularly common among patients with BN [19].

Mortality in BN

Mortality rates are reported to be elevated in patients with BN. For instance, in a recently published German study on long-term mortality in EDs which included 1930 people BN [20], the standardised mortality ratio (SMR) for people with BN was estimated at 1.49 indicating that the observed number of deaths in the BN population was about 1.5 times higher than expected in the general population standardised for sex and age. For this study, the research team had included people during their inpatient treatment in a hospital for behavioural medicine with a subspecialty for EDs.

Another example of a study looking at mortality rates has been published by Hoang et al. [21]. They used a dataset of the English national Hospital Episode Statistics (HES), linked with data from death records provided by the Office for National Statistics (ONS). They included all cases, where an ED was recorded as the main diagnosis on discharge of a treatment in hospital. They identified 838 people with BN and obtained an SMR of 4.1 for these patients [21]. This was probably a patient cohort of more severely affected people, as BN patients are usually not treated as inpatients in the UK, because the guidelines of UK’s National Institute for Health and Care Excellence (NICE) for the recognition and treatment of EDs specify that patients with EDs should only be admitted, if their physical health is severely compromised in such a way that they need medical stabilisation or refeeding [22]. In Germany, by contrast, there is a tradition of treating patients with EDs in psychosomatic hospitals, and the so-called S3 guidelines for the diagnosis and treatment of EDs specify comparably low-threshold indicators for inpatient therapy for patients with BN such as a lack of local outpatient treatment opportunities, the severity of the disorder, severe conflicts in the family and the social environment [23].

In a former systematic review and meta-analysis that included 12 quantitative studies and a sum of 2,585 individuals with BN across different countries, SMR was estimated at 1.9 [24] which might be a quite balanced SMR estimation.

There are different potential reasons, why BN is associated with increased mortality. Broadly speaking, increased mortality in this patient group might be related to the physical complications of BN [25] or due to a higher risk of suicidality [26, 27]. Potentially life-threatening complications of BN include gastrointestinal, endocrine and renal disturbances which can lead to severe electrolyte derangements and cardiac arrhythmias [25]. Among others, important factors contributing to the increased risk of suicide in people with BN could be the high level of impulsiveness [26], hopelessness [28] and—as already mentioned—comorbidities such as SUDs, PDs and ADs, like unipolar depression [28], which themselves are also associated with increased mortality [7,8,9,10,11,12,13,14, 29,30,31,32,33]. In the literature, SUDs, PDs, ADs, and anxiety disorders were found the most frequent mental disorders comorbid to BN [34,35,36]. Among those four, SUDs have specifically been reported to contribute to mortality in BN [37, 38], and PDs [39] as well as ADs [40] have each been shown to be associated with increased mortality—independently from the presence of BN, whereas a recent meta-analysis did not reveal any evidence of an increased mortality risk among anxious study participants. The increased mortality in patients with anxiety disorders previously found in small study samples has been considered to be most probably due to comorbid depression [41]. Therefore, we decided to focus on SUDs, PDs and ADs as the most promising potential factors associated with increased morbidity in patients with BN. Other demographic variables which have been associated with mortality in patients with EDs or ADs include age, gender, marital status, socioeconomic deprivation and ethnicity [42, 43].

Aim of the study

In this study, we sought to investigate the influence of major psychiatric comorbidities on all-cause mortality in people with BN in a case register for a large secondary mental healthcare service provider in southeast London, UK, with confounding from demographic and socioeconomic factors considered. Our hypothesis for this investigation was that AD, SUD and PD would have independent effects on increased mortality in people with BN.

Methods

Setting and study design

A retrospective cohort study was conducted in people with a diagnosis of BN. The BN cohort for the analysis was obtained from the South London and Maudsley National Health Service (NHS) Foundation Trust (SLaM) which provides near-monopoly secondary mental healthcare services to approximate 1.36 million residents in southeast London. Patients involved in this study were mainly from the London boroughs of Lambeth, Croydon, Lewisham, Southwark, Bromley, Bexley, and Greenwich. SLaM has used fully electronic records in all its services since 2016 and its NIHR-funded Biomedical Research Centre (BRC) supports an infrastructure for rendering its anonymised records available for research, using the its Clinical Records Interactive Search (CRIS) platform, established in 2008 [44, 45].

Inclusion criteria

The analysed cohort was extracted via CRIS, consisting of individuals who received an International Classification of Diseases (ICD-10) diagnosis [4] of BN (F50.2) within or before the observation window from 1 January 2007 to 31 March 2016. Structured information on diagnosis from drop-down fields in the source record was supplemented by information extracted from open-text fields using a bespoke algorithm generated using the Generalised Architecture for Text Engineering (GATE) software [44]. The exposure of interest was a further documented diagnosis of SUD (F10–F19), AD (F31–33) or PD (F60) at any stage during or before the observation window extracted from CRIS using identical procedures.

Primary outcome

The outcome of interest in this analysis was all-cause mortality occurring from 1 January 2007 to 31 March 2016. Information for each death was collected through nationwide tracing using data from the Office for National Statistics (ONS) linked to the SLaM database on a monthly basis. In the UK, all death certifications are linked by NHS number (a unique identifier for each UK NHS service user) to all healthcare providers.

Covariates

Date of birth (year and month), gender, ethnicity, marital status and residing borough were retrieved from CRIS database. Age in years was calculated from the individual’s first BN diagnosis date in the observation window or on 1 January 2007 if the BN diagnosis preceded the observation window. Recorded ethnicity was classified into three categories: ‘white’ (including ‘British’, ‘any other white background’ and ‘Irish’), ‘black’ (including ‘African’, ‘any other black background’ and ‘Caribbean’), and ‘others/unknown/mixed’. Recorded marital status was classified into four groups: ‘married’ (including ‘cohabitation’, ‘married’ and ‘married/civil partnership’), ‘single’, ‘divorced’ (including ‘divorced’, ‘civil partnership dissolved’ and ‘separated’), and ‘others’ (including ‘others’ and ‘not known’). Borough was classified by the subject’s residence: Lambeth, Croydon, Lewisham, Southwark, Bromley, Bexley, and Greenwich, defined by their primary care trust. Multiple deprivation score is a small-area-level measure of socioeconomic status, based on the individual’s address closest to the diagnosis of BN in the observation window (or 1 January 2007 for those with an earlier diagnosis of BN), covering the following seven components: ‘employment, income, education, health, barriers to housing and services, crime and the living environment’ with specific weightings. The Index of Multiple Deprivation core is a well-established measure which has been widely used as a regional indicator for socioeconomic status in previous studies [46,47,48]. The score is then transformed into percentiles (from 1 to 100) with a higher score, indicating greater deprivation (and thus a lower socioeconomic status in the neighbourhood). In the analysis, deprivation score was grouped into tertiles (33% percentiles).

Statistical analysis

Standardised mortality ratios (SMRs) were used to describe case mortality in relation to the general population. Observed number of deaths in the observation window was treated as the numerator and 5-year age and sex bands were defined to calculate expected numbers for the same bands from the population of England and Wales in 2012 according to ONS data. Since the study cohort was followed for up to 9.25 years, multipliers based on the average follow-up period for each corresponding age-and-sex band were applied for estimating general population denominators.

Descriptive statistics on demographic characteristics and psychiatric comorbidities were first carried out for the study cohort. We then used Cox proportional hazards regression to calculate hazard ratios (HRs) with 95% confidence intervals (CIs) on death as the ‘event’ for outcome within the observation period. Beginning of the follow-up period was defined as the start of the observation period (1 January 2007) if the first diagnosis of BN was given before the window; otherwise this was taken as the date of first BN diagnosis recorded in the window.

Multivariable analyses were also performed to estimate the effect of the considered psychiatric comorbidities as major exposure of interest, controlling for confounding of age, sex, ethnicity, borough, marital status and deprivation score. All the statistical analyses were performed using Stata 12.1 (StataCorp, College Station, Texas, USA) and the statistical significance (alpha level) was set at 0.05.

Ethical approval

Ethical approval was obtained from the Oxfordshire Research Ethics Committee C for the use of CRIS as a source of data for secondary analysis (reference 08/H0606/71 + 5) including the mortality linkage. This study has thus been performed in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments.

Results

Descriptive statistics

A total of 1501 individuals diagnosed with BN were identified from the CRIS system, of whom 1431 (94.4%) were female; the mean age was 28.5 years (standard deviation = 9.5 years), and the age range was 11.9–64.5 years old. More detailed information about the age distribution can be found in Table 1. During the observation window, 18 of the 1501 patients died. The mean age of the patients who died in the observation window was 34.9 years old (standard deviation = 12.9). Further detailed descriptive information on age, gender, marital status, the residential borough, ethnicity, social deprivation and comorbidities is given in Table 2.

Table 1 Age distribution of study participants with frequencies of deaths and psychiatric comorbidities by age groups (N = 1501)
Full size table
Table 2 Univariate analysis on demographics and clinical characteristics for mortality among people with BN by Cox regressions (N = 1501)
Full size table

The recorded causes of death among the 18 people with BN who have died in the observation window were: lethal intoxication from drugs or alcohol (n = 4), suicide (n = 3), circulatory or heart failure (n = 3), respiratory failure (n =  3) and cancer (n = 2). For one deceased person, the cause of death was documented as a result of the ED without further specification, and for three others, the cause of death could not be obtained from the database.

With regard to the deaths in relation to age groups, none of patients between 10 and 19 years (n = 249) of age died. Half of the deaths (n = 9) happened in patients aged 30 and over, even though this age group accounted for only about 35% of people in this study, whereas 65% of the study participants were younger than 30 years. Table 1 provides further details about the frequency of deaths and psychiatric comorbidities in each age group.

Standardised mortality ratio (SMR)

The SMR for our study cohort was 2.52 (95% CI 1.49, 3.97).

Univariate analysis by Cox regressions

Cox regressions revealed significant associations of mortality with older age, male gender and the presence of recorded SUD, AD and PD (Table 2).

Multivariable analyses

Controlling for age, sex, ethnicity, the residing borough, marital status and deprivation score as confounders, multivariable analysis outcomes by Cox regression are shown in Table 3. In these models, recorded PD remained significantly associated with mortality in patients with BN (adjusted HR = 3.36, 95% CI 1.05, 10.73), while associations with SUD (adjusted HR = 3.12, 95% CI: 0.81, 12.08) and AD (adjusted HR = 2.47, 95% CI 0.90, 6.78) were diluted in strength and no longer statistically significant. Detailed information regarding multivariable analysis outcomes for AD, SUD and PD as the major exposures of interest are presented separately in Supplementary Tables 1–3.

Table 3 Multivariable analysis on psychiatric comorbidity for mortality among people with BN by Cox regressions (N = 1501), controlling age, sex, ethnicity, borough, marital status and deprivation score as confounders
Full size table

Discussion

Main findings

Of the 1501 patients with BN, 18 died during the more than 9-year observation period. A more than 2.5-fold increased risk of general mortality was found after age and gender standardization, compared to the general population in England and Wales. Within the univariate analysis of the study cohort, age and male sex was found associated with an increased mortality. With confounders properly considered, the deceased patients suffered from PD significantly more often.

Comparison to previous research

SMRs of around 2 have been reported by previous research [24]. This figure might be even higher in certain populations of people with BN, for example, people within the first year after discharge from inpatient treatment [21].

Also, in accordance with our results, age is usually positively associated with mortality in patients with EDs in the literature [20, 49]. We found an association of male sex with increased mortality in patients with BN. In general, an association of male sex with increased mortality is well characterised in the literature, and potential explanations relate to the different effects of sex hormones [50, 51], the undertaking of hazardous behaviour [51] and the disadvantage of having only one X chromosome [50,51,52,53]. The higher mortality of men among patients with EDs [54,55,56] may be related to the fact that EDs are often considered as a female disorder and thus stigmatised and underdiagnosed in men, and it may also be related to findings of greater general psychopathology in men with EDs [57,58,59,60,61].

In our study sample, the proportions of people with comorbid psychiatric disorders are rather low as compared to similar studies. Whereas we found a documented percentage of PD of about 9% in this group of BN patients (see Table 1), a recent meta-analysis obtained a proportion of 54% of comorbid PDs in patients with BN [12]. Nevertheless, one has to keep in mind that our data did not derive from standardised interviews covering the whole spectrum of psychopathology. In fact, in our study, PD diagnosis means a diagnosis established and documented by the responsible clinician. One can assume that this was only documented, if it was clinically relevant.

Psychiatric comorbidity and mortality

In our dataset, the presence of a PD contributed significantly to mortality in patients with BN even when controlling for covariates. This is in accordance with previous literature, where PD has been reported to contribute to mortality [32]. Associations of mortality with SUD and AD were not significant in the fully adjusted model. Therefore, our discussion will focus on PD.

Even though we would like to keep the discussion of mechanisms to a minimum, since we did not study mechanisms in your paper, we will try to interpret the obtained association between increased mortality and PDs in patients with BN. It could be explained by increased risk-taking behaviour and self-injuries on the basis of high impulsivity, which is often seen in BN [62, 63]. Specific personality traits such as impulsivity could further lead to the use of alcohol and drugs and, in turn, increase morbidity in this patient group [49, 64, 65]. Additionally, the consequences of both BN symptoms, such as vomiting, in combination with substance abuse sequelae, such as mucosa damage in the gastrointestinal system due to alcohol intake, could lead to potentially deadly medical conditions, such as oesophageal adenocarcinomas [66].

Clinical implications

Our results advise to be aware of the increased mortality risk of older patients, patients with PDs, and men with BN. For further clarification of the presence of a PD in clinical practice, screening tools and clinical interviews are available such as the SCID-II Questionnaire (SCID-II-PQ) and SCID-II interview [67] for PDs.

Strengths and limitations

The main strengths of this study are the study design and the size of cohort using CRIS as our data source. CRIS is one of the largest collections of users for secondary mental healthcare services in Western Europe. Using these data, with a retrospective cohort study, we identified a large number of 1501 patients with BN. Eighteen patients with BN died in the observation period, since study subjects were fairly young. Therefore, statistical power is limited.

When considering systematic errors and generalisability of these findings, the research outcomes might be only generalizable to secondary and tertiary health care settings. Our study is a clinical cohort study. It did not include patients with BN who did not seek help, who were solely treated by their general practitioners, who were not correctly diagnosed with BN by their primary health care provider and subsequently referred to SLaM or who did not have access to SLaM due to funding issues.

This clinical cohort study included participants who were already suffering from BN. People in this cohort had not been included into the study before the appearance of BN like they would have been in a population-based cohort study. This means that we do not have a completely comparable control group of people without BN that would only differ in terms of the presence of BN. Strictly speaking, we compared people with BN from South London to the general population of England and Wales to calculate the SMR, because this was the most comparable data available.

Future research directions

In the current project, we calculated the SMR for people with BN, and we identified certain demographic variables and PDs as risk factors increasing mortality in patients with BN. We also reviewed the cause of death for the 18 patients who died in the observation window. Due to the small number of people who died in our study, future studies with a larger sample size and a longer observation period need to statistically confirm, whether lethal intoxication from drugs or alcohol, suicide, circulatory/heart failure and respiratory failure are really the top four killers for patients with BN.

Additionally, it would be clinically relevant to find out, whether death by these causes could be predicted based on certain psychopathological features such as the presence of PD symptoms, based on physical comorbidities or on laboratory parameters. Such clinical chemical parameters predicting lethal intoxication with alcohol could be biological markers of alcohol abuse such as gamma-glutamyl transferase (GGT), mean corpuscular volume (MCV), carbohydrate-deficient transferrin (CDT) or high-density lipoprotein (HDL) cholesterol [68, 69]. Alterations in the biological stress systems and the serotonergic system could be tested as predictors of suicidal behaviour [70]. And serum electrolyte concentrations which are often disturbed in patients with BN due to vomiting [71, 72] could be examined as biomarkers predicting circulatory, heart or respiratory failure.

Once the main causes of death and ways to predict them with a certain probability are established for patients with BN, this knowledge should be clinically applied for the individualised prevention of premature death. Such research projects could involve frequent monitoring of the mental and physical health state as well as the measurement of certain biomarkers. As PDs contributed significantly to the mortality risk of people with BN in the present study, clinical projects to develop and test therapies for patients with both BN and PDs would also seem sensible to find ways to prevent premature death in this patient group. Small clinical pilot studies in patients with BN and comorbid PD using dialectical behaviour therapy (DBT) have already been performed and published [73, 74]. However, the low remission rates and the maintained eating-related psychopathology suggest that those treatments need further improvement [74].

Taken together, future research should establish the main causes of death for people with BN, seek to predict the causes of death by focussing on physical comorbidities and biomarkers and try to develop specific therapies for people with both BN and PDs. This means that epidemiological, biomarker and therapy research need to cooperate to achieve further scientific and clinical progress.