Document resource
WHAT IS ALREADY KNOWN ON THIS TOPIC Eating disorders have a detrimental effect on mental and physical health with an increased risk of premature mortalityThe impact of eating disorders on longer term health is not well understoodGiven the prevalence of eating disorders and rising rates of diagnosis, a greater understanding of the short and long term outcome risks is neededWHAT THIS STUDY ADDS In a large cohort of people with eating disorders with patient records from both primary and secondary care sources, this investigation also looked at adverse physical and mental health outcomes in the long term and cause specific mortality riskAfter an eating disorder was first diagnosed, a marked increased risk of adverse physical health outcomes was found, including diabetes, renal and liver failure, osteoporosis and fractures, compared with individuals without an eating disorderHigh levels of pre-existing and coexisting mental health and behaviour problems, including depression, anxiety, and personality disorder, were found, with a sustained risk of death in the short term and long term for individuals with an eating disorderThe risks of adverse physical health outcomes, as well as mental health outcomes and mortality, extended beyond five years after diagnosis of an eating disorderHOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE, OR POLICY This study highlights the importance of ongoing monitoring for long term physical health outcomes in individuals with a history of eating disordersPrimary care has a key role in recognising and managing these long term health risks through sustained holistic follow-upThe study emphasises the need for a shift in the research agenda towards understanding the development and underlying mechanisms of physical comorbidities in this populationIntroduction Eating disorders are characterised by the development of abnormal eating or weight controlling behaviours and preoccupation with body shape. Based on the Diagnostic and Statistical Manual of Mental Disorders classification system, 1 eating disorders may be classified as anorexia nervosa, bulimia nervosa, binge eating disorder, and eating disorders not otherwise specified. Based on data from the then UK General Practice Research Database,2 the age standardised annual incidence of eating disorders was estimated at 37.2 per 100 000 (during the period 2000-09) among those aged 10-49 years. Based on data from the Secure Anonymised Information Linkage databank of routine clinical records in Wales,3 the incidence was estimated at 26 per 100 000 among those aged 10-65 years during the period 2016-17. In March 2020 to March 2022, the reported incidence of eating disorders was consistently higher than expected (33.1% difference in expected v observed incidence) in female children and young adults aged 10-24 years.4 Although some evidence exists indicating stabilising or reducing incidence rates over time before the covid-19 pandemic,3 5 6 rates have increased over time since the covid-19 restrictions in the UK.4 Referral to specialist clinics and hospital admissions also progressively increased between 2008 and 2018.3 6 7 More recent hospital admission data in England8 9 showed that hospital admissions of children and young people (<18 years) with eating disorders increased by 90% over a five year period, from 3541 in the financial year 2017-18 to 6713 episodes in the financial year 2020-21.8Physical health can be adversely affected by eating abnormalities because of the nature of anorexia nervosa, which is associated with weight loss and malnutrition, and because of purging behaviours characteristic of bulimia nervosa.10 The National Institute for Health and Care Excellence (NICE) guideline NG69 recommends that specialists in eating disorders should assess and monitor the physical health of all patients who have received a diagnosis of an eating disorder.11 Although thresholds for referrals to eating disorder services vary between children and young persons' services and adult services, the recent report from the Nuffield Trust and Health Foundation12 (sourced from Mental Health Services Monthly Statistics-NHS England Digital13) found that 79% of children and young people with urgent needs started treatment within one week and 83% of routine patients started treatment within four weeks in the first quarter of 2023. These figures fail NHS England's target of starting treatment in 95% of patients within one week for urgent patients and within four weeks in routine patients.Also, the short and longer term physical health outcomes of individuals with an eating disorder have not been studied extensively, although high levels of morbidity have been reported.3 In a general practice and hospital admission linked dataset from Wales, comorbid diagnoses and the frequency of prescribed drug treatments two years before and three years after a diagnosis of an eating disorder were presented. A relatively small cohort of 5777 individuals were eligible for hospital secondary care linkage, fulfilling matching criteria but with a relatively short outcome assessment at three years.3Psychiatric illness may coexist with eating disorders. In the Secure Anonymised Information Linkage Welsh study,3 individuals were 10 times more likely to have a personality disorder, six times more likely to have alcohol dependence, and almost six times more likely to have a depressive disorder two years before an eating disorder was diagnosed than those without an eating disorder. Eating disorders, although less common than other mental health disorders, are one of the most at risk groups for premature death,14–17 comparable only with deaths from substance misuse.17 In a meta-analysis of 25 studies published between 1966 and 2010, Arcelus and colleagues reported a standardised mortality risk of 5.9 for anorexia nervosa over a mean follow-up period of 14.2 years, with bulimia nervosa and other eating disorders both having a similar risk of 1.9.14 Suicide risk specifically was greatly increased in these individuals.18 In a meta-analysis of 59 studies published in 1983-2009 with five or more years of follow-up, a standardised mortality risk for suicide of 31.0 was reported for anorexia nervosa over a mean follow-up period of 11.1 years and a standardised mortality risk of 7.5 in bulimia nervosa over a mean follow-up period of 7.5 years.18 More recent data, however, are available from only one large population based primary care study in the UK that reported cause specific mortality.3Evaluating longer term morbidity in larger population based cohorts, particularly the development of new physical health problems, is necessary to improve our knowledge which can then help in designing services to meet the complex healthcare needs of patients with eating disorders. Adverse outcomes, including mortality, psychiatric illness and, to a lesser extent, physical health, have largely been investigated separately in respective studies. This approach results in differences in cohort inclusion criteria and classifications of outcomes, and other differences between cohorts, and makes comparing results difficult. Hence in this study, we examined a large population based cohort of primary care patients, with and without a diagnosis of an eating disorder, with linked hospital and mortality records. We compared short and longer term outcomes, including mortality, and physical and mental health outcomes.Methods Data sources Linked electronic health records were obtained from the Clinical Practice Research Datalink, including datasets from general practices using two different systems: GOLD (Vision) and Aurum (EMIS). Primary care records contain routinely collected personal information and patient-doctor interactions about symptoms, diagnoses, and prescribed drug treatments. Data for sex were taken from information in the Clinical Practice Research Datalink rather than from patient reported gender. For practices in England participating in the Clinical Practice Research Datalink linkage scheme, we obtained secondary care clinical data through linkage to Hospital Episode Statistics, including ICD-10 (international classification of disease, 10th revision) diagnostic coding of inpatient episodes from hospital data. The index of multiple deprivation (2015), an ecological socioeconomic indicator based on the patient's residential postcode, was obtained from additional linkage. We obtained mortality data, including date of death and cause, by linkage to the Office for National Statistics. The GOLD dataset included patients from 416 contributing English practices (as of August 2019) with 10 800 187 patients eligible for linkage. The Aurum dataset included patients from 890 Aurum English practices (as of August 2019), with 25 618 186 patients eligible for linkage.Study design All cohort members were registered with a general practice contributing to the Clinical Practice Research Datalink. The general practice was considered up to standard for research by the Clinical Practice Research Datalink for at least one year before cohort entry, and eligible for linkage to the Office for National Statistics, Hospital Episode Statistics, and index of multiple deprivation (at the patient level). A matched cohort study design was implemented in both the GOLD and Aurum datasets. A bridging file was used to identify general practices transferring from GOLD to Aurum, and data used only in GOLD, to avoid duplication of patients, and allowing the two datasets to be combined. The exposed cohort consisted of individuals who received a diagnosis of an eating disorder (ie, anorexia nervosa, bulimia nervosa, and other eating disorders) for the first time, aged 10-44 years, recorded between 1 January 1998 and 30 November 2018. This age range was chosen based on similar previous primary care studies where a higher prevalence of eating disorders was also seen in this age range. Our data sources necessitated the use of a broad categorisation of eating disorders, a classification approach widely used in many studies over the past two decades. Also, because of the heterogeneity of symptoms for different diagnoses, frequent transitions of people between diagnostic categories, and the high prevalence of non-specific diagnoses reported in electronic health records, we believed a transdiagnostic approach was appropriate.An eating disorder episode was identified as the first occurring date of event from primary or secondary care data through relevant Read, SNOMED (Systematised Nomenclature of Medicine), and EMIS (Egton Medical Information Systems) codes, and ICD-10 and ICD-9 codes where relevant for linkage to Hospital Episode Statistics for data on patients admitted to hospital. Online supplemental appendix 1 shows the frequency of codes from each source. The final code list was compiled by consensus of a general practitioner and child and adolescent psychiatrist (CAC-G and RE, respectively) and is available on GitHub (https://github.com/C5thyM/eatingdisorder/).SP110.1136/bmjmed-2025-001438.supp1Supplementary dataThe date of diagnosis of the eating disorder for each incident case was used as the index date. An incidence density based sampling approach with replacement was used to match patients with up to 20 comparator individuals who had no recorded history of an eating disorder by the index date. Matching was conducted based on age, sex, and general practice to control for potential confounders. For each patient, one or more comparators were randomly selected from the pool of eligible individuals at risk (those without a diagnosis of an eating disorder at the time of the index date of the patient). Selecting comparators was by replacement, where an individual could be selected in multiple matched sets for more than one patient if they remained eligible at the time of each patient's diagnosis. This sampling method ensured that the comparator group accurately reflected the population at risk during the period when incidents of eating disorders occurred, facilitating an unbiased estimation of the association between outcomes and a diagnosis of an eating disorder. Follow-up started on the index date, with all individuals required to complete at least one full day of follow-up, and ended on the date of transfer out of practice or when the practice stopped collecting data for the Clinical Practice Research Datalink, death, or the final date of the study's observation period (30 November 2018, and for the comparator group if they developed an eating disorder during follow-up), whichever date came first for each cohort member (online supplemental appendix 2).Physical health conditions and mental health behaviours likely to be a consequence of, or linked to, a diagnosis of a primary eating disorder10 and believed to be adequately recorded by primary care coding were identified by the first occurring event from a combination of Read, SNOMED, and EMIS codes for Clinical Practice Research Datalink data sources, and ICD-10 codes for linkage to Hospital Episode Statistics for patient care data after admission to hospital (online supplemental appendix 1). Death events included all cause mortality, natural and unnatural deaths, and all suicides. Fatal poisoning was also included as a separate outcome. All death events were identified by ICD-10 codes in the Office for National Statistics mortality records. Outcome code lists were compiled by consensus of the study team (CAC-G, RTW, and TWO) and published on GitHub (https://github.com/C5thyM/eatingdisorder/). Analyses were performed with Stata/SE 16.0.The first occurring coded episode of non-fatal self-harm, depression, anxiety disorders, obsessive compulsive disorder, personality disorder, fractures, osteoporosis, diabetes (type 1 and type 2, and combined with unspecified diabetes type), or severe disease or organ failure of the heart, kidney, or liver, was established as either occurring before (prevalent) or subsequent (incident) to the index date. Only those incident events occurring after the index date were included in subsequent outcome analyses (online supplemental appendix 3).Analysis Patient characteristics and prevalent conditions at the index date are summarised as frequencies and percentages for individuals with and without an eating disorder. Annual cumulative incidence with 95% confidence intervals (CIs) and excess risk for each adverse event outcome, including adjustment for competing risk (as detailed by Gooley et al 19) of other causes of death where relevant, were estimated with the number of new incidences of the event as the numerator and number of individuals at risk during that period and expressed as a percentage. Cumulative incidence and excess risk of coded adverse events in those with an eating disorder compared with those without an eating disorder were reported for one, five, and 10 year periods.Hazard ratios indicate how likely an individual with an eating disorder is to be coded with an adverse outcome compared with those without an eating disorder. Hazard ratios with 95% CIs were calculated with stratified Cox regression models accounting for matched sets and adjusted for the neighbourhood level index of multiple deprivation. Fatal adverse outcomes examined were all cause mortality, natural and unnatural deaths, and deaths from suicide and poisonings. Non-fatal outcomes were osteoporosis, bone fracture, diabetes, severe disease or organ failure of the kidney, liver, or heart, self-harm, anxiety disorders, depression, personality disorders, and obsessive compulsive disorders. Individuals with prevalent cases of adverse events were excluded from the analyses when calculating separate outcome cumulative incidences and hazard ratios. The proportional hazards assumption was assessed with Schöenfeld-Martingale residual testing for each Cox model.20 For many of the outcomes examined, the proportional hazard assumption was violated over the whole study period, and therefore hazard ratios were calculated specific to the first year, 1-5 years, and after five years of follow-up.Patient and public involvement Although patients and the public were not directly involved in the design or implementation of the study, we plan to approach large scale national eating disorder charities, such as Beat and FEAST (families empowered and supporting treatment of eating disorders), and the Royal College of Psychiatrists, to disseminate the results of the study. The protocol for the study ( online supplemental appendix 4) received approval from the Clinical Practice Research Datalink Independent Scientific Advisory Committee (ISAC protocol 19_208).Results Patient characteristics The study cohort comprised 24 709 individuals who had a new diagnosis of an eating disorder coded in their electronic health record: 14.5% (n=3577) were coded as anorexia nervosa, 20.6% (n=5085) as bulimia, 4.9% (n=1215) as binge eating disorder, and 60% (n=14 832) had an unspecified eating disorder code or other eating disorder not classified as above. The comparator group comprised 493 001 individuals. In each group, 89% of individuals were women and two thirds were aged 10-24 years. A greater proportion of individuals with eating disorders lived in the least deprived versus the most deprived areas (21.9% v 17.8%; P=0.001) (table 1). Median follow-up was 4.03 (interquartile range 1.56-8.78) years for the full cohort, with more than 10 years of follow-up for 19.9% (n=4918) of individuals with an eating disorder and 20.1% (n=101 202) of individuals without an eating disorder.Table 1Sociodemographic and baseline clinical characteristics of the cohort with a diagnosis of an eating disorder and the unaffected matched comparator cohortCharacteristicsEating disorder cohortUnaffected comparator cohortTotal No of individuals24 709 (4.8)493 001 (95.2)Sex: Women22 044 (89.2)439 779 (89.2) Men2665 (10.8)53 222 (10.8)Age at index (years): 10-2415 929 (64.5)318 012 (64.5) 25-448780 (35.5)174 989 (35.5)Index of multiple deprivation (group): 1 (least deprived)5368 (21.7)108 224 (22.0) 25061 (20.5)102 549 (20.8) 34794 (19.4)96 722 (19.6) 45073 (20.5)97 791 (19.8) 5 (most deprived)4413 (17.9)87 715 (17.8)Pre-existing mental health problem: Non-fatal self-harm3113 (12.7)17 830 (3.7) Depression7336 (29.7)50 763 (10.3) Anxiety disorder4098 (16.6)27 765 (5.6) Personality disorder542 (2.2)1360 (0.3) Obsessive compulsive disorder484 (2.0)1789 (0.4)Pre-existing physical health problem: Diabetes326 (1.3)3267 (0.7) Type 1191 (0.8)2034 (0.4) Type 2161 (0.7)1281 (0.3) Heart failure25 (0.1)180 (0.04) Renal failure119 (0.5)1442 (0.3) Liver disease32 (0.1)333 (0.1) Osteoporosis*80 (0.3)335 (0.1) Bone fracture4821 (19.5)80 157 (16.3)Data are number (%).*Excludes pathological fracture and vertebral collapse.Comorbidity Compared with their matched comparators, those with an eating disorder had a higher prevalence of all mental and physical health problems investigated before the index date: 30% versus 10% for depression, 17% versus 6% for anxiety, 0.1% versus 0.04% for heart failure, and 0.3% versus 0.1% for osteoporosis, respectively. Diabetes, predominantly type 1, was also more prevalent among those with an eating disorder (1.3% v 0.7%) (table 1). Comorbidities in men and women were similar, except for a higher prevalence of depression (31% vs 21%) and self-harm (13% vs 9%) in women, in individuals with an eating disorder (online supplemental appendix 5).Adverse outcomes Physical health We found increased risks for all coded episodes of adverse physical health outcomes ( figure 1 and online supplemental appendices 6 and 7). The short term risk (ie, in the first year after a diagnosis of an eating disorder) was higher than in other time periods for all physical health outcomes coded, except for fractures. For renal failure and liver disease, individuals with an eating disorder episode were six times more likely to have a coded episode of the outcome within the first year (hazard ratio for renal failure 6.0 (95% CI 4.2 to 8.5), excess events at one year 15 per 10 000 individuals (95% CI 11 to 21); hazard ratio for liver disease 6.7 (3.8 to 11.7), excess events at one year 6 per 10 000 (4 to 11)). For all physical health outcomes, except fractures, coded episodes of the outcome decreased over time but remained significantly high even five years after a diagnosis of an eating disorder (P<0.05).More than five years after an eating disorder was first diagnosed, individuals were six times more likely to have a coded episode of osteoporosis (hazard ratio 6.1, 95% CI 4.9 to 7.5; excess events at five years 117, 95% CI 102 to 134 per 10 000 individuals), two times more likely to have a coded episode of heart failure (hazard ratio 1.8, 1.1 to 2.8; excess events at five years 54 per 100 000, 22 to 108) or renal disease (hazard ratio 2.2, 1.8 to 2.6; excess events at five years 45 per 10 000, 34 to 57), and more than 2.5 times more likely to have a liver disease code recorded (hazard ratio 2.6, 1.8 to 3.8; excess events at five years 18 per 10 000, 12 to 25) (figure 1 and online supplemental appendices 6 and 7). For cumulative incidence and excess events observed at 10 years (table 2), those with an eating disorder had an excess of 199 osteoporosis events (95% CI 174 to 277) and 110 renal failure events (87 to 136) per 10 000 individuals compared with those without an eating disorder. For fractures, those with an eating disorder had an increased risk five years after diagnosis (figure 1), and an excess of 147 event fractures (13 to 309) per 10 000 individuals by 10 years after diagnosis of an eating disorder (table 2).Figure 1Hazard ratios with 95% confidence intervals (CI) for poor physical health outcomes, grouped by follow-up periodTable 2Cumulative incidence (absolute risk) and excess risk of coded adverse outcomes at 10 years in the cohort with a diagnosis of an eating disorder and in the unaffected matched comparator cohortCumulative incidence (95% CI) at 10 years (%)Excess risk (events or deaths per 10 000) (95% CI)Eating disorder cohortComparator cohortMortality: All deaths1.45 (1.22 to 1.72)0.50 (0.47 to 0.54)95 (75 to 118) Natural deaths1.02 (0.83 to 1.24)0.41 (0.38 to 0.44)61 (45 to 80) Unnatural deaths0.43 (0.32 to 0.59)0.093 (0.080 to 0.11)341 (236 to 479)‡ Suicides0.21 (0.14 to 0.32)0.041 (0.033 to 0.050)169 (103 to 266)‡ Fatal poisonings0.22 (0.13 to 0.35)0.031 (0.023 to 0.040)187 (108 to 308)‡Psychiatric health or behavioural event:* Non-fatal self-harm11.67 (10.88 to 12.48)2.99 (2.89 to 3.10)868 (799 to 938) Depression36.56 (34.87 to 38.24)16.60 (16.29 to 16.91)1996 (1858 to 2133) Anxiety disorder24.33 (23.06 to 25.63)10.59 (10.38 to 10.80)1374 (1268 to 1483) Personality disorder4.82 (4.41 to 5.25)0.53 (0.50 to 0.56)429 (391 to 469) Obsessive compulsive disorder2.24 (1.98 to 2.52)0.50 (0.46 to 0.53)174 (152 to 199)Physical health event:* Heart failure0.21 (0.13 to 0.33)0.11 (0.09 to 0.13)10 (3 to 20) Osteoporosis†2.19 (1.92 to 2.49)0.20 (0.18 to 0.22)199 (174 to 227) Bone fractures7.29 (5.57 to 9.30)5.82 (5.44 to 6.21)147 (13 to 309) Renal failure1.69 (1.43 to 1.99)0.60 (0.56 to 0.63)110 (87 to 136) Liver disease0.41 (0.30 to 0.56)0.15 (0.13 to 0.17)26 (17 to 39) Diabetes1.77 (1.49 to 2.07)1.03 (0.98 to 1.08)74 (51 to 99) Type 1 diabetes0.37 (0.25 to 0.52)0.20 (0.18 to 0.23)17 (7 to 29) Type 2 diabetes1.59 (1.34 to 1.88)0.90 (0.86 to 0.95)69 (48 to 93)*In instances where a health event of interest occurred before the index date, cohort members with an eating disorder, along with their matched comparator patients, were excluded from the analysis. Online supplemental appendix 3 shows the number of included patients for each event.†Excludes pathological fracture and vertebral collapse.‡Per 100 000 individuals.CI, confidence interval.Mental health We found an increase in the number of coded episodes of comorbid mental health problems in subsequent periods of follow-up after the initial diagnosis of an eating disorder. The risk was highest in the short term and then decreased ( figure 2 and online supplemental appendices 6 and 7). In the first year, individuals with an eating disorder were nearly 16 times (hazard ratio 15.9, 95% CI 12.8 to 19.9; excess events at one year 63 per 10 000 individuals, 95% CI 54 to 74), more than 12.5 times (hazard ratio 12.6, 10.1 to 15.6; excess events at one year 59 per 10 000, 50 to 71), and 9.4 times (hazard ratio 9.4, 8.2 to 10.7; excess events at one year 309 per 10 000, 279 to 342) more likely than their comparators to have personality disorders, obsessive compulsive disorder, and non-fatal self-harm, respectively. The risk of non-fatal self-harm remained at nearly three times more likely after five years (hazard ratio 2.7, 2.3 to 3.3; excess events at five years 713 per 10 000, 660 to 769), and two times more likely for depression (hazard ratio 2.0, 1.8 to 2.2; excess events at five years 1533 per 10 000, 1436 to 1631) and anxiety (hazard ratio 2.1, 1.9 to 2.4; excess events at five years 900 per 10 000, 830 to 972) at more than five years after a diagnosis of an eating disorder (figure 2 and online supplemental appendices 6 and 7). After 10 years of follow-up, we found an excess risk of coded events of 868 (95% CI 799 to 938) for non-fatal self-harm, 1996 (1858 to 2133) for depression, and 1374 (1268 to 1483) for anxiety per 10 000 individuals, comparing those with and without a diagnosis of an eating disorder (table 2).Figure 2Hazard ratios with 95% confidence intervals (CI) for mental health outcomes, grouped by follow-up periodCause specific mortality We recorded 200 deaths (0.81%) in individuals with an eating disorder and 1507 in those without an eating disorder (0.31%, hazard ratio for all cause mortality 2.2, 95% CI 1.8 to 2.7). Median age at death during the study's observation period was 38 (interquartile range 26-44) years versus 40 (29-47) years in those with and without an eating disorder, respectively. A patient with a diagnosis of an eating disorder was five times more likely to die from an unnatural cause within the first 12 months of the diagnosis and more than 5.4 times more likely 1-5 years after a diagnosis ( figure 3). Although the numbers of patients were small in the first 12 month period, we saw an increased risk of dying by suicide; 14 times more likely in the first 12 months (hazard ratio 13.7, 95% CI 4.8 to 38.8) and five times more likely 1-5 years after a diagnosis (hazard ratio 5.2, 2.8 to 9.6). The risk of fatal poisoning was 14 times more likely in the first 12 months (hazard ratio 13.8, 3.2 to 58.5) and six times more likely 1-5 years (hazard ratio 6.1, 2.8 to 13.6) after the first diagnosis of an eating disorder (figure 3 and online supplemental appendices 6 and 7).Longer term risk remained high for all causes of death; the risk of an unnatural death was three times more likely in those with an eating disorder than in those without an eating disorder, more than five years after diagnosis (hazard ratio 3.2, 1.9 to 5.4) (figure 3). At 10 years, excess unnatural deaths were 341 per 100 000 individuals (95% CI 236 to 479) in those with a diagnosis of an eating disorder compared with those without an eating disorder. For the same periods, the risk of death by suicide more than five years after diagnosis of an eating disorder was almost three times more likely than those without a diagnosis (hazard ratio 2.7, 1.3 to 5.8), with excess death by suicide of 169 per 100 000 individuals at 10 years (95% CI 103 to 266) (table 2), and 24 (10 to 52) and 105 (63 to 168) excess deaths per 100 000 individuals at one and five years, respectively (online supplemental appendix 7).Figure 3Hazard ratios with 95% confidence intervals (CI) for all cause and cause specific mortality, grouped by follow-up periodThe risk of all cause mortality in men was higher than that in women 1-5 years after a diagnosis of an eating disorder (hazard ratio 5.3, 95% CI 3.3 to 8.6 v 2.8, 2.1 to 3.7). We found a similar risk in both men and women in the first 12 months (hazard ratio 4.6, 1.9 to 11.3 v 4.6, 2.9 to 7.3) and more than five years (hazard ratio 2.6, 1.5 to 4.5 v 2.1, 1.7 to 2.7) after a diagnosis. Excess risk for all cause mortality was estimated at 140 per 10 000 individuals (95% CI 80 to 190) for men and 30 per 10 000 (20 to 40) for women at five years. Although the numbers of men were small, men were more at risk from natural causes of death, whereas women were more at risk from unnatural causes in the first year after a diagnosis of an eating disorder. Women were also at greater risk of fatal poisonings (online supplemental appendix 8).Discussion In this large population based study of eating disorders, we used health records with interlinked primary and secondary care sources. We investigated mental and physical, short and long term outcome risks, and cause specific mortality risks in the same cohort of individuals. As well as an increase in the risk of adverse mental health outcomes and death, we found an increased number of coded episodes of adverse physical health outcomes, such as diabetes, renal failure, liver disease, osteoporosis, and fractures, with increased risks after more than five years from the first diagnosis of an eating disorder. Our study provides more evidence about the longer term risk of outcomes in those who had a diagnosis of an eating disorder, and highlights the need for timely, integrated mental and physical health services and continued monitoring, even after apparent improvement in the eating disorder.Mental health outcomes We saw high levels of pre-existing and coexisting mental health and behaviour problems, including depression, anxiety, and personality disorder, consistent with previous published data, although based on a much smaller Welsh cohort. 3 In a larger Swedish clinical database of individuals attending an eating disorder specialist clinic, 71% of patients with an eating disorder had at least one other psychological disorder, including comorbid mood, anxiety, personality and obsessive compulsive disorders, as well as substance misuse.21 In a previous hospital based study, the risk of self-harm was reported as a standardised rate ratio of 13.3 (95% CI 12.4 to 14.3) in those admitted to hospital with an eating disorder.22 Although higher than our estimated risk, this cohort potentially had a more severe condition. Evidence of a comorbid relation between eating disorder behaviours and non-fatal self-harm should also be considered, with questions about their directional nature. Emerging evidence suggests that for some individuals, this relation may be bidirectional.23 We have shown that individuals with an eating disorder have a sustained increased risk of non-fatal self-harm up to 10 years after diagnosis, with no previous history of self-harm recorded in their electronic health record. A more recent study24 reported an absolute risk of 38% in those with anorexia or bulimia, higher than our reported absolute risk of 11% over 10 years. Their study was based on patients admitted to hospital, however, and included deaths from self-harm and not non-fatal self-harm, as in our study. Their study also reported a reduction in risk during the first year compared with subsequent years, although the risk was still increased, as seen in our study.24Physical health outcomes Because of the nature of anorexia, and bulimia in particular, regulation of homeostasis is disturbed, adversely affecting the cardiovascular and renal systems, bone physiology, hydro-electrolyte alteration, and acid-base balance. 10 Weight loss and malnutrition may directly result in health complications. In a study comparing morbidity in patients with atypical anorexia nervosa (where weight may be recovered to the normal range) with those with full threshold anorexia nervosa, outcomes were similar in both groups and indicated that despite adolescents presenting within or above the normal weight range, loss of substantial weight or rapid weight loss may itself be detrimental to physical health.25 We found an increased risk of osteoporosis, which is well reported, with the potential consequence of bone fracture. The cause of this increase in risk may be multifactorial, including reduced weight loading ability, reduced bone specific nutrients, including calcium,26 and hormonal imbalance caused by amenorrhoea. In younger people, failure to infill the skeletal envelope during rapid periods of bone growth and trauma may occur, with delayed puberty potentially exacerbating this effect. An increased risk of fracture from 12 months to more than five years after the first recorded eating disorder was similarly reported in a Danish nationwide study based on much earlier data (1977-98), with an increased risk of fracture persisting more than 10 years after a diagnosis of an eating disorder.27 We also found a high risk for the development of osteoporosis within the first 12 months of a recorded eating disorder. This finding might indicate a delay in patient disclosure, a delay in the recording of an eating disorder in the primary care record, or could be related to increased surveillance, with those with eating disorders more likely to be referred for scanning given the known risks of osteoporosis.We found a higher prevalence of type 1 diabetes before a recorded diagnosis of an eating disorder, potentially supporting reports in the literature.28 The association between type 1 diabetes and eating disorders was also highlighted in a recent UK parliamentary inquiry,29 noting that an eating disorder could develop after a diagnosis of diabetes because of an increased focus on food, weight monitoring, weight gain caused by drug treatments, and body dysphoria, particularly in younger individuals. In a meta-analysis of controlled studies investigating the prevalence of eating disorders in type 1 diabetes in women, no significant differences were found between patients with diabetes and controls in individuals with anorexia nervosa, but the prevalence in those with bulimia nervosa (1.73% v 0.69%) and with a combination of the two conditions (2.0% v 0.75%) was significantly higher in patients with than in those without type 1 diabetes.28 Although absolute risks were lower than for other physical conditions, we found an increased risk of type 2 diabetes after a diagnosis of an eating disorder, indicating that gastric and gastrointestinal problems, including metabolism and insulin response, may further increase the risk of diabetes.30We found a high cumulative incidence and excess events of renal failure at 10 years. In a recent retrospective case-control study of individuals aged 9-18 years admitted to hospital with anorexia nervosa, 37% were estimated to have impaired kidney function.31 Although the mechanism involved in renal failure in individuals with anorexia nervosa is complex and incompletely understood, low protein intake in restrictive diets, such as those seen in eating disorders, decreases glomerular filtration rate and renal plasma flow. This mechanism, in turn, reduces the ability of the kidneys to concentrate urine and excrete sodium and acid.31 In addition to purging, use of diuretics and laxatives can also contribute to chronic hypokalaemia and hypovolaemia, which left untreated may potentially cause renal failure. We found that the risk of renal failure remained increased for more than five years after a diagnosis of an eating disorder. This finding may indicate the potential damage sustained by the kidneys and the long term effects. Therefore, patients with eating disorders should be screened for evidence of renal impairment and referred to a renal specialist if concerns exist.Mortality We found that individuals with an eating disorder had a sustained risk of death, in the short and long term. Individuals had a 4.6 times higher risk of death from all causes during the first year after diagnosis, and 3.2 times at 1-5 years and 2.2 times at after five years. These values are higher than the crude hazard ratios reported in the Welsh primary and secondary care dataset (Secure Anonymised Information Linkage) of 1.9 at the end of the three year follow-up period. 3 The Welsh study, however, included 77 777 individuals compared with our study of 517 710 individuals.Self-poisoning is the most common form of self-harm in younger women with an eating disorder.24 We found similar risks for fatal poisonings and suicides, with much overlap in the cause of death. A previous study based on data from Hospital Episode Statistics, England, and mortality data for 1999-2011 on hospital attendees, reported that individuals with an eating disorder were about 8-9 times more at risk of dying by suicide (relative risk 8.4, 95% CI 5.0 to 13.4).22 Compared with our findings in the first year after diagnosis, we found that individuals were more than 13 times more likely and about six times more likely in subsequent years to die by suicide. The first 12 months after a diagnosis of an eating disorder or the start of treatment is a critical period when the risk of suicide is highest. During this time, individuals may have heightened emotional vulnerability, ambivalence about recovery, and more distressing emotions as they confront disordered eating behaviours. These factors, combined with underlying impulsivity (particularly relevant in bingeing and purging behaviours), cognitive rigidity, difficulty in adapting to new information or changing circumstances, and emotional dysregulation can increase the risk of suicidal thoughts and behaviours.32 The risk of mortality from suicide is greatly increased in those with an eating disorder combined with (6.8 times likely) or without (four times likely) a history of self-harm.33Although we need to interpret the results grouped by sex with caution because of the small numerator values and wide confidence intervals, especially for men, we found that men had a higher risk of all cause mortality 1-5 years after a recorded code for an eating disorder. The higher risk in men supports similar observations from a recent primary care study of electronic health records in Denmark34 and highlights the importance of recognising the difficulties in men as well as women. The National Confidential Inquiry into Suicide and Safety in Mental Health found that individuals with an eating disorder who died by suicide were twice as likely to be women and have died after an overdose than those with other psychiatric conditions.35 This finding further supports our observation of women being potentially at higher risk of fatal poisonings during the period of study.Strengths and limitations of this study Our study on eating disorders was based on a large number of patients with primary care data, linked to secondary care data and mortality records. We investigated both mental and physical health adverse outcomes as well as mortality in the same cohort of individuals. Also, the large primary care dataset of electronic health records allowed us to investigate the short term (<1 year), medium term (1-5 years), and longer term (10 years) adverse outcomes.Our study had some limitations. We relied on accurate and appropriate reporting of codes in the electronic health record by general practitioners and other primary care clinicians. We found that 60% of individuals had a non-specific eating disorder code rather than a more specific code for anorexia nervosa, bulimia, or binge eating disorder, and hence we could not accurately conduct subtype analyses. A transdiagnostic approach of using a broader categorisation of eating disorders, however, allows comparisons with previous studies that used a combined group. Also, from a policy perspective and societal focus, eating disorders are considered as a collective rather than specific eating disorder conditions.Potential surveillance bias should be considered because some conditions may be more likely to be coded if an individual has an existing eating disorder. The code description gives no indication of the severity of the eating disorder and so we could not investigate whether severe conditions are linked to worse outcomes. Some individuals who receive a diagnosis of an eating disorder may not, therefore, be at greater risk than others, with some results not applying equally to all individuals with eating disorders, particularly those receiving earlier interventions. The broad classification of eating disorders may mask considerable heterogeneity in absolute and relative risk that we could not elucidate because of the limitations of the primary care clinical coding. We also could not determine to what extent the excess risks for mental health and physical comorbidities between the eating disorder and comparator groups were a result of additional comorbidities rather than the eating disorder. For the analyses grouped by sex, the findings must be interpreted with caution because of the wide confidence intervals and possible under-reporting of eating disorders in men.Finally, all incidences of eating disorders might not have been identified because eating disorder behaviour may be hidden, with potential delays in presentation at primary care but also at the community level.36 A delay in diagnosis may be a result of the individual consulting their doctor with other symptoms as a consequence of their eating disorder, without disclosing their disordered eating behaviour.Implications for future research, policy, and practice This study provides valuable data for highlighting and recognising the range of physical and mental health risks, as well as mortality outcomes, associated with eating disorders. Our data describe the substantial long term effects of eating disorders and emphasise the potential opportunity for primary care to have a greater role in offering support and long term monitoring for individuals who are recovering from an eating disorder. A closer and more cohesive management approach in primary and specialist care may also be needed, for both physical (nephrology, cardiology, and endocrinology) and mental health services to provide this support. A potential gap exists in provision where patients' difficulties are too complex for low intensity brief interventions but not complex enough for specialist teams. Our study also highlights the need for clinicians and researchers to consider distinguishing and coding of subtypes eating disorders, because this level of diagnostic detail could make subsequent research findings more relevant and useful for understanding the different clinical subtypes of eating disorders.Initiatives such as the MARSIPAN (management of really sick patients with anorexia nervosa) guidance, published in 2014, and MEED (medical emergencies in eating disorders), updated in October 2023, have looked at the increasing number of hospital admissions for eating disorders and expanded the focus to all eating disorders.37 MEED outlines 12 recommendations across all levels of healthcare, including the role of primary care in monitoring and assessing physical investigations, and promoting collaboration between services and specialities when complex care is needed. Together with established NICE guidelines,11 these updates highlight the need for a targeted, multidisciplinary approach with a clear treatment pathway for managing people with eating disorders.NICE has published more specific guidance in primary care, including managing individuals with both suspected eating disorders in primary care and confirmed eating disorders.38 For example, guidance for individuals with a confirmed eating disorder includes ensuring clear agreement between primary and secondary or tertiary care, and monitoring the level of risk to mental and physical health. Also, for those with confirmed anorexia nervosa not receiving secondary care, physical and mental health checks should be done at least annually, including discussions about treatment, as well as measurements of weight or body mass index, blood pressure, electrocardiography, and assessment of the risk of physical and mental health with available risk tools. More practical questions and statements for assessing and treating eating disorders have been proposed in primary care in the US, such as recommended history and physical examination, clinical investigations, and an algorithm for medical management of individuals with suspected eating disorders or for those who have received a diagnosis of an eating disorder.39Primary care can potentially have an important role in providing long term support and monitoring for individuals, even after recovery from an eating disorder. Raising awareness among healthcare providers about the lasting effects of eating disorders and the need for ongoing support in managing current symptoms and recovery40 is essential. This approach also highlights the importance of appropriate commissioning and funding for such services.