A Meta-Analysis of Mortality in Eating Disorders

You often hear that anorexia nervosa has the highest mortality rate of any psychiatric illness, but you might struggle to find the rates for bulimia nervosa or EDNOS. Even for AN, the most common cause of death is rarely reported and the reported rates often vary a lot (depending on the study (and the media outlet).

I wanted to find out what are: (1) the mortality rates in BN and EDNOS and (2) the common causes of death in these disorders.  A relatively recent meta-analysis (click here for some background, pros and cons of meta-analyses) of 36 studies, which addressed some of my questions, was published by Arcelus and colleagues.

They excluded studies that had less than 15 patients and/or <1-year follow-up.


Mortality in Anorexia Nervosa:

  • 35 studies, mean follow-up time: 12.8 years (maximum 36.2), mean sample size was 361 (maximum 6009).
  • Total of 12 808 individuals, 639 deaths
  • Weighted annual mortality: 5.10 death per 1000 person-years (5.4 in female only studies and 4.55 in inpatient population)
  • 1 out of 5 deaths from suicide
  • SMR: 5.86 (based on  25 studies, 14.2 mean follow-up period, 12 189 patients

Mortality in Bulimia Nervosa:

  • 12 studies, mean follow-up time was 9.34 years, mean sample size was 200
  • Total of 2585 individuals, 57 death
  • Weighted annual mortality rate: 1.74 per 1000 person-years (2.22 in females only)
  • SMR: 1.93 (all 12 studies)

Mortality in EDNOS:

  • 6 studies, mean sample size was 313 patients, mean follow-up time was 9.1 years
  • Total of 18799 individuals, 59 deaths
  • Weighted annual mortality rate: 3.31 per 1000 person-years
  • SMR: 1.92 (4 studies)

SMR = standard mortality ratio: is a ratio between the observed number of deaths in the study group and the number of deaths that would be expected (matched by age and sex-specific rates). If the ratio of observed:expected deaths is greater than 1.0, there are more deaths in the study population than would be expected in the general population. Wikipedia gives a nice example:

“An SMR for bladder cancer of 1.70 in the exposed group would mean that there is 70% more cases of death due to bladder cancer in the cohort than in the reference population (in this case the national population, which is generally considered not to exhibit cumulative exposure to high arsenic levels).”

The discussion in this paper is insightful, and I’ll briefly summarize it here:

Limitations of previous research (which also limits the meta-analysis)

  • variable length of follow-up
  • low ascertainment rates
  • small numbers of patients with BN, EDNOS
  • variable classification used

What about diagnostic crossover?

My previous entries have touched on the issue of diagnostic crossover (Eddy et al., 2008, Eddy et al., 2010) where I mentioned that 20-50% of patients with AN develop BN over time (and often relapse back to AN), an around 14% from BN to AN (7 year follow-up). Why does this matter here? Well, patients were diagnosed at the start of the study, but, after, 5, 10, 20 years, maybe they died due to a different diagnosis. As a result, what’s reported as a mortality rate for AN, BN or EDNOS may not be accurate.

Cause of death?

Unfortunately, it is not always available. This means that not only is it highly likely that people diagnosed with one disorder died from another, but also that many could’ve have died of something unrelated to the disorder at all (car crash, for example). There were 12 studies that looked at the rate of suicide in AN patients and found that the weighted annual mortality from suicide was 1.39 (meaning 1 in 5 people initially diagnosed with AN died of suicide.) However, this is what the SMR should in theory control for, since a car crash for example is independent of an eating disorder, and should be factored into the denominator of the SMR ratio, thus, again, in theory, the SMR is the relative risk of dying due to the eating disorder.

Standardized mortality rates for AN (5.86) were much higher than for EDNOS and BN (1.92 and 1.93, respectively). As the authors write:

The low SMR suggests that a diagnosis of BN per se does not render an individual at increased risk of premature death, but this finding does not necessarily justify complacency, given that comorbid affective disorder and related behaviors may often accompany bulimic symptoms.

And, recall that the study by Eddy et al found that many patients with AN (75-85%) fit into the EDNOS category in a 7 year follow-up (counted as part of the partial recovery group). Agras et al (2009) found that only 18% of EDNOS patients never had or developed a full-syndrome ED during the 4 year study. 

Although Arcelus et al wanted to look at the predictors of death, the limited number of studies and the fact that many did not report the cause of death, made this aim difficult to accomplish in this meta-analysis.

In my next post, I’m going to cover a few studies individually, to see how the mortality rate varies based on sample size, age and diagnosis, as well as look into the most common causes of death in the studies that report it. In particular, I’m going to look at the study by Crow et al. (2009): Increased mortality in bulimia nervosa and other eating disorders and Rosling et al. (2010): Mortality in eating disorders: a follow-up study of treatment in specialist unit 1974-2000.


Arcelus, J., Mitchell, A., Wales, J., & Nielsen, S. (2011). Mortality Rates in Patients With Anorexia Nervosa and Other Eating Disorders: A Meta-analysis of 36 Studies Archives of General Psychiatry, 68 (7), 724-731 DOI: 10.1001/archgenpsychiatry.2011.74


Tetyana is the creator and manager of the blog.