Our best estimate, based on an analysis of the most recent UK survey data, is that around 3.5 million people in the UK have an eating disorder. This is based on analysis following advice from leading eating disorder epidemiologists and statisticians. However, like many prevalence estimates this is based on imperfect data, and the real figure may be lower or higher.
Nevertheless, our analysis shows that there is roughly a four-in-five (78%) chance that the true prevalence is higher than our previous estimate of 1.25 million, and it may be much higher.
Our new estimate is not directly comparable to our previous one, due to an unavoidable difference in methodology, and change in diagnostic criteria over time. More information on the methodology followed is set out below.
While the evidence indicates that eating disorders are more common than previously thought, the uncertainty in this estimate highlights the importance of further research in this area.
How this estimate was produced
The calculations drew from two different national surveys with different methods. This approach was considered reasonable in the absence of a single all-ages study. These surveys were:
These were both conducted in England. We have applied the rates derived from these to the UK population on the reasonable assumption that prevalence is likely to be broadly similar across England, Scotland, Wales, and Northern Ireland.
For this age group, the MHCYP survey was used. This estimated eating disorder prevalence amongst 11–16-year-olds at 2.6% (with a 95% confidence interval of 1.5-3.7%) or around 1 in 40. This would equate to around 100,000 children of that age group (the uncertainty range would equal around: 64,000-148,000). We applied the 2.6% rate to 11–15-year-olds to avoid double-counting 16-year-olds.
The MHCYP survey assigned diagnoses of eating disorder according to ICD-10 criteria. It did also attempt to assess the prevalence of ARFID; however, individuals identified as having ARFID were included within and reported as part of the ‘Other eating disorder’ category.
The approach taken for those aged 16 and over was much more complex. The 2023/24 APMS found that 9.1% of adults aged 16 and over screened positive for a ‘possible eating disorder’ on the 5-question SCOFF screening tool. Screening positive was defined as ‘scoring’ two or more. As a screening tool it sometimes misses people who do have an eating disorder (known as ‘false negatives’), and in other cases it wrongly flags some people who do not have an eating disorder (‘false positives’).
The analysis sought to correct the 9.1% figure for those two kinds of error by accounting for what is known about the accuracy of the SCOFF. The accuracy figures used were taken from the most recent study to validate the SCOFF in a community sample in the UK (Solmi et. al., 2015). In a multi-ethnic London community sample, it found that it:
The sample used in the validation study (a multi-ethnic London community sample) is not identical to the national survey samples and a screening tool’s sensitivity and specificity will vary depending on the populations it is used in.
The uncertainty around these estimates was significant, likely due to the relatively small sample, with only 145 people being interviewed.
It should also be noted that the SCOFF screening tool was designed specifically to screen for possible cases of Anorexia and Bulimia. However, the validation study cited (Solmi, et. al., 2015) did find some people experiencing Binge Eating Disorder and Other Specified Feeding and Eating Disorder (OSFED) that had screened positive, so these will have contributed to the sensitivity and specificity estimates cited above. It is possible that some of the people that screened positive on the SCOFF in phase 1 of the APMS survey could have had ARFID. However, the validation study did not feature anyone with Avoidant/Restrictive Food Intake Disorder (ARFID), and therefore it’s sensitivity and specificity estimates do not account for ARFID. We were unable to find any research that we could draw from to directly estimate the prevalence of ARFID in the UK.
A small overlap exists in this method for 16-year-olds. The 2.6% figure we applied to 11–15-year-olds was derived from the 11–16-year-old age-band reported by the MH CYP survey, while the APMS survey was completed by those aged 16 and over. Without the raw survey data, it is not possible to account for the inclusion of 16-year-olds in both surveys. However, the effect of this on an 11 years old-and-over estimate is not likely to be significant.
The analysis applied a standard, well-established formula that can be used to turn a screen-positive rate into a corrected prevalence estimate, given the screening tool’s sensitivity and specificity. This is called the Rogan-Gladen estimator. Once this had accounted for the contribution of false positives, it produced a corrected estimate that was lower than the raw 9.1% figure.
The figures on this page come from a Monte Carlo simulation. We also estimated the range using two other statistical methods (the Lang-Reiczigel adjusted interval and a Bayesian estimate) to act as checks, and all three produced similar results:
The overall estimate
To estimate uncertainty for the overall prevalence estimate, the totals for 11–15-year-olds and 16+ year olds were combined in terms of numbers of people, carrying every source of uncertainty through the same simulation, and then divided by the combined 11-and-over population to produce an overall rate (and uncertainty range).
Some probability statements can provide more clarity on the uncertainty. The analysis implies a roughly:
In short, the analysis suggests that the number of people with an eating disorder is very likely to be larger than our previous 1.25 million estimate.
Acknowledgements
We would like to acknowledge the valuable contributions made by others to this work. We are particularly grateful to a group of epidemiological researchers specialising in the study of eating disorders who kindly provided pro bono advice on the methodology. The group included Dr Clara Faria, Dr Maddie Davies Kellock, Professor Nadia Micali, and Professor Francesca Solmi. We also wish to thank Dr Seth Thomas for conducting the statistical analysis and advising on the communication of the findings.