Eating disorders · 4 items (not reproduced) · 0-4 · Developed by Cotton, Ball and Robinson at University College London (2003).
Licensed instrument
ESP: scoring, cutoffs & interpretation
A four-question screen for eating disorders in primary care, scored 0 to 4, designed to rule out rather than rule in. The questions are not reproduced here for copyright reasons.
The ESP is a four-question, yes-or-no screen for eating disorders in primary care. Each abnormal answer scores one point, giving a total of 0 to 4. The questions and item-level scoring key are not reproduced here; consult the original publication for authorised source material.
What the ESP measures is best described as the likelihood that a fuller assessment is warranted, not the severity of a problem. It carries no subscales, no dimensional score and no diagnostic categories. A high total does not indicate a more serious eating disorder; it indicates a higher post-test probability that one is present. The instrument was designed for a specific job in a specific setting - deciding, in a few seconds of a general practice consultation, whether to ask more.
02 - Origin & purpose
Where it comes from.
The ESP was developed by Mary Ann Cotton, Carol Ball and Paul Robinson and published in the Journal of General Internal Medicine in 2003, under the title "Four simple questions can help screen for eating disorders". Its purpose was to improve on the SCOFF, published four years earlier, which the authors suspected was less sensitive in practice than its derivation study had suggested.
The instrument's name and its item count disagree, and this causes persistent confusion. Five questions were drafted and administered. One of them, asking about family history of eating disorders, made no difference to screening performance and was dropped before the main analysis. The validated instrument therefore has four items, which is what the title refers to, even though many sources - including the United States Preventive Services Task Force - describe the ESP as a five-item measure while quoting performance figures that come from the four-item version.
03 - Scoring & cutoffs
How scoring works.
Each of the four questions is answered yes or no, and each abnormal answer scores one point, for a total of 0 to 4. There is no weighting. The item-level scoring key is available in the original publication 1.
Two or more abnormal answers is the standard threshold for further assessment. The original paper also published a three-band interpretation using likelihood ratios, shown below, which is more informative than the single cutoff because it distinguishes a borderline result from a clearly positive one. Note that the post-test probabilities in that analysis were calculated at the study's eating-disorder prevalence of 12%, which is high. In a general population, or among people browsing a website, the equivalent probabilities will be substantially lower.
The four questions are not reproduced on this page for copyright reasons, and there is no calculator here. Read the free full text at PubMed Central using reference 1.
Score
Severity
Interpretation
0–1
Screen negative
An eating disorder is very unlikely. The likelihood ratio for this band was 0.0 (95% CI 0.0 to 0.15) and no participant in the derivation sample with an eating disorder scored in it. This is the band the ESP is built to identify with confidence.
2–2
Borderline
Uninformative on its own. The likelihood ratio for exactly two abnormal answers was 0.85 (95% CI 0.38 to 2.0), meaning this result barely shifts the probability either way. It meets the standard threshold for further assessment, but the assessment, not the score, will decide the question.
3–4
Screen positive
Further assessment is clearly indicated. The likelihood ratio was 11 (95% CI 6.4 to 18), corresponding to a post-test probability of 59% at the derivation sample's 12% prevalence.
The ESP is a screening instrument, not a diagnostic test. It cannot diagnose an eating disorder on its own. If you have any concerns, discuss them with a healthcare professional.
04 - Validation evidence
How well it performs.
The ESP was validated in 225 people - 129 University of London students and 96 consecutive attendees at a large North London general practice - against the Questionnaire for Eating Disorder Diagnoses, a self-report operationalisation of DSM-IV criteria completed before the interview and sealed. Eating-disorder prevalence in the sample was 12%.
At the threshold of two or more abnormal answers, the ESP detected every case in that sample: sensitivity 100% (95% CI 90 to 100) with specificity 71% (95% CI 64 to 77), a negative likelihood ratio of 0.0 and a positive likelihood ratio of 3.4. The same study administered the SCOFF to the same participants and found it less sensitive, at 78% with specificity 88%, leading the authors to conclude that the SCOFF could not safely exclude a diagnosis.
Independent replication has been less favourable, and specificity is where it falls down. In 402 female US veterans, the ESP retained high sensitivity at 97% but specificity dropped to 40%. Reviewing both studies, the United States Preventive Services Task Force judged the evidence for the ESP consistent and precise for sensitivity but inconsistent and imprecise for specificity, and rated it insufficient overall. A 2025 scoping review reported an area under the curve of 0.684 for the ESP against 0.947 for the SCOFF.
100%
Detecting eating disorders
40-71%
Correctly clearing those without
LR− = 0.0
Ruling out
AUC 0.684
Overall discrimination
05 - How it compares
How it compares to the alternatives.
Instrument
Items
Time
When to reach for it
ESP
4
~1 min
Reach for it when ruling out is what matters and a high false-positive rate is acceptable.
The better-evidenced brief screen by volume, with pooled sensitivity 84% and specificity 80% across ten studies and 3,684 people. The more balanced choice when you cannot follow up every positive.
When you want a profile of disordered-eating attitudes rather than a risk flag. Copyrighted, so items are not reproduced here.
EDE-Q 6.0
28
10-15 min
The reference self-report severity measure across the last 28 days. Use it to monitor change, which no brief screen can do. Free for non-commercial research use only.
BEDS-7
7
1-2 min
When the presenting concern is binge eating specifically, which brief general screens detect poorly.
06 - When to use it
Right tool, wrong tool.
Reach for it when
-A general practice or general medical consultation where eating has come up and you need to decide in under a minute whether to ask more
-Situations where missing a case is the costlier error and you have capacity to assess everyone who screens positive
-As a documented negative: a score of 0 or 1 is reasonable evidence that an eating disorder is unlikely
Reach for something else when
-Measuring severity, or tracking anyone over time - the ESP supports neither
-Screening children or adolescents, where it has not been validated
-Settings where a 40% to 71% specificity would generate more positives than can be followed up
-Confirming a diagnosis, which requires clinical assessment
-Any use where the four questions would need to be reproduced without permission from the rights holder
07 - Confidence & precision
Reading the score with care.
No Cronbach's alpha and no test-retest coefficient have been published for the ESP. Both were searched for in the primary paper, the USPSTF evidence report and a 2025 scoping review, and neither appears in any of them. Internal consistency would in any case tell you little about a four-item checklist of heterogeneous symptoms containing a reverse-keyed item, since the items are not intended to measure one underlying trait.
This means there is no basis for interpreting a change in score, and no published threshold for meaningful individual change. The ESP should be read as a single yes-or-no decision about whether to assess further, taken once. It is not a repeatable measure and should not be used to track anyone over time.
08 - Limitations
What it cannot tell you.
Specificity is the central problem. At 71% in the derivation sample and 40% in US veterans, a positive result is common among people who do not have an eating disorder, and the volume of false positives will be high in any low-prevalence setting. The instrument is built to rule out, and it should be used for that and little else.
The derivation sample was not fully representative. The student half was recruited by posters and lecture announcements rather than consecutively, and the authors acknowledged that this probably drew in people already concerned about their eating. The observed prevalence of 12% was high compared with other studies. The study was also too small to separate the individual likelihood-ratio strata, which is why the published bands are merged and their confidence intervals wide.
The ESP has not been validated in adolescents or children, in men as a separate group - both validation samples were majority or entirely female - or in non-Western and non-Anglophone settings. It was administered by a psychiatrist in an interview in the original study; completing it alone on a screen is a different task, and no published data establish that the two are equivalent. More broadly, the USPSTF rates the evidence for screening asymptomatic adolescents and adults for eating disorders as insufficient, which is the honest context for any brief screen in this area.
09 - Licensing, explained
How licensing works.
The ESP is published in the Journal of General Internal Medicine and is copyright 2003 the Society of General Internal Medicine, with rights administered by Springer Nature. The full text is free to read in PubMed Central, but the article is not open access, carries no Creative Commons licence, and contains no statement permitting reproduction. Free to read and free to reuse are different things, and this instrument sits in the gap between them. Several clinical websites describe the ESP as free to use; nothing in the primary source supports that. We therefore describe the instrument here but do not reproduce its questions or offer a calculator. The full text, including the four questions, can be read without charge at PubMed Central (PMC1494802): https://pmc.ncbi.nlm.nih.gov/articles/PMC1494802/
See how Aisel removes friction where it costs most. A 20-minute walkthrough tailored to your clinic.
We value your privacy
We use cookies to analyse site usage and improve your experience. Analytics and embedded media (e.g. YouTube) only load if you accept. Read our cookie policy.