Eating disorders (ED) are a group of psychiatric disorders characterized by disordered eating habits and excessive focus on one's weight. They are most common in young females, among whom they present an important cause of physical and psychosocial morbidity. It has been estimated that 0.3–2.2% of young Western females suffer from anorexia nervosa 1234 and 1–2% from bulimia nervosa 124. Mortality from anorexia nervosa is as high as 5.0% 5. Early detection and treatment improve the prognosis of eating disorders 67.

Detecting eating disorders early is difficult because of the cryptic presentation of the disorder. A simple screening method for detecting eating disorders would be very useful for primary care and for student healthcare. Such screen would be valuable for research purposes as well. A few screening tools have been developed 8910, but they are often lengthy and may be difficult to interpret for a non-specialist. Recently, a new screening tool, the SCOFF questionnaire, was developed to overcome these limitations 6. The SCOFF is a simple and memorable instrument of five questions intended to raise suspicion of an existing eating disorder. So far, only a limited number of studies concerning the validity of the SCOFF have been performed (table 1). The previous estimates of the sensitivity, specificity, positive predictive value, and negative predictive value of the SCOFF have varied between 53.3–100%, 21–94.4%, 24.4–81%, and 88.7–99.3%, respectively, depending on sample characteristics (table 1). While sensitivity and specificity are properties of the instrument, positive predictive value depends on both the test and the population in which it is used. Lower disease prevalence in the population leads to lower positive predictive value. Accordingly, positive predictive value has been best in studies of treatment-seeking young adults and in adolescents and lowest in studies on unselected primary care patients. The SCOFF has also been used in other recent studies which did not specifically focus on its validity as an eating disorder screen 11121314.

A general population study setting is appropriate for investigating whether a screen is suitable for detecting cases when there is no selection based on symptoms in the study population. In this study, we tested the validity of the SCOFF in screening current eating disorders in Finnish young adults. To our knowledge, this is the first study reporting the validity of the SCOFF in a population-based sample of young adults.

SCOFF was designed to be a simple, memorable screening instrument for primary care, intended to raise suspicion that an eating disorder might exist before rigorous clinical assessment 6. This is the first study in which the SCOFF is being evaluated in young adults from the general population. We found fairly similar sensitivity and specificity for the recommended SCOFF cut-off point as in previous studies conducted in primary care 27313233 or among students 3132343536. However, the same cut-off produced unacceptably low positive predictive (PPV) value for screening eating disorders from the general population. On the other hand, none of the subjects with current eating disorder scored zero point in the SCOFF, and thus SCOFF might be useful for ruling out eating disorder.

Our results concerning the sensitivity and specificity of the SCOFF when using two or more positive answers as the cut-off point were quite consistent with previous studies made with similar research frame 343536 or with samples representing primary care patients 27313233. However, PPVs were considerably better in these studies than in our study. The PPV is important especially when identifying relatively rare disorders such as eating disorders, because their detection can be ineffective among general population even with instruments with relatively high specificity and sensitivity 37. The problem of low PPV has also been discussed previously by Mond and colleagues 33.

The SCOFF seems to work considerably better among risk population than among general population. The studies testing the SCOFF among patients with probable eating disorder diagnosis versus healthy controls 6738 report almost exclusively excellent validity figures. It might be that these excellent figures are at least partly due to the research frame; the subjects are in easily recognizable advanced stages of the disease while the controls consist of healthy people only. This leads to a marked difference between these two groups and can therefore give overoptimistic results 37. Even people attending general practice already differ from the general population because they present some symptoms and seek treatment for some problem. People with eating disorders consult general practitioners more frequently than healthy young adults even prior to the diagnosis 39 and thus there might be an over-representation of eating disorders in these study samples. In unselected population samples of young adults eating disorder prevalences are rather low; in our sample the prevalence estimates for lifetime anorexia nervosa and bulimia nervosa among women were 2.14% and 2.25%, respectively 40, which are comparable with previous estimates in studies with similar research frame 4.

It also might be that the SCOFF works best among adolescents. This would explain the greater PPV of the study made by Caamaño et al 34, in which the sample consisted of 11–13-year-olds.

Other screens for eating disorders in general (Stirling Eating Disorder Scales (SEDS), BEDT, Eating Disorder Diagnostic Scale (EDDS)) have shown PPVs comparable to our results, but population-based knowledge of these instruments is limited 41424344. Mond et al reported relatively high PPV (30%) for Eating Disorder Examination Questionnaire (EDE-Q) in a population-based study of adult women 45. Many screens are limited to a specific disorder like Eating Attitudes Test (EAT-40 and EAT-26) to anorexia nervosa and The Bulimic Investigatory Test (BITE) and (BULIT) to bulimia nervosa. Compared to SCOFF most of the screens mentioned above are longer and more difficult to interpret and have shown relatively low positive predictive values 46. Our results are in line with Jacobi et al. suggesting that further research is needed to identify target populations for SCOFF 46. The negative predictive value of the SCOFF has been excellent in this and previous studies, suggesting that it could be used to rule out eating disorder, while persons with at least some symptoms of disordered eating should be assessed in more detail.

There are some limitations in our study. As discussed earlier, eating disorders are rare in the general population. In our study sample, we identified ten persons with current disorder from 546 interviewed persons, and twenty with lifetime eating disorder who were currently in remission. Because of the small number of cases with current eating disorder in our general population sample, the confidence intervals of our estimates were relatively wide. This could have only been avoided by increasing the sample size considerably. The SCID interview also has its limitations. It is widely used for diagnosing psychiatric disorders, but the denial of the disease and non-disclosure by patients, common in eating disorders, may complicate identification of the disorders even in the SCID interview. On the other hand, we were able to utilize case notes and health care register data in our diagnostic assessment, making them exceptionally reliable for a general population study.

Altogether, the screening questionnaire was sent to 1863 persons and returned by 1316 (70.6%). This might have affected the generalizability of our results, if eating disorders were more common in non-responders. Our previous study showed that those who returned the MEAF questionnaire were younger, more often women, had more often graduated high school, and had less often than non-responders been treated for any mental health problem in a mental or general hospital according to the Finnish Hospital Discharge Register 16. There were no other socioeconomic differences, nor differences in the frequency of psychiatric symptoms reported in the baseline interview 16. However, although persons with hospital-treated disorders were less likely to return the questionnaire than others, we had access to their case notes and could ascertain their diagnoses. There was only one person with a lifetime history of eating disorder in this group, suggesting that the questionnaire non-response did not introduce any major bias in the study.

There was attrition in the MEAF interview as well: only 55.5% of those invited participated in the interview. However, attrition in the MEAF interview depended on age, sex, and education, but none of the scores in any of the screens we used for the mental health interview differed between interview participants and non-participants 16. However, the item in SCOFF describing attitude related to appearance differed between questionnaire responders and non-responders. The difference was too small (9.1% vs. 13.4%) to introduce any major bias, but it may suggest that young people with a negative attitude toward their appearance are less likely to participate in a survey.

Even though the SCOFF did not prove to be an efficient tool in screening eating disorders from the general population, it could be a valid tool in ruling one out. When utilizing the threshold of one or more positive answers, all the eating disorder cases were detected giving us a sensitivity of 100%. This leads to conclusion that if a person has zero positive answers in the SCOFF questionnaire, it is very unlikely that she/he suffers from an eating disorder. The SCOFF could be a helpful tool in ruling out eating disorders in primary care and student health services.

Detecting eating disorders among the general population has its own challenges. In such a population, the SCOFF could be used at most as an aid in ruling eating disorders out. Further studies are still needed to establish the performance and ideal target population of the SCOFF.

The authors declare that they have no competing interests.

SL drafted the manuscript. JMS conducted the statistical analyses. JMS, JP, TA-S and SIS did final diagnostic assessments. JMS, TA-S and JL obtained funding for the study. All authors contributed to the interpretation of data, participated in revising the manuscript for intellectual content, and approved the final version of the manuscript.