The aim of this study was to use the Eating Attitudes Test-26 (EAT-26) as a screening instrument on a specific population with a marked prevalence of binge eating disorder (BED) and eating disorder not otherwise specified (EDNOS). The EAT-26 questionnaire was used in order to identify the high-risk subjects for referral to clinical evaluation.
EAT-26 was administered to 845 subjects who, for the first time, came to the Nutritional Medicine Service looking for a diet between January 1999 and December 2002. From this initial sample, subsequently, 250 subjects were randomly selected and administered a semistructured clinical interview for DSM-IV (SCID I, version 2.0).
Discriminant analysis provided a cutoff value of EAT-26=11. Logistic regression analysis indicated high Dieting (D) or Bulimia (B) subscale scores as a risk factor of EDNOS or bulimia nervosa (BN) cases, respectively; on the other hand, a high Oral Control (O) subscale score represented a protecting factor for BED cases.
Our study tried to assess the usefulness of EAT-26 as a screening instrument for obese patients attending a Medical Nutritional Service. Results from this study suggest that a cutoff score of 11, lower than that indicated in the literature, improves the diagnostic accuracy of the EAT-26 in a high-risk setting regarding sensibility level (68.1%) and leading to a reduction of the false negative rate (31.9%).
Eating disorders (ED), which include anorexia nervosa (AN), bulimia nervosa (BN), binge eating disorder (BED) and eating disorders not otherwise specified (EDNOS), are characterized by extreme emotions, attitudes and behaviours surrounding weight and food issues.1
In western women aged 12–30 years, the prevalence of AN and BN is 5 and 14%, respectively.2 Epidemiological information regarding EDNOS is poorly represented and this may be due to the uncertainty of the diagnostic criteria. Cotrufo et al.3 and Stein et al.4 reported, respectively, a prevalence rate of 14.4 and 34% in a non-clinical population, whereas Mizes and Sloan5 and Nicholls et al.6 found a prevalence rate of 26 and 58% in clinical samples.
BED has only recently been recognized as a distinct condition. To date, according to DSM-IV, only research criteria have been investigated.
When psychometric instruments are applied for both clinical and epidemiological purposes to assess EDs they fail to be adequately specific and accurate.
The Eating Attitudes Test (EAT), proposed by Garner and Garfinkel,9 is one of the most widely used self-report ED instruments to date. Its application as a clinical and research tool has produced a large amount of literature in this sector.10
Originally developed to diagnose AN, it has been used as a screening method for non-clinical populations.11
EAT is presented in a self-report format which is easily administered and scored even in a non-specialized setting.
EAT-26 is a highly sensitive and reasonably specific instrument. It is able to identify subjects at high risk of ED,12, 13 but further clinical evaluation is required in order to make a correct diagnosis.14
The low specificity of EAT-26 could depend on the presence of subjects with eating symptoms related to a disturbance in eating habits, whose frequency or intensity do not meet DSM criteria for EDs diagnosis.11
To date there is little data on the instrument as a screening tool for BED and EDNOS cases when applied to a population at high risk of ED.
The use of EAT-26 as a screening instrument for outpatients at high risk of BED and EDNOS could produce a high rate of false negativity.
If the test were administered to a population of Clinical Nutrition Unit attenders, and thus at high risk of BED and EDNOS, its loss of sensibility could be solved by identifying a new cutoff point, in order to minimize the probabilities of misclassification (false positive and false negative).
The aim of this study was to assess the psychometric characteristics of the EAT-26 questionnaire in a population considered to be at high risk of EDNOS and BED, so as to identify a new threshold capable of better differentiating ED cases from non-cases, and of discriminating among ED diagnoses.
The study was carried out at Udine University Hospital's Psychiatric Clinic and at Udine's General Hospital Clinical Nutrition Unit in Italy.
EAT-26 was administered to all subjects who, between January 1999 and December 2002, came to the Clinical Nutrition Unit for a diet for the first time. Informed consent was obtained from all participants and confidentiality ensured.
Then, a sample of 250 subjects was randomly selected from the initial population to administer a semistructured clinical interview (SCID I, version 2.0). In this study, SCID I data were chosen as ‘gold standard’, to identify ED cases from non-cases, and to differentiate ED disorders (AN, BN, BED and EDNOS).
Eating Attitudes Test-26
EAT-26 is an abbreviated 26-item version of the EAT-40, created by Garner et al.13 Similarly to the EAT-40, a cutoff at 20 is used to determine ED cases from no-cases.
The questionnaire is characterized by a good correlation between the emotional distress of the subject and body image, but it provides limited data regarding bulimic behaviour. Furthermore, the result seems to be entirely independent of variations in weight.
The subdivision of EAT-26 into three subscales B (bulimia), D (dieting), O (oral control) allowed us to obtain more information from the same questionnaire:
Factor 1-D (13 items) is closely correlated with a distorted body image;
Factor 2-B (six items) is closely associated to body weight; it provides information about body image and tendency towards bulimic behaviour;
Factor 3-O (seven items) reflects the tendency to self-control. High scores in this area are related to low weight and to the absence of bulimia.
SCID I -version 2.0. (Structured Clinical Interview for DSM-IV)
SCID I18 is a semistructured interview used for the formulation of DSM-IV axes I and II diagnoses. It brings together simple and easy administration of the interview with the advantage of a reliable and accurate diagnosis.
In this study SCID I was chosen as ‘gold standard’, to identify ED cases from non-cases, and used as a diagnostic instrument (AN, BN, BED and EDNOS).
The Shapiro–Wilk and Levene's tests, respectively, were used to assess normality and homogeneity of variances of the subjects' descriptive variables. Then, on an exploratory mission to compare their distributions, we used χ2 and Fischer's exact tests for categorical variables, depending on the assumptions check. To optimize EAT-26 as a screening test, a discriminant analysis was performed after verifying the assumptions.
For the discriminant analysis, we used the SCID I (version 2.0) results, as gold standard, to determine ED cases from non-cases. Age and body mass index (BMI) were chosen as covariates to be included in the model.
To assess the validity of EAT-26 to discriminate between ED typologies, a logistic regression was carried out. This analysis allowed us to determine which subscale of EAT-26 could better predict every ED typology.
To test if high scores to one of the subscales could identify subjects at risk of a specific ED, we used the SCID I (version 2.0) results to determine different ED typologies and thus, to dichotomize outcome variables.
Five subjects (two positive and three negative to SCID) were removed from logistic analysis, because of the incompleteness of the EAT-26 subscales scores.
Data were analysed with SPSS software 11.5th edition. Results were considered statistically significant when P⩽0.05.
EAT-26 was administered to 845 people who came to the Clinical Nutrition Unit. Among these patients, 250 subjects were randomly selected and 231 (92%) completed the SCID-I interview. Reasons for non-inclusion were refusals to be interviewed. No differences in age, gender or marital status arose between interviewed and non-interviewed subjects.
All 231 subjects were 15–65 years old (mean age: 42; s.d. 13 years) and their mean BMI was 32.5 kg/m2 (s.d. 6.7 kg/m2); 192 subjects were females (mean age: 41.8; s.d. 13.1 years; mean BMI: 32.24; s.d. 6.9 kg/m2) and 39 males (mean age: 45.15; s.d. 12.77 years; mean BMI: 33.87; s.d. 5.53 kg/m2).
Table 1 shows the gender distribution by age, BMI and diagnosis. No AN cases turned up in this sample. As compared with male subjects, in the female sample all diagnoses, and in particular EDNOS, were much more represented (χν=32=13.4; P=0.004).
In order to carry out the discriminant analysis, we first checked that all statistical assumptions were respected. Age and BMI showed a normal distribution, while EAT-26 total scores were normalized through a square root transformation, which also allowed a stabilization of the variances in the positive (n=119) and negative (n=112) SCID groups.
Discriminant analysis provided a cutoff value of EAT-26=11; its overall misclassification rate (OMR) was 33.3%, with a false positive rate of 34.8% and a false negative rate of 31.9% (Table 2). Age and BMI, introduced as covariates, did not show any significant advantages in order to maximize the separation between the groups.
Table 3 shows the SCID-I diagnoses of the 231 subjects according to three EAT-26 groups of patients with different scores. The cutoff scores that were chosen were those identified by our discriminant analysis (11) and the score proposed in 1982 by Garner et al.13 (20). Medium (11–19) and high (⩾20) EAT-26 scores were more representative of ED patients than of patients without ED (χν=22=26.7; P=<0.001). Moreover, EAT-26 scores were evenly distributed among ED diagnoses. The highest EAT-26 scores were reported by 80% of BN patients and by 35.6% of EDNOS patients. On the contrary, they were also reported in a minority of patients without diagnosis (12.5%) and in BED patients (9.1%) (χν=62=51.1; P<0.001).
Table 4 shows the EAT-26 mean scores on each of the D, B and O subscales by SCID-I diagnoses. A preliminary comparison among ED showed that the highest subscale D mean scores were reported by subjects with BN (17.5±7.6) and EDNOS (11.4±6.5), the highest subscale B mean scores were reported by subjects with BN (9.1±5.7), whereas the mean scores on O subscale were lowest for subjects with BED (0.8±1.4).
The logistic regression analysis showed an association between different SCID diagnoses and EAT-26 subscales:
A high score in subscale B was the only risk factor for subjects with diagnosis of BN (OR=1.46; 95% CI: 1.23–1.74).
A high score in subscale D was the only risk factor for subjects with diagnosis of EDNOS (OR=1.07; 95% CI: 1.03–1.12).
A high score in subscale O, instead, was the only protecting factor for subjects with diagnosis of BED (OR=0.77; 95% CI: 0.59–0.99).
EAT-26 can adequately screen ED cases when applied in a specialist setting14 and Garner et al.13 demonstrated that 20 was the best cutoff score in this setting. A different issue is the detection of cases in non-psychiatric settings not specifically addressed to treat major ED.
Differences in ED between subjects attending non-specialist health services and psychiatric services have been observed. For example, the prevalence of extreme weight control (self-induced vomiting, laxative misuse, extreme dietary restriction, etc.) is low in our sample of attenders to a Nutritional Unit. Therefore, we cannot expect a cutoff score of 20 in EAT-26 to perform well in detecting BED and subsyndromal cases of AN and BN.19 Other instruments, or a different EAT-26 cutoff score is therefore needed. In this regard, Mond et al.20 proposed the use of another questionnaire, the EDE-Q,21 as an alternative to the EAT-26 to screen general population samples. These authors criticized EAT-26 stating that this questionnaire, although frequently used to detect probable ED cases in general population surveys, was actually developed to assess the specific behaviours and attitudes of AN in a specialist setting. Consequently, its validity as a case-finding instrument is not adequately supported. However, according to the same authors, even the EDE-Q represents the problem of an underestimation of subclinical variants of BN and BED.20 In conclusion, this problem reflects the inherent difficulty to assess subclinical cases by self-report measures, rather than a particular weakness of a specific questionnaire.
Our study showed that the use of a cutoff score of 20 on a group of subjects suffering from weight problems led to a large amount of FN (68.1%), with a very low sensibility (31.9%). On the other hand, the discriminant analysis showed that the cutoff score of 11, which was associated with the lowest OMR (33.3%), allowed to improve the validity of EAT-26 as a screening questionnaire. Compared with the great improvement of the sensibility (+113%) and a reduction of the OMR (−19%), a decrease from 20 to 11 in the cutoff score did not substantially worsen the specificity (−25%).
Logistic regression showed that high D or B subscale scores are risk factors for EDNOS cases and for BN cases, respectively. Moreover; a high O-subscale score, which measures the ability to demonstrate feeding self-control, represented a protective factor for the BED subjects subgroup. These results confirm the data in the literature,13 which showed an association between high B-subscale scores and negative prognosis cases.
The main limitation of our study is the absence of AN subjects. Our aim was to assess the usefulness of EAT-26 as a screening instrument in a population attending a Clinical Nutrition Unit, where patients seek help for obesity or nutritional intervention and AN subjects are not common. The latter subjects usually seek help from specialists working in a psychiatric setting.
In this situation, the availability of a screening instrument that can be used by another non-specialized personnel is important. EAT-26 with a cutoff score of 11 demonstrated to be a useful tool. Moreover, even if it is not a proper diagnostic instrument, the possibility to differentiate the information of the three subscales allowed us to obtain broader information about diagnostic and prognostic matters. The application of this easy and quick questionnaire therefore seems to be of great usefulness in a non-specialist setting.
American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorder, 4th edn. Washington, DC 1994, trad. it. DSM-IV. Manuale diagnostico e statistico dei disturbi mentali. Masson: Milano, 1996.
Szmukler G, Dare C, Treasure J . Handbook of Eating Disorders, Theory, Treatment and Research 1995. Wiley: New York.
Cotrufo P, Barretta V, Monteleone P, Maj M . Full-syndrome, partial-syndrome and sub clinical eating disorders: an epidemiological study of female students in southern Italy. Acta Psychiatr Scand 1998; 98: 112–115.
Stein D, Meged S, Bar-Hanin T, Blank S, Elizur A, Weizman A et al. Partial Eating Disorders in Community Sample of Females Adolescents. J Am Acad Child Psychiatry 1997; 36: 1116–1123.
Mizes JS, Sloan DM . An empirical analysis of eating disorder not otherwise specified: preliminary support for a distinct subgroup. Int J Eat Disord 1998; 23: 233–242.
Nicholls D, Chater R, Lask B . Children into DSM don't go: a comparison of classification systems for eating disorders in childhood and early adolescence. Int J Eat Disord 2000; 28: 317–324.
Cotrufo P . Il disturbo da alimentazione incontrollata: dati epidemiologici e caratterizzazione clinica. Giornale Italiano di Psicopatologia 1999; 5/3: 229–232.
Yanovski SZ, Nelson JE, Dubbert BK, Spitzer RL . Association of binge eating disorder and psychiatric comorbidity in obese subject. Am J Psychiatry 1993; 150/10: 1472–1479.
Garner DM, Garfinkel P . The eating attitude test: an index of the symptoms of anorexia nervosa. Psychol Med 1979; 9: 273–279.
Garner DM, Olmstead MP, Polivy J . Development and validation of a multidimensional eating disorder inventory for anorexia nervosa and bulimia. Int J Eat Disord 1983; 2: 15–34.
Mintz LB . The eating attitude test: validation with DSM IV Eating Disorder Criteria. J Pers Assess 2000; 74/3: 489–503.
Dotti A, Lazzari R . Validation and reliability of the Italian EAT-26. Eat Weight Disord 1998; 3: 188–194.
Garner DM, Olmsted MP, Bohr Y, Garfinkel P . The eating attitudes test: psychometric features and clinical correlates. Psychol Med 1982; 12: 871–878.
Williams P, Hand D, Tarnopolsky A . The problem of screening for uncommon disorders – a comment on the eating attitudes test. Psychol Med 1982; 12: 431–434.
Anstine D, Grinenko D . Rapid screening for disordered eating in college-aged females in the primary care setting. J Adolesc Health 2000; 26/5: 338–342.
Morgan HG, Macquire-Samson IMP, Purgold J, Vassilas CA . Outcome assessment. In: Herzog W, Deter HC and Vandereycken W (eds.). The Course of Eating Disorders. Springer-Verlag: Berlin, 1992, pp 30–39.
Garner DM, Garfinkel PE . Socio-cultural factors in the development of anorexia nervosa. Psychol Med 1980; 10: 647–656.
First MB, Spitzer RL, Gibbon M, Williams JBW . Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I), Clinician Version 1997. American Psychiatric Publishing Inc.: Arlington, VA, USA.
Hay PJ, Marley J, Lemar S . Covert eating disorders: the prevalence, characteristics and help-seeking of those with bulimic eating disorders in general practice. Primary Care Psychiatry 1998; 4: 95–99.
Mond JM, Hay PJ, Rodgers B, Owen C, Beumont PJV . Validity of the Eating Disorder Questionnaire (EDE-Q) in screening for eating disorders in community samples. Behav Res Ther 2004; 42: 551–567.
Fairburn CG, Beglin SJ . Assessment of eating disorders: interview or self-report questionnaire? Int J Eat Disord 1994; 16: 363–370.
About this article
Cite this article
Orbitello, B., Ciano, R., Corsaro, M. et al. The EAT-26 as screening instrument for clinical nutrition unit attenders. Int J Obes 30, 977–981 (2006). https://doi.org/10.1038/sj.ijo.0803238
- screening instrument
- eating disorders
- nutritional service
Clinical Nutrition (2020)
Psychology of Women Quarterly (2018)
Frontiers in Psychology (2018)
“Trait EI in the relationship between needs fulfilment and symptoms and attitudes associated with EDs”
Mental Health & Prevention (2018)