Il Diabetes Eating Problem Survey-Revised (DEPS-R) può identificare in modo affidabile la diagnosi di disturbo alimentare nelle donne con diabete di tipo 1?

INTRODUCTION

Type 1 diabetes (T1D) is associated with an increased risk to develop disordered eating behaviours (DEB) and eating disorders (ED) compared to individuals without T1D, 1 and concurrent T1D and DEB/ED are associated with acute and chronic diabetes complications. 2 Moreover, insulin omission is a dangerous diabetes‐specific ED behaviour accelerating morbidity and mortality. 3

Self‐report measures for DEB have been developed for early identification and subsequent intervention. The Diabetes Eating Problem Survey‐Revised (DEPS‐R) is the most prominent diabetes‐specific self‐report measure, including 16 items on dietary habits, diabetes control, insulin misuse and other unhealthy compensatory behaviours for weight management, where a total score of ≥20 suggests further ED assessment. 4 The past 15 years, the DEPS‐R has been translated and validated for several languages with good evidence of validity and reliability. Accordingly, it has been concluded that the DEPS‐R is the best validated screening tool to capture DEB in adolescents and adults with T1D 5 and routine‐based screening with this questionnaire is recommended by international guidelines. 6

Although self‐report questionnaires can be useful for screening and monitoring, diagnosis should be based on a clinical diagnostic interview that assesses EDs according to the Diagnostic and Statistical Manual of Mental Disorders, 5th edition (DSM‐5). 7 , 8 Therefore, the degree to which a self‐report questionnaire correlates with an ED diagnosis is of clinical interest and relevance.

Studies examining this relationship for the DEPS‐R remain limited. Ryman et al. 9 found that the DEPS‐R provided low specificity when adolescents who scored above the DEPS‐R cutoff were referred to an ED specialist unit for further assessment. However, results may be interpreted with caution as the sample size was small (N = 116) and only cases with DEPS‐R ≥ 20 were referred to a diagnostic interview (N = 24), of which only half participated in further assessment. In the Italian validation study of the DEPS‐R, Pinna et al. 10 found that individuals who met the criteria for a DSM‐5 ED showed significantly higher DEPS‐R scores compared to those without an ED (median 22 vs. 10, p < 0.0001). The generalizability of these findings may be limited as the population (N = 211) included adult men and women (median age 38 years) with T1D and insulin‐dependent type 2 diabetes (T2D). Klinker et al. 11 investigated the ability of a shortened 10 non‐insulin‐specific items‐version of the DEPS‐R, the DEPS‐10, to capture binge eating disorder (BED) in men/women (N = 679, mean age 53.8 years) with T1D and T2D and found, using receiver operation characteristic (ROC) analysis, strong discriminatory power (AUC 92%), as did the full DEPS‐R. Even though this study evaluated the performance of the DEPS‐R, the sample was a mix of individuals with T1D and T2D, and it focused solely on BED. Moskovich et al. 12 showed a significant positive association between the DEPS‐R and Eating Disorder Examination using logistic regression analysis. They found that DEPS‐R ≥ 20 accurately detected 81.8% of cases with an ED with a sensitivity of 76.7% and a specificity of 88.2%. The sample was relatively small (N = 83) and included adult women with T1D (mean age 41.5 years). The authors emphasize that future studies should investigate the accuracy, sensitivity and specificity of the established DEPS‐R cutoff in a larger and more diverse sample, using ROC analysis.

Therefore, the overall objective of this study was to empirically evaluate the DEPS‐R against a clinical diagnostic interview and describe other clinical characteristics using baseline data from a large multinational intervention study. Specific aims were to (1) assess the ability of the DEPS‐R to identify DSM‐5 EDs using a ROC analysis, including sensitivity and specificity of the current DEPS‐R cutoff ≥20; and (2) determine which individual items correlated with the presence of any DSM‐5 ED.

RESEARCH DESIGN AND METHODS

The sample consisted of 293 women (age range 14–35 years, 12% adolescents ≤18 years, 88% adults >18 years, mean age 25.5 years ± 5.5) with T1D (mean diabetes duration 11.5 ± 6.2 years) and any level of self‐reported body image concerns, enrolled in a multinational RCT at four sites (Oslo University Hospital, Amsterdam University Medical Center, Stanford University, Joslin Diabetes Center) in three countries (Norway n = 77, the Netherlands n = 60 and the United States n = 156) evaluating the efficacy of a novel ED prevention programme (Diabetes Body Project) (ClinicalTrials.gov NCT05399446). Eighty‐six percent were European‐American, 79% used insulin pumps and 98% used glucose sensors. Exclusion criteria were ED‐related hospitalization or diabetes ketoacidosis in the past year. Ethics approval was obtained at all sites. Written consent was provided by participants and parents for minors. Baseline data from this trial was used for the current paper. More details about the trial can be found elsewhere. 13 , 14

The Diabetes Eating Problem Survey–Revised (DEPS‐R) (16‐items) assesses diabetes‐specific DEB over the past 28 days on a scale from 0 (never) to 5 (always), where higher scores indicate higher levels of disordered eating behaviours. 4 Sum‐scores range between 0 and 80 and a threshold of ≥20 was established to identify cases that should undergo further assessment for ED. 4 The DEPS‐R has been translated and validated into a range of languages 5 and has shown internal consistency (Cronbach's α = 0.84–0.89), 4 , 15 construct validity (r = 0.68) 15 and test–retest reliability (ICC = 0.95). 10 Internal consistency of DEPS‐R was calculated using α.

The semi‐structured Eating Disorder Diagnostic Interview (EDDI) assesses DSM‐5 ED symptoms and ED diagnoses, including threshold and subthreshold anorexia nervosa (AN), bulimia nervosa (BN) and BED, as well as purging disorder (PD). 16 It has demonstrated 1‐week test–retest reliability (α = 0.79) and inter‐rater agreement (α = 0.75), sensitivity to detecting the effects of ED prevention and treatment interventions, and correlates with emotional distress, mental health utilization and functional impairment. 16 Phone interviews were conducted by trained clinicians experienced in EDs. Assessors completed 24 hours of training on structured interviews and ED criteria, practised with simulations, and were required to achieve inter‐rater reliability (k > 0.80) on 12 recorded interviews.

The Satisfaction and Dissatisfaction with Body Parts Scale (nine items) assesses satisfaction with various body parts (e.g., ‘The past month, how satisfied have you been with your waist’) on a scale from 1 to 5. The answers are reverse scored, where higher scores indicate higher levels of body dissatisfaction. 17 The scale has demonstrated internal consistency (α = 0.94), 3‐week test–retest reliability (r = 0.90), and predictive validity for future onset of BN, BED and PD. 16

The Ideal‐Body Internalization Scale–Revised (eight items) assesses the level of pursuit of the thin appearance ideal (e.g. ‘Slim women are more attractive’) on a scale from 1 to 5, where higher scores indicate higher levels of internalization of the thin appearance ideal. The measure has demonstrated internal consistency (α = 0.91), 2‐week test–retest reliability (r = 0.80), sensitivity to detecting intervention effects and predictive validity of future onset of ED. 16 , 18

The Problem Areas In Diabetes (PAID) (20 items) scale assesses diabetes‐related psychosocial distress (e.g. ‘Are you feeling overwhelmed by your diabetes?’) on a scale from 1 to 5, where higher scores indicate higher levels of diabetes distress. 19 The measure has demonstrated internal consistency (α = 0.95), test–retest reliability (ICC r = 0.79) and convergent validity (r = 0.45–0.54). 20

Type 1 Diabetes And Life (T1DAL) (adolescents 23 items, young adults 27 items, adults 27 items) assesses diabetes‐specific health‐related quality of life in T1D (e.g. ‘I am comfortable asking other people for help with my diabetes if I need a hand’) on a scale from 1 to 5, where higher scores indicate higher levels of quality of life. Transforming the scores to a 0–100 scale enables comparison between the age‐specific versions. T1DAL has demonstrated internal consistency (α = 0.88–0.89), test–retest reliability (r = 0.80–0.82) and convergent validity (r = 0.48–0.76). 21 , 22

Objectively measured HbA1c was obtained. Oslo, Amsterdam and Stanford provided self‐administered HbA1c kits to participants through mail, which were returned and analysed at laboratories of the respective sites. Joslin measured HbA1c as part of the baseline visit at the hospital.

Objectively measured TIR, referring to the percentage of time a person's glucose levels are within target range (3.9–10 mmol/L, 70–180 mg/dL) was assessed using continuous glucose monitoring (CGM) with approved commercial glucose sensors.

Statistical analysis was conducted using R version 4.4.2. 23 Distribution of data was assessed using scatter plot, and normality of data was checked through histograms and Q‐Q plots. Completeness of data was 99%–100%, except for HbA1c and TIR which had 85% and 86%, respectively. ROC Analysis was performed with R package ‘pROC’ 24 using DEPS‐R composite sum‐scores and dichotomized diagnostic status for any ED. Area under the curve (AUC) of the ROC ranges from 0.5 to 1.0 where higher values indicate better discrimination of the model. Univariate logistic regression was performed to assess the correlation between individual DEPS‐R items and the presence of any DSM‐5 ED. Linearity between the continuous DEPS‐R items and the logit of the outcome was tested. To characterize and compare the population scoring ≥20 and <20 on the DEPS‐R, a two‐sample t‐test was performed. To assess homogeneity of variances, Levene's test was used. Welch's t‐test was applied if variances were not homogenous between the groups. Cohen's d effect sizes between DEPS‐R group ≥20 and <20 were calculated. To account for multiple comparisons, Bonferroni correction was applied.

RESULTS

The total sample (N = 293) consisted of 35 (12%) adolescents ≤18 years (Norway n = 22, 28%; Netherlands n = 6, 10%; United States n = 7, 4%) and 258 (88%) adults >18 years. In total 32 (11%) women met DSM‐5 criteria for an ED diagnosis (Norway n = 8, 10%; Netherlands n = 7, 12%; United States n = 17, 11%), of which four were adolescents (Norway n = 2; Netherlands n = 1; United States n = 1). ED diagnosis included: AN (n = 2), BN (n = 13), BED (n = 16) and PD (n = 1). For individuals with an ED, the mean age was 25.8 years, the mean BMI was 26.9 kg/m2, mean TIR was 63%, mean HbA1c was 54 mmol/mol (7.1%), mean illness duration was 12.9 years, and the mean DEPS‐R score was 32.

The ROC analysis showed an AUC of 0.82 (95% CI 0.79–0.94) (Figure 1). The current DEPS‐R cutoff of ≥20 yielded a sensitivity of 87.5% (95% CI 83.6%–91.3%) and a specificity of 60.4% (95% CI 54.8%–66.0%). Internal consistency of the DEPS‐R was 0.88 (95% CI 0.84–0.89).

ROC curve of DEPS‐R against Eating Disorder Diagnostic Interview (EDDI). AUC, area under the curve; AUC of 100% indicates perfect discrimination.

Two alternative DEPS‐R cutoffs were identified. Based on the Youden Index, a cutoff of ≥25 showed an ideal trade‐off between sensitivity and specificity with 75.0% (95% CI 70.0%–79.9%) and 77.9% (95% CI 73.1%–82.6%), respectively. A cutoff of ≥17 yielded a sensitivity of 96.8% (95% CI 94.8%–98.8%) and a specificity of 49.2% (95% CI 43.4%–54.9%).

Univariate logistic regression revealed significant correlations between 12 items (1, 3, 4, 5, 6, 8, 9, 11, 13, 14, 15, 16) and the presence of any DSM‐5 ED, where one unit increase on the item scales increased the odds of having an ED by the factor of the respective odds ratio (OR) (Table 1). The strongest correlations were found for items 9 (‘I try to keep my blood sugar high so that I will lose weight’), 8 (‘I make myself vomit’), 13 (‘After I overeat I skip my next insulin dose’), 14 (‘I feel that my eating is out of control’), 15 (‘I alternate between eating very little and eating huge amounts’) and 5 (‘I eat more when I am alone than when I am with others’), with ORs of 3.64 (95% CI: 2.15–6.24), 2.85 (95% CI: 1.60–5.22), 2.36 (95% CI: 1.55–3.69), 2.23 (95% CI: 1.71–2.99), 1.99 (95% CI: 1.55–2.59) and 1.99 (95% CI: 1.51–2.69), respectively.

Correlation between DEPS‐R items and the presence of any DSM‐5 eating disorder.

Note: p‐values, Bonferroni adjusted p‐values, odds ratios (OR) and 95% CI are reported. Inclusion of 1 in the 95% CI indicates p ≥ 0.05.

Abbreviations: DEPS‐R, Diabetes Eating Problem Survey‐Revised; DSM‐5, Diagnostic and Statistical Manual‐5 eating disorders; NS, not significant.

In total 131 (44.7%) participants scored ≥20 on the DEPS‐R (Norway: n = 38, 49%; Netherlands: n = 24, 40%; United States: n = 68, 44%). Compared to the group scoring below 20, they showed significant differences with large effect sizes for diabetes distress (d = 1.13), ED symptoms (d = 1.03), diabetes‐specific quality of life (d = −0.99) and body dissatisfaction (d = 0.86); and with small to medium effect sizes for HbA1c (d = 0.53), BMI (d = 0.48), TIR (d = −0.46) and thin ideal internalization (d = 0.39). There was no significant difference in age (p = 0.65) or illness duration (p = 0.68) between the groups (Table 2).

Group differences between positive and negative screen at DEPS‐R cutoff ≥20.

Note: Two‐sample t‐test/Welch's t‐test* was conducted. Mean, SDs, p‐values, Bonferroni adjusted p‐values, 95% CI and between group effect sizes (Cohen's d) are reported.

Abbreviations: DEPS‐R, Diabetes Eating Problem Survey‐Revised; NS, not significant.

DISCUSSION

This is the first study to investigate the ability of the DEPS‐R to identify any DSM‐5 ED diagnosis, established in a diagnostic interview, using a ROC analysis. An AUC of 0.82 implies that the overall model shows good performance. There is an 82% chance that the model correctly identifies whether a person has an ED or not based on the DEPS‐R score. Further, results demonstrate that the threshold ≥20 identifies most cases (sensitivity 87.5%) with an ED. However, specificity is moderate with 60.4%, hence the false positive percentage is high with 39.6%.

The findings of our study in a sample of young women with T1D and body image concerns are somewhat different from results reported by other studies addressing sensitivity and specificity of the DEPS‐R. Moskovich et al. showed lower sensitivity and a higher specificity in adult women with T1D, 12 Ryman et al. showed lower specificity when adolescents above the threshold were referred to ED assessment, 9 and Klinker et al. reported similar sensitivity but higher specificity in adults (men/women) with T1D and T2D using the DEPS‐10 for BED. 11 Differences in study populations, objectives and analytical approaches limit the ability to make direct comparisons between findings and highlight the need for future studies to replicate the current findings in large representative samples.

Our results may spark a discussion on alternative DEPS‐R cutoff scores. An ideal trade‐off between sensitivity and specificity was found for a threshold of ≥25, whereas ≥17 identified all individuals with an ED, except for one case with a DEPS‐R sum‐score of 5 and a partial AN diagnosis. Moskovich et al. 12 tested alternative thresholds and found that cutoff values between 16 and 22 showed good accuracy, whereas Klinker et al. 11 identified an optimal threshold of 15 for the DEPS‐10 for BED. In line with our results, there is a common understanding that simply lowering the cutoff would not increase the ED detection rate. 12 Also, our findings show that lowering the threshold to ≥17 is accompanied by half of the population being identified as false positives, where high referral rates to mental health services would be costly and needlessly distressing for the individuals. This may support a potential stepwise screening approach, as it has been suggested for the DEPS‐10. 11 On the contrary, one might argue that a self‐report screening measure should aim for identifying all ED cases and therefore should be intentionally low, especially given the severe consequences of comorbid T1D and ED and the lack of specialized treatment facilities. 2 , 3 , 25

Identifying which DEPS‐R items are associated with having an ED is of interest, especially considering stepwise clinical screening approaches. In total, 12 items (1, 3, 4, 5, 6, 8, 9, 11, 13, 14, 15, 16) were significantly correlated with the presence of any ED, whilst 4 items where not (2, 7, 10, 12). This indicates that most items in the DEPS‐R address ED symptoms and behaviours that are relevant for ED‐positive cases. The strongest correlations in descending order were found for items 9, 8, 13, 14, 15 and 5. The first three items—9, 8 and 13—identify ‘intentional hyperglycaemia’, ‘vomiting’ and ‘insulin omission after overeating’, respectively. They do entail pathological, compensatory ED symptoms of greater severity. Following, items 14, 15 and 5 capture ‘loss of control’, ‘alternating between binge/restrictive eating’ and ‘eating in secrecy’, respectively. These capture broader DEB which may vary in severity. Priesterroth et al. 26 identified four distinct DEB profiles for individuals scoring ≥20 performing latent profile analysis, where two profiles were classified as moderate DEB (restrained eating, disinhibited eating) and two were classified as severe DEB (maintaining high blood glucose, dual compensatory behaviours). While our study examines the correlation between individual items and any ED diagnosis, the findings may extend this classification by identifying specific items that are indicative of moderate and severe DEB. Overall, these items support a binge/purge ED phenotype described in literature for comorbid T1D and ED. 1 Further, Abild et al. 27 reported correlations between DEPS‐R items and the Youth Eating Disorder Examination Questionnaire (YEDE‐Q) in adolescents. Although the findings are not directly comparable with the current results due to differences in measures and population, both studies indicate that item 10 ‘I try to eat to the point of spilling ketones in my urine’ may has limited relevance and could be omitted, as was done in the Dutch version. 28

In our population, 45% of women scored above the DEPS‐R threshold ≥20 and about 1 in 10 qualified for a DSM‐5 ED diagnosis, supporting our findings that specificity is moderate. Prevalence of DEB (indicated by DEPS‐R scoring ≥20) varies, depending on the population studied and has been reported up to 48% in adolescents, 29 which is similar to 49% in the Norwegian sub‐population with the highest percentage of adolescents. Since most participants in this study were adults, one might hypothesize that the percentage scoring above cutoff would be even higher in a solely adolescent population. 30 Positive correlations of DEPS‐R ≥20 with HbA1c, BMI and diabetes distress are in line with previous literature. 4 , 10 , 12 , 31 Noteworthy is that some of the largest effect sizes were observed for diabetes distress and diabetes‐specific quality of life, underscoring the significant correlation with other psychological aspects. Merwin et al. 32 suggested individuals with a DEPS‐R score above 20 could have a ‘Bulimia’, ‘Binge Eating’ or ‘Overeating’ profile. While the current study has focused on DEPS‐R thresholds regardless of ED phenotypes, the most frequent EDs in our study were BED and BN, underpinning the findings from Merwin et al. 32 Future studies might explore the sensitivity and specificity of phenotype‐based DEPS‐R thresholds.

The present findings may provide potential clinical implications. Based on international guidelines, routine‐based screening for DEB in adolescents with the DEPS‐R is recommended 6 and outpatient clinics should strive to utilize the DEPS‐R in clinical practise. To increase sensitivity, lowering the threshold to ≥17 could be considered. For individuals scoring above this threshold, one might use selected items to initiate a conversation in a clinical setting. For example one might utilize items 14, 15 and 5 as entry questions to explore if and to what extent a patient identifies with ‘I feel that my eating is out of control’, ‘I alternate between eating very little and eating huge amounts’ or ‘I eat more when I am alone than when I am with others’. As clinicians seem to be hesitant assessing eating problems in fear of causing harm, 33 items 14, 15 and 5 do not reveal potentially harmful ED behaviours such as insulin omission. Items 9, 8 and 13 might be utilized as an extension to the initial questions, assessing the severity of DEB. Endorsing in ‘intentional hyperglycaemia’, ‘vomiting’ or ‘insulin omission’ for weight loss purposes may indicate referral to specialized mental health services for further ED assessment. Introducing this as an intermediate step for those who screen positive could reduce unnecessary referrals to mental health services and mitigate the high false‐positive rate associated with lowering the threshold. Moreover, this approach could enhance clinicians' skills in addressing sensitive issues like DEB. Finally, the Diabetes Body Project, a T1D‐specific ED prevention programme, has shown promising acute effects in reducing ED risk factors and symptoms in two randomized trials, 13 , 34 and may present a targeted intervention for individuals with body image concerns in adolescent and adult diabetes care.

Strengths and limitations need to be mentioned. Strengths include a large, multinational sample, with few exclusion criteria and overall high completeness of data. Limitations are that the study population only included young women, who were predominately European‐American, limiting generalizability. We faced challenges in recruiting adolescents, which may limit representativeness of the results for this population. Though the overall sample is large, the relatively small number of participants diagnosed with an ED (primarily adults, primarily BED and BN), compared to the overall sample size, may impact the stability and generalizability of the ROC curve and the AUC estimate. Also, enrolling in the Diabetes Body Project ED prevention trial could have led to a selection bias, increasing the prevalence of DEB. Future research should confirm these findings in larger, representative samples and assess the DEPS‐R in more diverse groups, including males.

In conclusion, the DEPS‐R showed good performance to identify correct ED status. The current DEPS‐R cutoff ≥20 showed high sensitivity and moderate specificity. The strongest significant correlation with having an ED was found for items 9, 8, 13, 14, 15 and 5. Collectively, the findings expand the knowledge of the DEPS‐R and suggest potential implications for clinical practise.

AUTHOR CONTRIBUTIONS

Severina Haugvik contributed to the planning of the study, collecting data, data analysis and writing the manuscript. Mareille H. C. L. Hennekes contributed to the planning of the study, collecting data and writing the manuscript. Maartje de Wit has contributed to the planning of the study, collecting data and writing the manuscript. Elena Toschi contributed to planning the study, collecting data and writing the manuscript. Christopher D. Desjardins is the project statistician and provided statistical support during the analysis and revising the manuscript. Torild Skrivarhaug has contributed to planning the study and revising the manuscript. Knut Dahl‐Jørgensen has contributed to planning the study and revising the manuscript. Eric Stice is co‐PI of the study, has contributed to planning the study, collecting data and writing the manuscript. Line Wisting is PI of the study, has contributed to planning the study, data collection and writing the manuscript.