Introduction

The adolescent brain is characterized by a heightened drive for reward and an inhibitory control system, which is not fully developed [1, 2]. Partly due to this imbalance, adolescence represents a vulnerable period with regard to development of addiction of substances such as alcohol, marihuana, cocaine, etc. [1, 3, 4]. Addiction to other and more accessible substances with rewarding potential may, however, also be prevalent in adolescence.

Research from the past decade has shown that highly processed foods with high levels of saturated fat and refined carbohydrates (like chips, cookies, pizza, and chocolate) trigger reward-related neural responses resembling those seen for conventional substances of abuse [5, 6]. In some individuals, this may trigger addictive-like consumption of highly processed foods—now commonly referred to as “food addiction” [7, 8]—a phenotype carrying all the hallmarks of addiction of conventional substances of abuse (including continued use despite negative consequences, loss of control, tolerance, withdrawal, and craving), only with food as the abused substance.

It is well established that addictive disorders (alcohol, cannabis, cocaine, etc.) often co-occur with (other) mental disorders, both among adults [9] and adolescents [10]. Therefore, it has been investigated that the same could apply for mental disorder and food addiction. In adults, symptoms of depression, anxiety, and eating disorder have already been widely investigated in relation to food addiction. For instance, the meta-analysis by Burrows et al. [11] found moderately strong associations between food addiction and symptoms of depression 0.459 (95% CI 0.358; 0.550), anxiety 0.483 (95% CI 0.228; 0.676), and binge eating 0.602 (95% CI 0.557; 0.643), respectively. Also, results from the adult FADK sample [12] showed that those with a clinically verified mental disorders had a significantly higher prevalence of food addiction compared to controls from the general population. For instance, 29% of adults with an affective disorder met the criteria for food addiction, compared to only 9% of the general population controls [13, 14]. However, there are only few studies on the potential comorbidity between mental disorders and food addiction among adolescents—most of which suggest a positive relationship as that observed for adults [7, 15,16,17,18,19,20]. However, these studies have typically been based on self-reported symptoms of mental disorder [21,22,23,24], which likely has resulted in report bias. Furthermore, sampling procedures have often been based on self-inclusion, while information on those not participating is lacking, introducing a substantial risk of selection bias. To further elucidate the likely association between food addiction and mental disorder among adolescents, while minimizing the impact of the biases described above, we conducted a study with three sequential aims:

  1. (I)

    To examine the psychometric properties of the dimensional Yale Food Addiction Scale for children 2.0 (dYFAS-C 2.0) via a survey in a random sample of adolescents with clinically verified mental disorders from six major diagnostic categories.

  2. (II)

    To evaluate the generalizability of the results of the survey informed by data from the Danish nationwide registers on sociodemographic and health-related factors available for both the survey respondents and the non-respondents (attrition analysis).

  3. (III)

    To estimate weighted dYFAS-C 2.0 scores across the major diagnostic categories of mental disorders, taking attrition into account.

The main hypotheses of the study were that the dYFAS-C 2.0 is a psychometrically valid measure of food addiction in adolescents with mental illness and that food addiction is highly prevalent among adolescents with mental illness.

Materials and methods

Study design and setting

The study is based on data from the Food Addiction Denmark (FADK) Project, a large survey and register-based study conducted in the country of Denmark in February 2018. The project is described in depth by Horsager et al. [12] and outlined below in brief for the data used in the present study.

Participants

A total of 3529 adolescents aged 13–17 years were randomly drawn from the Danish Psychiatric Central Research Register (DPCRR). The DPCRR contains information on all diagnoses assigned in relation to inpatient, outpatient, and emergency room contacts at Danish psychiatric hospitals since 1995 [25]. To be available for extraction, the adolescents had to have Danish-born parents, and be born in Denmark (to ensure language proficiency). Furthermore, the adolescents had to live with at least one parent/legal guardian with a valid, non-protected Danish postal address. Adolescents living in an institution were therefore not eligible for extraction. If siblings were randomly sampled, they could all participate.

To ensure that all major mental disorders were studied, we sampled randomly within six major diagnostic categories according to the ICD-10 Classification of Mental and Behavioral Disorders—Diagnostic criteria for research (ICD-10) [26]. A total of 625 adolescents were randomly drawn from each category; psychotic disorders (ICD-10 F20), affective disorders (ICD-10 F30), anxiety disorders (ICD-10 F40), eating disorders (ICD-10 F50), autism spectrum disorders (ICD-10 F84), and Attention Deficit Disorders (ICD-10 F90.0, F90.1, F90.8, and F98.8). To be included, one should be registered in the DPCRR with a contact to psychiatric inpatient or outpatient hospital service in the period from 2013 to 2017. Contacts to psychiatric emergency rooms were excluded due to the potential lower validity of the diagnoses assigned in relation to these short encounters. The extraction from the diagnostic categories was hierarchal, with the most severe diagnostic category drawn first (first F20, then F30, and F40, and so forth), and the adolescents could only be included once. Because psychotic disorders are not commonly diagnosed among adolescents, all 404 adolescents diagnosed with a psychotic disorder in the defined time period were included.

Survey procedure

The adolescents were invited via the electronic secure mail system (eBoks) used by Danish public authorities [27]. The invitation was sent to the cohabiting legal guardian’s (and to both in the case, there were two) eBoks account to ensure that they were informed of the purpose of the study, and able to give informed consent on behalf of their child. The invitation included a personal web-link to a web-based questionnaire, and it was specified that the adolescents should fill in the questionnaire themselves. If the invitees had not responded to the invitation within 6 weeks, a reminder and invitation was sent via surface mail.

Ethics

The invitation material included concise information on the study purpose and clearly stated that participation was voluntary, and that the consent to participate could be withdrawn at any time. In Denmark, the legal age is 18 years. Therefore, the legal guardian had to decide whether their child should have the opportunity to participate in the study. For this reason, the invitation was sent to the legal guardian(s) as described above. In Denmark, Ethical Review Board approval is not required, when a survey-based study does not involve biomedical intervention [28]. Access to data from the registers as well as the invitation and study methodology were approved by Statistics Denmark and the Danish Health Data Authority. The project was registered with record number 2008-58-0028 at the Danish Data Protection Agency. After the survey was completed, data from the survey and from the Danish registers were de-identified and stored by Statistics Denmark.

Measures

The included measures are equivalent to those used in prior validation studies of the YFAS 2.0 and the dYFAS-C 2.0 [7, 14]:

The dimensional Yale Food Addiction Scale for children 2.0 (dYFAS-C 2.0) In children and adolescents, symptoms of food addiction can be assessed via the dYFAS-C 2.0 [7], which is modeled upon the DSM-5 criteria [29] for substance use disorder, and represents an updated version of the YFAS-C, which was modeled upon the DSM-IV [15]. The dYFAS-C 2.0 consists of 16 items, which reflects symptoms of addictive behavior (e.g., continued use despite negative consequences, loss of control, tolerance, withdrawal, and craving) and the scoring is based on the year leading up to the assessment. Each of the 16 dYFAS-C 2.0 items are scored from 0 to 4 and the range of the total score is consequently 0–64. The dYFAS-C 2.0 demonstrated good psychometric properties in the original validation study in adolescents with overweight or obesity [7]. In the present study, complete response to the dYFAS-C 2.0 required having answered all of the 16 dYFAS-C 2.0 items. Participants with partial response to the dYFAS-C 2.0 (e.g., only answering three out of 16 items), hindering the calculation of a total score, were categorized as non-respondents.

The Eating Disorder Examination Questionnaire (EDE-Q) The EDE-Q originates from the clinical interview Eating Disorder Examination (EDE), which is validated across age and a variety of settings [30]. For the analyses of the convergent validity, the subscales on eating-, weight-, and shape concern and the total score were used [31]. For the incremental validity and convergent validity analyses, binge eating frequency was used. Finally, for the analyses on the discriminant validity, the subscale on dietary restraint was used [31, 32].

The ADHD Subscale of the Symptom Checklist-92 (SCL-92) The SCL-92 is a well-validated self-report rating scale, which includes several subscales examining a broad range of psychopathology [33]. In the present study, the SCL-92 ADHD subscale was included as a measure of impulsivity and ADHD symptomatology—both of which are associated with food addiction [34]. The SCL-92 ADHD subscale was included as a convergent measure in the construct validity analyses.

The Alcohol Use Disorder Test (AUDIT) The AUDIT is a self-report measure of alcohol dependence and problematic use of alcohol. The AUDIT has been validated across various populations including adolescents [35]. The AUDIT was included as a discriminant measure in the construct validity analyses, as studies have found alcohol dependence to correlate negatively or not to correlate with food addiction [36].

Weight and height Weight and height were self-reported. In children and adolescents, the Body Mass Index (BMI) varies with sex and age, and therefore, the BMI Z-score [37], which takes common growth according to age and sex into account, was used [38]. The BMI Z-score was categorized according to the WHO [38]; underweight/thinness < − 2 SD, normal weight − 2 SD >  + 1 SD, overweight + 1 SD <  + 2 SD, and obese >  + 2 SD. The BMI Z-scores were assessed for outliers, and BMI Z-scores >  + 5.5 and < − 4.5 were considered to be biologically implausible (none in the current study).

Register data for the attrition analyses

Data on demographics, socioeconomics, and health were obtained from the nationwide Danish registers, which contains data on all Danish citizens [39]. Data were available for both respondents and non-respondents, and were linked at an individual but de-identified level. The data included: information on age, sex, parents’ marital status, degree of urbanization (population density at the place of living), and place of living (region) from the Danish Civil Registration System [40]. Information on parental education (highest completed level) from the Population Education Register [41]. Information on parental occupational status from the Registers on Personal Labor Market Affiliation [42]. Information on the family’s equivalized disposable income, which takes the number of family members into account, from the Income Statistics Register [43]. Information on lifetime mental disorder for both parents and invitees from the Psychiatric Central Research Register [25]. Information on lifetime use of psychotropic medication for both parents and invitees from the National Prescription Register [44], and parental and adolescent lifetime medical comorbidity operationalized by the Charlson Comorbidity Index [45] from the National Patient Register [46]. The specified categorization of each demographic, socioeconomic, and health variable is provided in supplement S1.

Statistics

Before all statistical analyses were conducted, the underlying model assumptions were checked. All analyses were conducted using Stata version 15.1.

I: Psychometric validation of the dYFAS-C 2.0

Validation of the dYFAS-C 2.0 The psychometric validation analyses were conducted in accordance with the original work by Schiestl and Gearhardt [7], including confirmatory factor analyses examining the fit for a single factor model using the following fit indices; comparative fit index (CFI), Tucker–Lewis index (TLI), root-mean-square error of approximation (RMSEA), and standardized root-mean-square residual (SRMR) [47], and an examination of the internal consistency quantified using Cronbach’s alpha.

Construct validity The convergent validity and the discriminant validity [7, 15, 32] were examined by Pearson’s correlations comparing the dYFAS-C 2.0 total score with theoretically associated or non-associated constructs (see further description of the hypothesized convergent and discriminant constructs in the section on Measures). Only complete responses on the measures of theoretically associated constructs were used for the correlation analyses, and n is therefore always reported for a given analysis. Pearson correlation coefficients at (|r|) ≥ 0.30 were considered for relevant associations [48] with the significance level at p < 0.05.

Incremental validity For the incremental validity analysis, we used multiple hierarchical regression analysis [7, 32]. The predictive effect on the BMI Z-score was estimated for binge eating frequency and the dYFAS-C 2.0 total score, respectively. In the first model, binge eating frequency was entered as the only explanatory variable for the BMI Z-score. In the second model, the dYFAS-C 2.0 total score was entered along with binge eating frequency to evaluate the unique variance in the BMI Z-score accounted for by each of the two measures.

II: Attrition analysis

The attrition analysis compared the respondents (complete response to the dYFAS-C 2.0) with the non-respondents using descriptive statistics. For categorical variables, proportions were reported, and for continuous variables, means, and standard deviations (SDs) were reported. Comparisons between respondents and non-respondents were carried out using Chi-square test/Fisher’s exact test and by Student’s simple t test, as appropriate.

III: Estimation of the weighted dYFAS-C 2.0 total score

The crude mean dYFAS-C 2.0 total score and 95% CI were reported for the whole population and stratified on the six diagnostic categories. Furthermore, estimates were stratified on sex, and the difference in dYFAS-C 2.0 score between the sexes was tested via the student’s t tests. Because the dYFAS-C 2.0 total score was not normally distributed, bootstrapping with 1000 replications was used to estimate the standard error. The weighted mean dYFAS-C 2.0 scores were computed using augmented inverse probability weighted (AIPW) estimator [49, 50]. The same variables as used in the attrition analyses were included in the weighting of the scores, namely age, sex, parental marital status, parental socioeconomic factors (parental highest educational level, parental occupation status, and equivalised disposable income), urbanization, place of living, prior lifetime somatic illness of the invitee (the Charlson Comorbidity Index), prior lifetime mental disorders of the invitee, and finally prior lifetime use of psychotropic medication of the invitee. Due to a relatively low number of participants in the different diagnostic categories, it was not possible to stratify on specific diagnoses within these categories, as this could have led to identification of individual participants. In addition, the relatively low number of participants in the diagnostic categories violated the assumptions for the AIPW estimator in the sex-stratified analyses. Therefore, only the crude dYFAS-C 2.0 scores are reported for males and females, respectively.

Results

Survey response rate

Figure 1 illustrates the overall survey participation based on complete respondents of the dYFAS-C 2.0, and the response rate of the invitees from the six diagnostic categories.

Fig. 1
figure 1

Flowchart

Full size image

The overall response rate was 12.0% (n = 423), the mean age of the respondents was 15.5 years, and 63.8% of the respondents were female. The response rates for the diagnostic categories ranged from 10.0% for adolescents with psychotic disorders and attention deficit disorders, to 21.5 and 22.5% among those with eating disorders and anxiety disorders, respectively.

I: Psychometric validation of the dYFAS-C 2.0

Factor structure and internal consistency The factor loadings for the 16 dYFAS-C 2.0 items are provided in Supplement S2. The confirmatory factor analyses demonstrated good fit to a single factor model with factor loadings in the range from 0.38 (“I ate certain food all day long”) to 0.87 (“I really wanted to cut down or stop eating certain kinds of foods, but I just couldn’t.”), all with p values < 0.001.The fit indices were; the CFI = 0.85, the TLI = 0.82, the RMSEA = 0.12, and the SRMR = 0.06. The internal consistency Cronbach’s alpha was 0.94.

Construct validity The results of the analyses of convergent validity and discriminant validity of the dYFAS-C 2.0 are provided in Table 1. All measures of eating pathology were moderately or strongly correlated with the dYFAS-C 2.0 total score. The same applied for the BMI Z-score, the ADHD subscale, and AUDIT score who were all positively and moderately correlated with the dYFAS-C 2.0 total score.

Table 1 Correlation matrix illustrating convergent validity and discriminant validity for the dimensional dYFAS-C 2.0 symptom score
Full size table

Incremental validity of the dYFAS-C 2.0 In the model having binge eating frequency as the only explanatory variable for BMI Z-score, binge eating was a significant predictor of the BMI Z-score [t = 3.91, coeff. = 0.25 (0.12; 0.38), p < 0.001] explaining 3.7% of the variance. When adding the dYFAS-C 2.0 total score to the model (n = 395), binge eating frequency was no longer a predictor of the BMI Z-score [t = 0.04, coeff. = 0.003 (− 0.15; 0.15), p = 0.968], but the dYFAS-C 2.0 total score was [t = 5.33, coeff. = 0.033 (0.02; 0.05), p < 0.001], accounting for additional 6.5% of the variance.

II: Attrition analysis

The results of the attrition analysis are shown in Table 2.

Table 2 Attrition analysis comparing respondents with non-respondents
Full size table

Respondents were more likely to be female than non-respondents. With regard to age, parental marital status, and parental occupational status, there were no substantial differences between respondents and non-respondents. Parents of respondents had a higher education level and equivalated disposable income compared to parents of non-respondents. With regard to urbanization and place of living, there were not substantial differences. The parents of the respondents were less likely to have a lifetime mental disorder compared to parents of non-respondents. The same applied for the adolescents, where respondents were less likely to be registered with lifetime use of psychotropic medication compared the non-respondents.

III: Estimation of the weighted dYFAS-C 2.0 total score

The crude and weighted estimates of the mean dYFAS-C 2.0 score are presented in Table 3.

Table 3 dYFAS-C 2.0 total score in the six diagnostic categories of mental disorders, stratified on sex
Full size table

The crude mean dYFAS-C 2.0 score was 13.8 (95% CI 12.6; 14.9), and higher for females 16.0 (95% CI 14.5; 17.5) than males 9.9 (95% CI 8.3; 11.4), p < 0.001. The weighted mean dYFAS-C 2.0 total score 13.8 (95% CI 12.6; 14.9) did not differ from the crude. For all six diagnostic categories, females had higher dYFAS-C 2.0 scores compared to males; however, the difference was only significant for psychotic disorders (p = 0.006) and anxiety disorders (p = 0.031). For females, participants with psychotic disorder had the highest dYFAS-C 2.0 score 23.2 (95% CI 17.4; 29.0). For males, the highest score was found in participants with affective disorders 15.7 (95% CI 8.3; 23.1).

As shown in Fig. 2, the mean dYFAS-C 2.0 score increased going from a lower to a higher weight category.

Fig. 2
figure 2

Mean dYFAS-C 2.0 scores for BMI Z-score weight categories

Full size image

Discussion

In the present study based on data from the FADK project [12], it was demonstrated that the psychometric properties of the dYFAS-C 2.0 were sound among adolescents with clinically verified mental disorder. Attrition analyses informed by sociodemographic and health-related register data on all invitees showed that the parents of the respondents had a slightly higher educational level and income and were less likely to have a lifetime mental disorder compared to the parents of non-respondents. The weighted estimates of the mean dYFAS-C 2.0 total score—taking attrition into account—suggested that symptoms of food addiction are particularly prevalent among adolescents with affective and psychotic disorders.

The psychometric analyses showed that the dYFAS-C 2.0 is a psychometric valid measure of food addiction in adolescents with mental disorder. Specifically, the one-factor model was confirmed and the factor loadings were slightly higher than in the first (and only) validation study of this scale [7]. This also applied to both the internal consistency and the fit indices for the single factor model. Hypothesis testing with theoretically convergent and discriminant measures was used to examine the construct validity. The results were comparable to those from other validation studies of both the adult YFAS 2.0 [14, 32] and the dYFAS-C 2.0 [7]. All measures of eating pathology correlated positively with the dYFAS-C 2.0 score, which is in accordance with the high prevalence of food addiction found in populations with eating disorders [11, 51]. Furthermore, the strong positive association between food addiction and BMI (obesity) observed in studies among adolescents with less psychiatric morbidity [7, 15, 19] seems to extend to adolescents with clinically verified mental disorder. Finally, the dYFAS-C 2.0 accounted for 6.5% of the unique variance in the BMI Z-score, predicting the BMI Z-score over and above binge eating frequency. For comparison, the dYFAS-C 2.0 only accounted for 3.4% of the variance in BMI Z-score in the original validation study by Schiestl and Gearhardt [7]. While these results are promising with regard to the validity of the dYFAS-C 2.0, both research studies and clinical settings would benefit greatly from establishment of clinically informed cut-offs for the YFAS-C 2.0, which would allow for identification of children and adolescents with clinically significant food addiction.

Interestingly, the AUDIT score (problematic alcohol use) correlated positively with the dYFAS-C 2.0 score. This was somewhat surprising as the AUDIT was hypothesized to be a discriminant measure that would “compete” with food addiction [52]—as reported in studies among adults [36, 53]. However, in a prior study of adolescents, Mies et al. also found that food addiction was associated with substance use [54]. This co-occurrence of addictive-like eating and alcohol-related problems in adolescence may be explained from a developmental perspective. In the emerging phase of addiction, adolescents may have a problematic use of several addictive substances at the same time, like alcohol and processed food, since the “drug of choice” is yet to be consolidated. To get a better understanding of the relationship between food addiction and problematic alcohol (and other substances) use in adolescence—and their temporal association—longitudinal studies are needed.

The results of the attrition analysis practically showed the same attrition pattern as for the adult populations [13, 14] with a slight overrepresentation of respondents (or parents of the respondent) with higher education and income. However, when weighting the dYFAS-C 2.0 total scores based on the attrition analysis, they were virtually identical to the crude scores—for all diagnostic categories. This suggests that despite the quite low response rate, attrition from the survey did not impact the estimation of the dYFAS-C 2.0 scores substantially. This is in line with the results from our prior studies of food addiction in both adolescents and adults from the FADK sample, in which the crude and weighted estimates of the prevalence of food addiction did also not differ substantially [13, 14, 55].

Notably, across all diagnostic categories of mental disorder, females had higher dYFAS-C 2.0 total scores compared to males. This is in accordance with prior findings (see the review by Penzenstadler et al. [51]) and indicates that the female preponderance in food addiction is also present for adolescents with mental disorder.

The weighted dYFAS-C 2.0 total score varied substantially across the diagnostic categories, ranging from 9.3 (95% CI 7.3; 11.3) for autism spectrum disorders to 19.4 (95% CI 16.3; 22.5) for affective disorders. For comparison, we recently reported a weighted mean dYFAS-C 2.0 total score of 12.0 (95% CI 11.2; 12.9) for adolescents from the general Danish population [55]. Hence, adolescents with affective and psychotic disorders have substantially higher weighted dYFAS-C 2.0 scores compared to adolescents from the general population.

Somewhat surprisingly, the dYFAS-C 2.0 total scores of participants with eating disorder were relatively low, and among the lowest when compared to the other diagnostic categories. This is in contrast to other studies finding that symptoms of food addiction were highly prevalent among adolescents with eating disorder [21, 22]. The low food addiction symptom load may be at least partly explained by the quite large representation of anorexia nervosa among the participants with eating disorder (n = 59/91). Furthermore, within the eating disorder spectrum, diagnostic crossover between diagnoses are quite common—most often the crossover goes from anorexia nervosa to a bingeing eating disorder [56]. Consequentially, it could be hypothesized that participants with anorexia nervosa—due to their low age—have not yet experienced a diagnostic crossover to a potentially more bingeing eating pattern. Actually, Cinelli and colleagues speculated that food addiction symptomatology in adolescents with anorexia nervosa could predict such diagnostic cross over to binge eating behavior [22].

Among participants with ADHD, the food addiction symptom load was relatively low. Due to the well-known association between ADHD and substance use disorders [57], and between ADHD and obesity [58], this finding was surprising. In addition, several studies have found an association between impulsivity (a core symptom of ADHD) and food addiction [34]. The low food addiction symptom load may be explained by the male preponderance in ADHD [59], and the female preponderance in food addiction. Also, side-effects (appetite suppression) of ADHD medication may play a role [60].

Adolescents with affective disorders, including both depressive disorder and bipolar disorder, had the highest weighted dYFAS-C 2.0 score of all diagnostic categories. Symptoms of depression have been widely investigated in relation to food addiction, and a positive association has been reported both among adults [11, 36] and adolescents [24, 61]. While the relationship between depression and obesity has typically been described as being bidirectional [62, 63], recent studies have proposed that the effect predominantly goes from obesity to depression [64, 65]. This would suggest that food addiction—via obesity—may have the potential of inducing depressive disorder. However, the temporal aspect of the associations between food addiction, obesity, and depression remains unknown, and should be subjected to further study.

The second highest dYFAS-C 2.0 total score was found among respondents with psychotic disorder. This is in accordance with the findings reported by Teasdale et al. (using the adult YFAS 2.0) who found a high prevalence of food addiction among adolescents with psychotic disorder [23]. Among adults, a few studies investigating food addiction in the context of psychotic disorder (schizophrenia) have also found it to be quite prevalent [66, 67]. The appetite stimulating effect of many antipsychotic drugs may explain part of this association [68]. Accordingly, Teasdale et al. found that drug naïve adolescents with psychotic disorder had a lower daily energy intake compared to those in treatment with antipsychotics [23]. The impaired executive function and negative symptoms commonly seen in psychotic disorders may be other contributors to the association with food addiction [69] as these symptoms often manifest as sedentary lifestyle and poor diet with resulting obesity [70]. Also, addiction disorders are frequent comorbidities to psychotic disorder [71], which may explain the proneness toward rewarding highly processed foods. If food addiction indeed plays a role in the development of obesity in psychotic disorders, it could represent an important target for prevention and treatment.

Limitations

The general limitations of the FADK project, which provides data to this study, have been discussed in depth by Horsager et al. [12]. For that reason, the primary focus will be on the limitations related specifically to the examination of food addiction in adolescents with mental disorder. First and foremost, data from the survey were cross-sectional. Consequently, the temporal aspect of the association between food addiction and mental disorder cannot be investigated. Second, there was a quite low response rate of 12.0% (n = 423), and even lower for some diagnostic categories (e.g., ADHD 10.0%). For ethical reasons, the adolescents were invited via their legal guardian(s), and this “barrier” may partly explain the low response rate. Furthermore, it is likely that parents of adolescents with a mental disorder may find that participation would be too stressful for the adolescent. This tendency may be more pronounced among parents of adolescents with more severe psychopathology. Based on the apparent association between food addiction and more severe psychopathology [11], this could have resulted in an underrepresentation of participants with severe psychopathology and a consequential underestimation of the dYFAS-C 2.0 scores. Third, while the availability of register data on a large variety of sociodemographic and health-related data for both respondents and non-respondents enabled the estimation of weighted dYFAS-C 2.0 scores, it is important to note that the AIPW model can only weigh based on the variables included in it. For that reason, it is an important limitation that BMI Z-scores on non-respondents were not available and we were therefore not able to factor in the quite low mean BMI Z-scores of the respondents (compared to the general adolescent Danish population [72]). Due to the strong association between food addiction and BMI [51], the dYFAS-C 2.0 total score estimates are likely underestimated. This is probably also the reason why the dYFAS-C 2.0 total scores in the present study were considerably lower than those found in the original study including a non-clinical sample of adolescents with overweight or obesity [7]. Fourth, to ensure that invitees were able to understand the questionnaire, invitees had to have Danish-born parents and be born in Denmark (approximately 87% of the population in Denmark [73]). This limits the generalizability of the findings. Relatedly, adolescents who were institutionalized or otherwise in the care of the authorities were not included. Fifth, invitees had to be registered in the DPCRR with a contact to a psychiatric inpatient or outpatient hospital service in the period from 2013 to 2017. The period of 5 years ensured that a sufficient number of cases were available for random sampling from each diagnostic category (this was not possible for psychotic disorders where all adolescents with psychotic disorders were included). As a consequence, at the time of the survey, the invitees did not necessarily belong to the diagnostic category from which they were initially drawn, or their mental disorder could have remitted. This adds some uncertainty to the food addiction prevalence estimates across the diagnostic categories investigated in this study. Finally, it is important to note that the construct of food addiction is still a subject of debate [74,75,76]. However, Schulte et al. [77] recently applied Blashfield’s criteria for a new diagnostic category on the food addiction construct, and concluded that a large body of literature now supports that food addiction indeed seems to have clinical utility.

Conclusion

This is the first study to validate the dYFAS-C 2.0 and investigate symptoms of food addiction symptomatology in a random sample of adolescents with clinically verified mental disorder. The findings suggest that the dYFAS-C 2.0 is also a sensitive measure for detecting emerging symptoms of food addiction among adolescents with mental disorder. Furthermore, the results indicate that symptoms of food addiction are quite prevalent among adolescents with clinically verified mental disorder—especially affective and psychotic disorders. As obesity is a tremendous problem in individuals with mental disorder—often having its onset in adolescence—further investigation of food addiction in young people is called for. Such studies will likely contribute to our understanding of the temporal aspects of the association between mental disorder, food addiction and obesity, and thereby aid in the identification of potential transdiagnostic targets for prevention and treatment.

What is already known on this subject?

Adolescence is a high-risk period for development of addictive behavior. Having a mental disorder may increase the risk of developing food addiction as addiction disorders often co-occur with mental disorders.

What this study adds?

The study found food addiction symptomatology to be relatively prevalent among adolescents with affective and psychotic disorders and to correlate positively with BMI.