Attitudes towards healthy eating and dietary behavior have changed in western societies in recent years. Increasing attention is being paid to eating high-quality food [1
]. The desire to eat healthy foods is not in itself a disorder [2
]; however, restrictive eating patterns may contribute to increasing health risks [3
]. An extreme adherence to clean eating (characterized by proper nutrition, restrictive eating patterns, and strict avoidance of foods considered to be unhealthy or impure) [1
] may result in a pathological fixation with healthy eating, called orthorexia nervosa (ON) [4
]. ON is defined as an obsession, a fixation, or preoccupation with healthy food consumption. The concern about healthy diet results in the attention captured by food (preoccupation); thus, evolving to a persistent and disturbing thought (obsession) and a stereotyped behavior (fixation) [5
]. ON is related to the attempt to achieve optimal health by paying attention to healthy dietary intake [6
]. Four classification approaches suggest the diagnostic criteria for ON [7
] which refer to: (1) Obsessional or pathological preoccupation with “healthy”, “pure”, or “clean” foods, (2) rigid avoidance of foods considered “unhealthy” or “unclean”, (3) emotional consequences for transgressing self-imposed dietary rules (distress at violation of food rules/distress when violating food rules), and (4) impairment to social, physical, and/or psychological wellbeing resulting from these food beliefs and behaviors [5
]. It is worth noting that ON has so far not been recognized either by the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) or by the International Classification of Diseases (ICD 11). This may be explained by the still ongoing discussion about the conceptualization of ON and its categorization among other mental disorders (a distinct disorder, a variant of eating disorder, or a variant of obsessive–compulsive disorder). Previous studies have shown that ON is related to disordered eating attitudes [5
], eating disorders symptoms (e.g., dietary restraint, shape and weight concern) [14
], body dissatisfaction [16
], as well as obsessive–compulsive disorder symptoms [18
Despite the lack of a universally accepted definition or diagnostic criteria, ON has been studied progressively in the last two decades, which has resulted in an increased amount of ON measures. To the best of our knowledge, there are thirteen distinct ON assessment tools (in alphabetical order): The Barcelona Orthorexia Nervosa Scale [20
], the Body-Image Screening Questionnaire (BISQ) [21
], Burda-Orthorexia Risk Assessment (B-ORA) [22
], Bratman Orthorexia Test (BOT) [23
], the Düsseldorf Orthorexia Scale (DOS) [24
], the Eating Habits Questionnaire (EHQ) [25
], the Eating Habits Questionnaire-Revised (EHQ-R) [26
], the Orthorexia Nervosa Inventory (ONI) [13
], the Orthorexia Nervosa Scale (ONS) [27
], the ORTO-15 [28
], the ORTO-R [29
], the Scale to Measure Orthorexia in Puerto Rican Men and Women [30
], and the Teruel Orthorexia Scale (TOS) [31
]. The detailed characteristics of ON tools are presented in the recent systematic review by Opitz et al. [32
], showing that existing measures demonstrate questionable psychometric properties (BOT, ORTO-15), challenge preliminary diagnostic criteria (DOS, TOS), and require further evaluation (e.g., EHQ-R, ONS, B-ORA). No gold standard exists for the assessment of ON so far [32
]. Nevertheless, the recent study [33
] found that BOT, EHQ, and DOS are internally reliable self-report instruments (a unidimensional structure—higher ordered for the EHQ, good internal reliability) and their high intercorrelations (r
s > 0.70) support convergent validity, indicating that they essentially measure the same construct. In contrast, the ORTO-15 was found to have an unacceptable model fit, low internal reliability, and medium-sized correlations with the other three ON questionnaires, suggesting that it measures a different construct and/or captures ON tendencies less precisely [33
]. Thus, according to Meule et al. [33
], research on ON should preferably use the BOT, EHQ, or DOS. To have more solid evidence, other researchers [13
] are also considering using the EHQ and DOS questionnaires, which seem to be reliable assessments of ON symptomatology.
Although the ORTO-15 [28
] has been the most widely used tool to assess ON worldwide [35
], the EHQ has increasingly been used. The EHQ [25
] evaluates cognitions (knowledge of healthy eating), behaviors (problems associated with healthy eating), and feelings (feeling positively about healthy eating) towards an extreme focus on healthy eating. The conceptualization of ON, on which the EHQ is based, defines it as an “overwhelming” preoccupation about eating healthfully. To the best of our knowledge, a total of seventeen studies [36
] have used the EHQ. This questionnaire has been translated into Italian [46
] and French [42
] so far. Australian validation of the EHQ also exists [43
]. However, the EHQ has never been validated in a Polish population. The EHQ is an already established tool with promising psychometric properties [43
] (except for the lack of criterion-related validity [47
], and no criticisms have been raised towards EHQ. Therefore, this study aimed to validate the Polish version of the EHQ in a general population sample. In addition, we investigated the association between ON, eating behaviors (cognitive restraint, uncontrolled eating, emotional eating), and body dissatisfaction in a large general population sample of Polish adults.
3.1. Exploratory Factor Analysis (EFA)
The EFA was conducted on a first randomly split sample (n = 502). The Mardia’s coefficient z score was greater than 5 (|z| = 119.080, p < 0.001), indicating that the multivariate normality condition was not met, as well as the univariate normality of the data, supported by a significant Shapiro–Wilk test (p < 0.001). Hence, nonparametric statistics were employed for the Polish validation of the EHQ.
The Bartlett’s sphericity test and the Kaiser–Meyer–Olkin index (KMO
) had correct values (Bartlett test: χ2
(210) = 4472.741, p
< 0.001; KMO
= 0.92), indicating that the data were appropriate for the EFA. The factor extraction method used was the principal axis method, which did not require a normal distribution of the data [57
], and the oblimin rotation was applied.
According to the correlation matrix, all items were significantly correlated with each other. The Cattell index and the Kaiser criterion (eigenvalue > 1), obtained using a parallel analysis, suggested a 3-factor solution. While exploring the factorial solution of the scale, two different models were tested. Initially, six items (items 3, 4, 7, 10, 15, and 20) were removed from the first model because they loaded significantly on two factors. However, after conducting an AFC on both gender groups, the indexes revealed that the model fit was poor in the male sample (CFI and TLI values were below 0.90). Therefore, considering the low R2
= 0.182) of item 16 in the male group, as well as the modification indices, item 16 was excluded from the second and final model, and the correlation between the errors from the items 6 and 8 as well as 8 and 18 were added. The remaining 14 items explained 47% of the total variance. The factors correlated moderately with each other. The correlation between F1 and F2 was 0.34, between F2 and F3 0.36, and between F1 and F3 0.59. The results from the EFA are presented in Table 1
3.2. Comfirmatory Factor Analysis (CFA)
The CFA was carried out on the second randomly split sample (n
= 465) using the MLR estimator. The results presented in Table 2
revealed a good adjustment of the model.
Considering the content of each subscale, the terms “knowledge” (knowledge of healthy eating) incorporating items 1, 5, and 11, “problems” (problems associated with healthy eating) including items 2, 6, 8, 13, 14, 17, and 18, and “feelings and behaviors” (feeling positively about healthy eating and behaviors associated with healthy eating) composed of items 9, 12, 19, and 21 were given to the three subscales of the Polish version of the EHQ (Figure 1
3.3. Multigroup Confirmatory Factor Analysis
Multigroup confirmatory factor analyses were conducted next in order to explore the EHQ’s measurement invariance across gender (female vs. male).
The 967 participants were divided in two groups according to their gender. Thus, the female group was composed of 573 women (Mage = 23.86, SD = 5.77 years) and the male group was composed of 394 men (Mage = 22.62, SD = 3.34 years).
First, a regular CFA was performed separately on each of the two gender groups. The fit indexes of the model tested in the two groups were satisfactory (Table 3
Next, four different models of measurement invariance were explored. The comparison of the configural and metric measurement invariance models reported the following indexes: ∆CFI = 0.022; ∆SRMR = 0.003; ∆RMSEA = 0.01. Even though the ∆CFI was higher than the cut-off value of 0.05, we decided to pursue the analysis, given the fact that the values of the other two indexes were correct and the ∆χ2 was not significant (∆χ2 = 38.93, p > 0.05).
The comparison between models 3 and 2 showed that the additional constraint did not significantly weaken the fit (∆CFI = 0.001; ∆SRMR = 0.000; ∆RMSEA = 0.002; ∆χ2 = 7.696, p > 0.05). Finally, the last two models’ comparison (models 4 and 3) argued that the added constraint did not significantly alter the fit (∆CFI = 0.002; ∆SRMR = 0.003; ∆RMSEA = 0.002; ∆χ2 = 9.619, p > 0.05). These results demonstrate that the factor structure, the factor loadings, as well as the intercepts and measurement errors, were equivalent according to gender.
3.4. Reliability and Sensibility Analyses
Cronbach’s alpha coefficient and the split-half method were performed in order to explore the reliability of the Polish version of the EHQ. The intra-class coefficient with a Spearman–Brown correction, calculated between even and odd items, was equal to 0.86. The values of the Cronbach’s alpha coefficients were 0.85 for “knowledge”, 0.81 for “problems”, 0.81 for “feelings and behaviors”, and 0.88 for the whole scale. Important correlations between items, as well as good Cronbach’s alpha coefficient values, implied a satisfactory reliability of the Polish version of the EHQ. Ferguson’s δ coefficient was 0.98 and the discrimination analysis of the items was satisfactory with the discrimination index being superior to 0.40 for all the items. These results supported the hypothesis of the adequate reliability and sensibility of the Polish version of the EHQ scale.
3.5. Relationships with Sociodemographic Variables
Since the data was not distributed normally, a non-parametric Spearman’s rhô
correlation coefficient was calculated in order to explore the links between different scales. All the correlations are displayed in Table 4
BMI was correlated with EHQ total score (rhô = −0.06) and the “knowledge” subscale (rhô = −0.08). The total EHQ score and its three subscales were strongly correlated with each other. Positive correlations were observed between the total EHQ score and cognitive restraint (rhô = 0.32), uncontrolled eating (rhô = 0.10), and emotional eating (rhô = 0.11). While the “problems” and “feelings and behaviors” scales both presented positive correlations with body dissatisfaction (rhô = 0.14 and rhô = 0.1, respectively), uncontrolled eating (rhô = 0.11), emotional eating (rhô = 0.10 and rhô = 0.12, respectively) and cognitive restraint (rhô = 0.24 and rhô = 0.30, respectively), the “knowledge” scale was only positively related to cognitive restraint (rhô = 0.24).
Non-parametric Wilcoxon–Mann–Whitney U tests were conducted in order to compare the scores from different scales according to participants’ gender. There was a significant difference in the EHQ total score and “feelings and behaviors” across gender: Women had higher scores than men (p
< 0.001). However, the size effect of those differences was quite small: d
= 0.14 and d
= 0.17, respectively. Significant differences across gender were also observed in the mean scores of the cognitive restraint, uncontrolled eating, and emotional eating subscales. The results revealed that men had higher sores on these three variables (p
< 0.001). The size effect of these differences was moderate: d
= −0.54, d
= −0.75 and d
= −0.43, respectively. The results are presented in Table 5
The objective of the present study was to investigate the psychometric properties of the Polish version of the EHQ in a general population sample. The CFA yielded a shorter version composed of 14 items (items 1, 2, 5, 6, 8, 9, 11, 12, 13, 14, 17, 18, 19, 21) for which the three-factor solution showed satisfactory goodness-of-fit (Table 2
). Compared to the original scale, the Polish validation encompassed the same number of factors but a reduced number of items (14 items instead of 21). The Polish version of EHQ also had a similar factorial structure to the French validation [42
]. However, the number of items was not the same: 5 items were removed from the French version [42
], while 7 items were removed from the Polish validation. Furthermore, those two versions had only two removed items in common, items 15 and 16. However, the content of their subscales were quite similar. The subscale “positive feeling of control” from the French version [42
] embodied all the items from the “feelings and behaviors” subscale of the Polish validation. Likewise, the “problems of attention control” and “social relation” from the French version [42
] consisted of almost all the items from the “problems” subscale of the Polish validation. Apart from the same two items being deleted from the French [42
] and Polish validations, there were no other similarities relating to the withdrawn items between the different EHQ validations. In addition, no statistical parameter could explain the removal of the items from the Polish validation. Nevertheless, the majority of the items removed from the Polish validation belonged to the “problems” subscale of the original version. Hence, the withdrawal of those specific items may have been due to the ambiguity of the item’s content in Polish.
The EHQ has been validated in Italian [46
] and French populations [42
] so far. Novara et al. [46
] investigated the psychometric properties of the Italian version of the EHQ among a non-clinical female sample (n
= 204). Results from EFA and CFA (Table 2
) revealed that the 21-item Italian version of the EHQ [46
] represented three ON factors: Knowledge of healthy eating (knowledge), problems associated with healthy eating (problems), and feeling positively about healthy eating (feelings). These findings were consistent with the original three dimensions of the EHQ. The Italian version of the EHQ total score correlated with the Eating Disorder Inventory-3-Referral Form (EDI-3-RF). The psychometric properties of the EHQ were also investigated in a large sample of French adults (n
= 2065) [42
]. Results of the CFA revealed a very good fit of the improved measurement model (Table 2
). A 16-item EHQ [42
] represented three ON dimensions: Rigid eating behavior (eating behavior characterized by the rigid pursuit of a strict diet with many rules; REB), positive feeling of control (feeling of one’s capacity to maintain a healthy diet involving a positive feeling of control; PFC), and problems of attention control and social relationships (difficulty to control attention and to maintain the quality of social relationships because of intrusive thoughts related to the obsession of healthy eating; PACSR). The reliability analysis showed that REB (α = 0.82), PFC (α = 0.76), and PACSR (α = 0.75) measurements had satisfactory internal consistency [42
]. The reliability and validity of the EHQ was also examined on Australian adult women (n
= 286) [43
]. The EFA established that the EHQ represented four ON dimensions: Healthy eating cognitions, dietary restriction, diet superiority, and social impairment. Cronbach’s α coefficient was 0.89 for the total EHQ scale and ranged from 0.72 to 0.80 for the four subscales [43
] Criterion-related validity showed a significant moderate to strong negative correlation between the EHQ and ORTO-10 (a ten-item version of ORTO-15) [43
Reliability analysis for the Polish version of the EHQ across the whole questionnaire showed strong internal consistency (α = 0.88; ICC
= 0.86). The internal consistency of the EHQ subscales, measured by Cronbach’s alpha coefficients, was 0.85 for “knowledge” (knowledge of healthy eating), 0.81 for “problems” (problems associated with healthy eating), and 0.81 for “feelings and behaviors” (feeling positively about healthy eating and behaviors associated with healthy eating). These values were similar to those reported in other validations, i.e., alpha values between 0.75 and 0.90 [25
Our results demonstrated that the total EHQ score was positively correlated with its three subscales (“knowledge”, “problems”, and “feelings and behaviors”), cognitive restraint, uncontrolled eating, and emotional eating, as well as negatively correlated with BMI. This suggests that EHQ score was associated with eating behaviors related to, on the one hand, the intention to control food intake in order to maintain or lose weight and, on the other hand, to consuming more than usual due to a loss of control over intake, accompanied by subjective feelings of hunger and the tendency to eat in response to negative emotions. When cognitive control was undermined, dietary restraint led to reduced sensitivity to internal cues for satiety and, therefore, resulted in overeating [69
]. In the case of ON (using the EHQ), preoccupation with healthy food consumption may be related to decreasing sensitivity to internal cues for satiety and may have an effect on uncontrolled eating (e.g., consumption of foods considered to be unhealthy or impure) or emotional eating. The total EHQ score as well as the “feelings and behaviors” and “problems” subscales were positively correlated to cognitive restraint, body dissatisfaction, uncontrolled, and emotional eating, while the “knowledge” subscale was only correlated to cognitive restraint. The absence of significant correlations between the “knowledge” scale and body dissatisfaction, uncontrolled, or emotional eating may be explained by the fact that individuals who scored high on the “knowledge” scale were satisfied with their body, did not have a problematic eating behavior, and were restraining their food intake in order to keep a healthy weight and avoid health consequences related to overweight and obesity. These findings support those of Depa et al. [70
], which have identified the motivations for weight control as being involved in orthorexic behavior. The authors explained that, related to the orthorexic behavior, weight control can reduce or prevent health consequences engendered by overweight and obesity.
Our results were consistent with the recent study [42
] revealing the negative association between two ON dimensions (REB and PFC) and BMI. In addition, this study noticed that people with high levels of ON symptoms did not typically present high or low BMI, which indicated that ON dimensions were only marginally related to BMI [42
]. Our results were in contrast to a previous study [50
] that found no relationship between ON (measuring by the EHQ) and BMI, and another one [47
] that showed that high BMI was associated with greater ON symptomatology for men (high-BMI and low-BMI women were at equal risk for ON). It is worth pointing out that research investigating the link between ON and BMI have reported inconsistent results. The majority of studies noticed that both high and low BMI are risk factors for developing ON [71
Additionally, our findings showed that women had higher scores on the EHQ total score and on the “feelings and behaviors scale” when compared to men. However, the size effect of those differences was quite small (d
= 0.14 and d
= 0.17, respectively). Findings regarding the gender differences in ON were inconsistent. The recent meta-analysis on gender differences in ON [72
] pointed out that men represented higher normative healthy eating behaviors and showed more problems from this rigid eating behavior, whereas pathologically healthful eating behaviors among women were more linked with having positive feelings about their healthy eating [72
A strength of our study was the use of a large population-based sample, representative of the Polish general population in terms of age and gender. Our study had some limitations. First, the cross-sectional design did not allow for the assessment of test–retest reliability of the questionnaire. Then, due to the self-report nature of the data, the results may have been susceptible to a common method and social desirability biases. Although the risk was reduced by relying on anonymity and assuring voluntary participation, it was impossible to completely eliminate the problem. Finally, a selection bias was present because of the convenience sampling technique followed.
The objective of the present was to validate the Polish version of the EHQ in a general population sample. Future study should evaluate the reliability and consistency of the EHQ in more specific clinical groups (e.g., individuals with ON, individuals with ED) as well as in non-clinical groups (e.g., adolescents). Adolescents may be at high risk for disordered eating behavior. The previous study [73
] has shown that ON prevalence in the adolescent population was similar to that found in the adult population.