Exercise addiction risk in recreational athletes: a cross-sectional study of psychological distress and food choice motivations in Jordan Authors: Nour Amin Elsahoryi, Tamara Shukri Abu-Alhija, Omar Amin Alhaj Published: 2026 ABSTRACT BackgroundExercise addiction (EA) represents an emerging mental health issue among athletic communities, manifested by compulsive exercise with negative psychological and physiological consequences. Although EA phenotypically coincides with overtraining syndrome and burnout, no study has explored this athlete mental health continuum in recreational athletes, especially regarding comorbid psychological distress and maladaptive food choice motivations.ObjectiveTo explore EA risk as an athlete mental health concern among Jordanian recreational athletes, examining the EAI screening-positive rate, associations with psychological distress and maladaptive food choice motivations, and multifactorial risk predictors including demographic, behavioral, and motivational variables.MethodsA cross-sectional study recruited 1,036 recreational athletes (18–45 years) across Jordanian fitness facilities. Subjects completed validated Arabic versions of the EA Inventory, Depression Anxiety Stress Scale-21, and Athlete Food Choice Questionnaire. Binary logistic regression determined independent predictors.ResultsOverall, 47.8% screened at risk for EA (22% low activity, 30% moderate, 66% high; p ≤ 0.001). At-risk individuals showed considerably high stress, depression, and anxiety (p ≤ 0.001), and valued performance, weight control, and nutritional optimization over sensory appeal and cultural food. Multivariable modeling (R2 = 0.336; accuracy = 72.9%) identified independent predictors: male gender (OR = 1.66), high physical activity (dose–response: low OR = 0.13, moderate OR = 0.20 vs. high reference), lower stress levels compared to extremely severe stress (ORs 0.29–0.47; Normal through Moderate levels reaching conventional significance), and performance-oriented eating (OR = 1.12). Living alone (OR = 0.60) and hedonic food motivations showed protective associations.ConclusionsThis first integrated assessment of EA risk in an Arabic-speaking recreational-athlete sample identified a high EAI screening-positive rate within a fitness-oriented Jordanian sample, together with multifactorial associations involving male gender, higher recent activity, psychological stress, and maladaptive eating patterns; however, this screening-positive rate should be interpreted cautiously because the EAI is a screening instrument rather than a diagnostic tool. Introduction Physical activity offers numerous health benefits, including reducing the burden of chronic diseases and improving the mental and physical health of athletes across various sports [1, 2]. Nevertheless, among a group of people, exercise habits could shift to patterns of compulsive exercise where one cannot control and experiences negative outcomes [3, 4]. The concept of exercise addiction (EA) and its psychological and nutritional correlates is an essential mental health issue on the athletic continuum, both at the recreational and elite levels [4]. Theorization of EA as an addiction behavior has six basic components such as salience, mood change, tolerance, withdrawal, conflict, and relapse [5]. Interestingly, EA does not meet Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM-5) or International Classification of Diseases, 11th revision (ICD-11), criteria; available instruments such as Exercise Addiction Inventory (EAI) are screening tests that are capable of identifying persons at risk but not diagnostic [4]. The literature differentiates between the primary EA—where compulsive exercise develops on its own and secondary EA—where excessive exercise is a compensatory mechanism in the pathology of eating disorders [6, 7]. In any particular study without parallel screening of the eating disorder this difference cannot be operationalized and not taking it into consideration may overblow the prevalence estimates as well as confound etiologic mechanisms [4]. The current research thus regards any result about weight-control and nutritional-attribute patterns as a possible indicator of uncontrolled eating as opposed to peculiarities of EA. These trends are representative of both true differences of population and methodological heterogeneity. In addition, psychiatric comorbidity studies also indicate higher levels of mood conditions, anxiety, symptoms of ADHD and childhood trauma exposure in individuals at risk of EA, implying that behavioral addictions share similar vulnerability mechanisms [8]. Published estimates of exercise addiction risk vary across athletic populations and are strongly influenced by methodological features, particularly the screening instrument, cut-off threshold, and sampling frame. Accordingly, the present findings should be interpreted primarily against studies using comparable EAI-based assessment. In one such study, 7.6% of national-level elite athletes screened positive using the EAI [9]. This comparison is informative because the present sample was recruited from a self-selected recreational fitness context rather than a structured, professionally supervised elite-sport environment. Differences in recruitment context, exercise setting, and participant self-selection may therefore contribute meaningfully to variation in EAI screening-positive rates across studies. The systematic reviews report strong linkages between EA risk and high levels of depression, anxiety, and stress symptoms, and medium-large effect sizes were observed in cross-sectional studies [10, 11]. Obsessive–compulsive characteristics and disturbed body image are also linked to EA risk, but the causation is not clear [12]. These comorbidity patterns also indicate that EA needs to involve multidimensional psychological assessment instead of considering compulsive exercise as a single occurrence [11]. EA is conceptually and phenomenologically related to established mental health issues in athletes, especially overtraining syndrome (OTS) and burnout [13]. Similar to OTS, EA implies fatigue, injury, or medical contraindication training, physiological indicators such as hormonal dysregulation and immune suppression, and psychological symptoms such as mood disturbance, anxiety, and depression [14, 15]. Likewise, burnout among athletes which is manifested through emotional/physical depletion, decreased feelings of accomplishment and devaluation of the sport often accompanies compulsive exercise behaviors where training is no longer optional but an obligatory requirement [13]. The triangular interrelationship between EA risk, psychological distress, and maladaptive eating behaviors could reflect a single mental health syndrome of athletes at all levels of competition, but most studies have centered on elite athletes who are already the minority of all regular exercisers when it comes to subjected pressure to appear, perform, and identify [4]. The conceptualization of EA as a component of the athlete mental health continuum, as opposed to a behavioral pattern, allows more holistic screening and intervention strategies to be used in a recreational, amateur, and elite environment. It is also emerging evidence that people in the EA risk group exhibit unique food choice habits that put more emphasis on performance, weight control, and nutritional optimization over sociocultural or sensory value of food [7, 16]. The conceptual similarity of this instrumental eating orientation to orthorexia nervosa pathological fixation on healthy eating, moderately correlated with EA measures, demonstrates itself [17, 18]. The neurobiological mechanism on high-volume exercise, which may be associated with hormonal changes (depressed leptin and T3), results in food preoccupation coupled with dietary restraint when chronic high-volume exercise is combined with insufficient energy intake, which provokes feedback loops supporting compulsive exercise and fixed eating behavior [14]. Directional relationships between EA risk and psychological distress and food choice motivations are, however, poorly described, especially in non-athlete populations [4]. The EA literature is still disproportionately Western, and there is little evidence of the Middle East [4]. The existing studies in the region are highly specific: a Saudi study (n = 150) found the prevalence of EA risks 13.1% and associated it with depression and anxiety but did not determine the motivation behind food choices [19], and a Lebanese study (n = 352) modeled the EA-eating disorder pathways through body image but did not examine the food choice motivation [20]. There are no existing studies that have concomitantly investigated EA risk, psychological distress, and holistic food choice motives among Arabic-speaking populations of athletes using validated and culturally adjusted measures [20]. This loophole restricts theoretical generalizability in sociocultural contexts and practical creation of culturally competent athlete mental health screening guidelines in an area where the fitness industry is developing rapidly, and sports participation is on the rise [6, 19]. Previous studies usually analyze EA correlates individually and do not attempt to model the psychological symptoms and dimensions of food choice simultaneously [4]. The Athlete Food Choice Questionnaire (AFCQ) offers a trusted nine-domain scale of food choice drivers beyond binary eating disorder screening that is capable of measuring subtle motivational configurations [16, 21]. Patients can be performance-oriented and inflexible regarding their food preferences and perpetuate the cycle of compulsive physical activity and still have a lower score than clinical eating disorders [7]. Thus, the proposed research was designed to explore EAs risk as an athlete mental health issue in the group of Jordanian recreational athletes, the prevalence of this issue, the associations with psychological distress (depression, anxiety, stress), and maladaptive food choice behavior, as well as multifactorial risk indicators in the range of demographic, behavioral, and motivational factors- extending the research on athlete mental health to the under researched group of recreational athletes. Materials and methods Study design, participants, and setting The proposed cross-sectional mental health study of recreational athletes was carried out in the period between March and May 2025 with the help of a self-administered online questionnaire, which was sent via Google Forms [22]. Ethical approval of the study was granted by the Research Ethics Committee of the Faculty of Pharmacy and Medical Sciences of the University of Petra (Approval Number: S/2/3/2025) and it was conducted in accordance with principles of the Declaration of Helsinki in terms of ethical conduct during observational studies and reporting observational studies according to the guidelines of Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) [22]. The participants were recruited based on the interest of the fitness groups, university athletic clubs and local fitness centers in Jordan to the convenience and snowball sampling of the target social media platforms (Facebook, Instagram, WhatsApp) [23]. This multi-platform strategy aimed to maximize recruitment of recreational athletes while acknowledging the inherent limitations of non-probability sampling. The survey's first page provided informed consent detailing study purpose, estimated completion time (15–20 min), voluntary participation, and data confidentiality [24]. The consent was expressed by the participants choosing between the following options: I agree to participate or I do not agree to participate; the participants who consented to participate were given a chance to fill out the questionnaire. Out of 1,068 participants who visited survey link, 16 did not provide consent and 16 more individuals were disqualified because of incomplete information (completion rate: 97.0%), which resulted in analysis sample of 1,036 people. The eligibility criteria were as follows: 18–45 years of age; physical activity self-reported according to the World Health Organization (WHO) recommendations (at least 150 min/week moderate-intensity or 75 min/week vigorous-intensity physical activity) sustained at least 6 months; no current injuries or medical contraindications to exercise; not pregnant or lactating; Arabic literacy; and provision of informed consent [2]. It should be noted that these criteria reflected participants' habitual exercise status at the time of recruitment; the IPAQ-S, administered as part of the study instruments, assessed physical activity during the preceding seven days only, meaning that some participants who met the habitual activity eligibility threshold may have recorded lower activity levels during that specific week due to normal variation in weekly training load. The exclusion criteria were: professional or elite athlete status (i.e. having participated in national/international events or having been given an athletic scholarship), a diagnosis of severe psychiatric disorders that may affect questionnaire understanding (e.g. schizophrenia, bipolar disorder), and undergoing active eating disorder treatment [25]. Naturally, these requirements were coded in terms of screening questions at the beginning of the survey, where the ineligible respondents were automatically dismissed [26]. This focus on recreational athletes targets the largest population in athlete mental health research—individuals who face similar psychological stressors around performance, appearance, and identity as elite athletes but remain understudied [4, 27]. Every eligibility criterion was self-reported, which is a known limitation of large-scale online surveys [28]. Data collection Electronic data collection was carried out during the period between March and May 2025, with all answers being saved automatically in a password-restricted database of Google Forms, available to the research team [24]. The recruiting was spread across various geographic areas in Jordan and different types of platforms (social media groups, university networks, fitness center partnerships) to reduce selection bias that is present in non-probability sampling and increase the heterogeneity of the sample in terms of age, education, and socioeconomic status [23]. Missing data were assessed prior to analysis: of the 1,068 individuals who accessed the survey link, 16 did not provide informed consent and 16 were excluded for incomplete responses (overall exclusion rate: 3.0%). Among the 1,036 included participants, no missing values were identified on any study instrument, as the survey platform required completion of all items before submission. Sample size calculation Multivariable logistic regression was to be used with events-per-variable (EPV) method to determine sample size [29]. A conservative EPV of 15 was defined to guarantee the stability of coefficients and reduce the effect of the small sample size typical of complex models- it was higher than the 10 EPV threshold of traditional methods and in line with current methodological advice on complex models [30]. The minimum sample size was computed as N = (EPV × k)/event proportion = (15 × 12)/0.30 = 600 participants with the number of candidate predictors (k) = 12, and a hypothetical prevalence of the EA risk is 30%. This minimum was exceeded by the obtained sample (N = 1,036). Ultimately, 27 candidate predictors were entered, and the post-hoc EPV value was found to be 495 events/27 predictors = 18.3, which is greater than the conservative EPV of 15, and so the sufficient model stability and generalizability of the results are ensured in spite of the increased final set of predictors [29, 31]. Study instruments The online questionnaire comprised five validated sections (instruments) that measured sociodemographic traits, EA risk, physical activity level, food choice motives, and psychological distress. The sociodemographic characteristics were age, sex, education level, marital status, job, body mass index (BMI), health/sport field membership, sports nutrition education, personal trainer supervision, chronic disease condition, smoking, living situation, and monthly income. The EAI is a 6-item assessment tool measured on a 5-point Likert scale (1 = strongly disagree to 5 strongly agree), and the items were related to salience, mood modification, tolerance, withdrawal, conflict, and relapse [5]. The scoring scale is used to give each item a mark between 1 and 5 with a total mark of 6–30. A minimum score of 24 would declare the participants to be at-risk of EA, 13–23 would reflect that they are symptomatic, and 12 and below would reflect that they are asymptomatic [32]. In the current research, all analyses used the binary (at-risk ≥ 24 vs. not-at-risk < 24) classification as the dependent variable. The level of physical activity was assessed based on the International Physical Activity Questionnaire-Short Form (IPAQ-S) that is a seven-day recall objective instrument that measures amount of walking, moderate-intensity, and vigorous-intensity activity in domains (leisure, domestic, occupational, transport), and the total weekly metabolic equivalent (MET) minutes were categorized as low (< 600), moderate (600—2999), or high (≥ 3000) activity [33]. The 32-item (AFCQ) was used to assess the motives of food choice. The rating of each item is based on the 5-point Likert scale (1 = not at all important to 5 = extremely important), and the items will be divided into nine conceptual domains, including nutritional attributes, emotional influences, health awareness, social influences, habitual practices, weight control, values/beliefs, sensory appeal and performance. Domain scores were computed by adding the items in each domain as a higher score means that motive is more important in food selection [21]. Using the Depression, Anxiety, and Stress Scale-21 (DASS-21) a 21-item self-report tool, the psychological distress was measured by assessing the level of depression, anxiety, and stress through a 3-subscales scale (depression, anxiety, and stress) rated on a 4-point frequency scale, and subscale scores were multiplied by two to make a comparison with the normative data and be classified as normal, mild, moderate, severe, or extremely severe [34]. All instruments were administered in validated Arabic versions. Sociodemographic information Sociodemographic variables were assessed through structured self-report items (Table 1). Age was categorized as < 25, 25–30, or > 30 years to facilitate comparison with prior EA research in young adult populations [4]. BMI was calculated from self-reported height and weight (kg/m2) and classified per WHO criteria as underweight (< 18.5), normal (18.5–24.9), overweight (25.0–29.9), or obese (≥ 30.0) [35]. Smoking status distinguished never-smokers, former smokers, and current smokers (any tobacco product: cigarettes, electronic devices, waterpipe). Monthly income was reported in Jordanian Dinars and categorized based on regional income distributions. Physical activity level was quantified using the Arabic-validated International Physical Activity Questionnaire-Short Form (IPAQ-S), which assesses frequency and duration of walking, moderate-intensity, and vigorous-intensity activities across four domains (leisure, domestic, occupational, transport) during the preceding seven days [36]. Standard Metabolic Equivalent of Task (MET) value (walking = 3.3, moderate = 4.0, and vigorous = 8.0) was multiplied by the activity duration and added up in categories to calculate MET-minutes per week [33]. The total number of minutes of MET per IPAQ scoring protocols was low (< 600), moderate (600–2999), and high (≥ 3000) [33]. The analysis of data involved elimination of biologically implausible data (> 960 min/day in any category) to increase the reliability [37]. Table 1Social Demographics Characteristics of Participants (n = 1036)VariableCategoryFrequency (n)Percentage (%)GenderMale59457.3Female44242.7Age (years)Less than 2547645.925–3031230.1More than 3024823.9Education level ≤ High School12512.1Diploma/Bachelor80477.6Postgraduate10710.3Marital statusSingle78976.2Married22221.4Other252.4BMIUnderweight373.6Normal59957.8Overweight30429.3Obese969.3Job StatusNo job1009.7Student43742.2Part-time12512.1Full-time35033.8Retired242.3Health or sports fieldYes48747.0No54953.0Courses in sports nutritionYes26125.2No77574.8Personal trainerYes25224.3No78475.7Chronic diseasesYes1039.9No93390.1Smoking statusNon-smoker49047.3Ex-smoker16515.9Current smoker38136.8HousingAlone13813.3With family89886.7Income (JOD)Less than 30039838.4300–50034533.3More than 50029328.3Values are expressed as frequency (n) and percentage (%). Percentages may not total 100% due to rounding. Body Mass index (BMI) categories were defined according to WHO classifications. Income values are in Jordanian Dinar (JOD). Categories were grouped as follows: less than 300 JOD (approximately 423 USD), 300–500 JOD (approximately 423–705 USD), and more than 500 JOD (above 705 USD) based on the exchange rate at the time of analysis (1 JOD ≈ 1.41 USD) Table 1Social Demographics Characteristics of Participants (n = 1036)VariableCategoryFrequency (n)Percentage (%)GenderMale59457.3Female44242.7Age (years)Less than 2547645.925–3031230.1More than 3024823.9Education level ≤ High School12512.1Diploma/Bachelor80477.6Postgraduate10710.3Marital statusSingle78976.2Married22221.4Other252.4BMIUnderweight373.6Normal59957.8Overweight30429.3Obese969.3Job StatusNo job1009.7Student43742.2Part-time12512.1Full-time35033.8Retired242.3Health or sports fieldYes48747.0No54953.0Courses in sports nutritionYes26125.2No77574.8Personal trainerYes25224.3No78475.7Chronic diseasesYes1039.9No93390.1Smoking statusNon-smoker49047.3Ex-smoker16515.9Current smoker38136.8HousingAlone13813.3With family89886.7Income (JOD)Less than 30039838.4300–50034533.3More than 50029328.3Values are expressed as frequency (n) and percentage (%). Percentages may not total 100% due to rounding. Body Mass index (BMI) categories were defined according to WHO classifications. Income values are in Jordanian Dinar (JOD). Categories were grouped as follows: less than 300 JOD (approximately 423 USD), 300–500 JOD (approximately 423–705 USD), and more than 500 JOD (above 705 USD) based on the exchange rate at the time of analysis (1 JOD ≈ 1.41 USD) Social Demographics Characteristics of Participants (n = 1036) Values are expressed as frequency (n) and percentage (%). Percentages may not total 100% due to rounding. Body Mass index (BMI) categories were defined according to WHO classifications. Income values are in Jordanian Dinar (JOD). Categories were grouped as follows: less than 300 JOD (approximately 423 USD), 300–500 JOD (approximately 423–705 USD), and more than 500 JOD (above 705 USD) based on the exchange rate at the time of analysis (1 JOD ≈ 1.41 USD) Exercise addiction inventory (EAI) The EAI demonstrates acceptable psychometric properties across diverse populations, with internal consistency ranging from α = 0.80–0.84 [5, 32]. In this study, the EAI was translated into Arabic and culturally adapted in accordance with established cross-cultural validation guidelines [38]. The adaptation was limited to linguistic translation and minor contextual adjustments to ensure comprehension; the item content, number of items, response format, and scoring procedure were retained unchanged from the original instrument to preserve comparability with prior EAI-based studies. A pilot sample was used to psychometrically test the Arabic version before deployment of the main study. Bilingual experts translated the EAI into Arabic using forward–backward translation, with discrepancies resolved through expert panel consensus [38, 39]. Pilot psychometric validation (N = 30; 50% female; age M = 29.43 ± 2.91 years; exercise frequency M = 4.47 ± 1.01 times/week) was done on the members of a fitness center in Amman, through convenience sampling. Internal consistency was excellent (Cronbach's α = 0.92, 95% CI [0.88–0.96]; McDonald's ω = 0.93), with strong two-week test–retest reliability (r = 0.89, p ≤ 0.001) and item-total correlations (r = 0.72–0.86). Five experts rated content validity as excellent (CVI = 0.94). The Exploratory factor analysis (EFA) provided a one-factor solution, which explained 74.18 per cent of variance (loadings: 0.77–0.89), which was similar to the original EAI structure [5]. Confirmatory factor analysis demonstrated excellent model fit: χ2(9) = 10.26, p = 0.33; Comparative Fit Index (CFI) = 0.99; TLI = 0.98; RMSEA = 0.049 (90% CI [0.00–0.078]); SRMR = 0.038. Positive correlation with the frequency of exercise per week (r = 0.76, p ≤ 0.001) supported convergent validity. Although the pilot validation sample (n = 30) was sufficient according to the cross-cultural adaptation rules [40], the entire study sample (n = 1,036) made it possible to confirm a high level of internal consistency (α = 0.91) and factorial structure, which validates the reliability and validity of the measure in this group. The athlete food choice questionnaire (AFCQ) The AFCQ is a 32-item instrument rated on a 5-point Likert scale (1 = not at all important to 5 = extremely important), organized into nine conceptual domains; domain scores are obtained by summing items within each domain, with higher scores indicating greater importance of that motive in food selection. The original AFCQ [21] was translated into Arabic following forward–backward translation procedures [38], and semantic and conceptual equivalence was assessed by a five-member expert panel (nutrition scientists, sports psychologists) and tested by 10 Arabic speakers who exercised to ensure the clarification of terms and cultural appropriateness. The participants who were recruited in pilot psychometric validation (N = 100; 50% female; age M = 26.47 ± 4.29 years) through recruitment in fitness centers of Jordan proved to be good internal consistency (Cronbachs alpha = 0.91; Spearman-brown split-half = 0.88). A four-component output of the exploratory factor analysis including varimax rotation (KMO = 0.84; Bartlett test: 2185.32; 496 = 2185.32, p ≤ 0.001) obtained a four-component solution with a 72.13% variance (loads: 0.61–0.89), which are comparable to the previous validations on athletic populations [21]. This four-component framework summarizes the initial nine conceptual domains into factors that are obtained through empirical means without losing content coverage. Convergent validity was also deduced by moderate-strong correlations between related subscales (r = 0.62–0.78, p ≤ 0.001) and discriminant validity by weak-moderate correlations between different subscales (r = 0.22–0.35, p < 0.05). Mental health assessment The level of psychological distress was measured with the help of the Arabic-validated Depression, Anxiety, and Stress Scale-21 (DASS-21) that has been shown to have high psychometric characteristics in Arabic-speaking populations [41]. The Arabic version of the DASS-21 has a strong three-factor structure as compared to its original version with high internal consistency of depression (α = 0.88), anxiety (α = 0.82), and stress (α = 0.90) and overall scale (α = 0.93) subscales [41]. The scores on the subscales (response range: 0–42 post multiplication with 2) were classified based on the severity cut-offs (depression: normal (0–9), mild (10–13), moderate (14–20), severe (21–27), extremely severe (28 and above); anxiety: normal (0–7), mild (8–9), moderate (10–14), severe (15–19), extremely severe (19 and above); stress: normal (0–14), mild (15–18) moderate (19–25), severe (26–33), extremely severe (34 +). Statistical analysis Data were analyzed using SPSS version 25.0 (IBM Corp., Armonk, NY). Normality was assessed using the Shapiro–Wilk test [42]. Data are presented as means ± standard deviations (M ± SD) or frequencies (n) and percentages (%). Between-group comparisons (at-risk vs. not-at-risk for EA based on EAI ≥ 24 cutoff) employed independent-samples t-tests for continuous variables and chi-square tests for categorical variables, with Cohen's d and Cramér's V effect sizes reported respectively. However, backward stepwise methods carry risks of model instability and capitalization on chance, particularly with large predictor sets. Recent methodological advances demonstrate that machine learning approaches, including controlled feature selection, bootstrapping, random forests, and gradient boosting, achieve superior predictive performance and model stability compared to traditional stepwise methods [43]. In the most directly relevant exercise dependence prediction study to date, Zandonai et al. [43] applied machine learning with controlled feature selection and bootstrapping across 1099 participants recruited across three diverse samples and demonstrated that perfectionism, body image concerns, and sport level were robust cross-validated predictors of exercise dependence, with machine learning models achieving classification accuracy exceeding that of conventional logistic regression. While our approach served the current exploratory aims and enabled comparison with prior EA research employing similar methods, future studies should employ these advanced methodologies alongside cross-validation to enhance prediction accuracy, assess model stability across resampling iterations, and validate findings in independent samples. The convergence of traditional regression and machine learning approaches would strengthen confidence in identified risk factors [7, 10, 43]. Regression diagnostics included: (1) multicollinearity assessment via variance inflation factors (VIF < 5.0) and tolerance (> 0.20); (2) outlier detection using standardized residuals (± 3 SD) and Cook's distance (< 1.0), with influential cases retained given robust sample size; (3) linearity of continuous predictors with log-odds assessed via Box-Tidwell test. Model fit was evaluated using omnibus chi-square (p < 0.05), Hosmer–Lemeshow goodness-of-fit test (p > 0.05 indicating adequate fit), and Akaike Information Criterion (AIC) for model comparison across steps. Nagelkerke R2 quantified explained variance, and classification accuracy was reported. For exploratory analyses within the at-risk subsample only, associations among physical activity level (ordinal), psychological distress subscales (continuous), and AFCQ domain scores (continuous) were assessed using Spearman's rank-order correlations given non-normal distributions in this subgroup. All tests were two-tailed with α = 0.05. A p value < 0.05 was considered statistically significant throughout. Results Participants characteristics The analytic sample finalized (n = 1,036) included recreational athletes (Table 1). The participants were mainly male (57.3%), young (< 25 years: 45.9% 25–30 years: 30.1%), single (76.2%), and students (42.2%), or full-time workers (33.8%). Most participants held diploma/bachelor's degrees (77.6%), reported normal BMI (57.8%), and resided with family (86.7%). Monthly income distribution was relatively balanced across categories (< 300 JOD: 38.4%; 300–500 JOD: 33.3%; > 500 JOD: 28.3%). Approximately half (47.0%) reported health or sports-related fields of study/work, yet formal sports nutrition education (25.2%) and personal trainer supervision (24.3%) were uncommon. Smoking status was heterogeneous: non-smokers (47.3%), former smokers (15.9%), and current smokers (36.8%). Chronic disease prevalence was low (9.9%). Physical activity levels, assessed via IPAQ-S, were distributed as follows: low activity (n = 91, 8.8%), moderate activity (n = 412, 39.8%), and high activity (n = 533, 51.4%), indicating that over half the sample engaged in ≥ 3,000 MET-minutes/week—substantially exceeding WHO minimum guidelines. Prevalence of risk of exercise addiction overall and across the physical activity levels Nearly half of the sample (47.8%) crossed the EAI screening threshold for EA risk (Fig. 1). The EAI screening-positive rate increased systematically across IPAQ-S physical activity categories (Fig. 2; p ≤ 0.001): 22.0% among low-activity, 29.9% among moderate-activity, and 66.1% among high-activity participants, demonstrating a dose–response association between physical activity volume and EA risk. Fig. 1Exercise Addiction among physically active individuals (n = 1036) Exercise Addiction among physically active individuals (n = 1036) Fig. 2Exercise Addiction risk across the Physical Activity Level (n = 1,036) Exercise Addiction risk across the Physical Activity Level (n = 1,036) Psychological distress and risk of exercise addiction EA risk prevalence showed a statistically significant gradient across DASS-21 severity levels for all three dimensions (Figs. 3, 4 and 5). The sharpest gradient was observed for stress, where 66.7% of those with extremely severe stress screened at risk compared with 44.1% at the normal level (p = 0.003). Depression (p = 0.012) and anxiety (p = 0.005) showed qualitatively similar but numerically smaller gradients. Independent-samples t-tests comparing mean DASS-21 subscale scores between EA risk groups revealed significantly higher distress among at-risk individuals across all three dimensions: depression (at-risk M = 15.2 ± 9.8 vs. not-at-risk M = 12.8 ± 8.6; t(1034) = − 3.92, p ≤ 0.001, d = 0.26), anxiety (at-risk M = 12.6 ± 8.4 vs. not-at-risk M = 10.4 ± 7.8; t(1034) = − 4.18, p ≤ 0.001, d = 0.28), and stress (at-risk M = 18.9 ± 10.2 vs. not-at-risk M = 16.1 ± 9.4; t(1034) = − 4.45, p ≤ 0.001, d = 0.29). Effect sizes were small-to-medium, consistent with meta-analytic findings on EA-distress associations [11]. The cross-sectional design precludes causal inference regarding the temporal relationship between psychological distress and compulsive exercise. Fig. 3Exercise Addiction risk by Depression Level (n = 1036) Exercise Addiction risk by Depression Level (n = 1036) Fig. 4Exercise Addiction risk by Anxiety Level (n = 1036) Exercise Addiction risk by Anxiety Level (n = 1036) Fig. 5Exercise Addiction risk by Stress Level (n = 1036) Exercise Addiction risk by Stress Level (n = 1036) Food-choice motives and exercise addiction risk At-risk individuals demonstrated significantly higher scores across eight of nine AFCQ domains (Table 2), reflecting an instrumental eating orientation. The largest effects emerged for performance motives (d = 0.50), weight control (d = 0.44), and nutritional attributes (d = 0.45), with at-risk participants prioritizing performance optimization and body composition over sensory appeal. Medium effects were observed for social influences (d = 0.38), health awareness (d = 0.35), and habitual practices (d = 0.33). Sensory appeal showed no between-group difference (p = 0.108), indicating taste preferences remain stable regardless of EA risk status. Table 2Group Differences in Food Choice Motives by Exercise Addiction Risk Status (n = 1036)VariableNot at risk of EAAt risk of EAtEffect size (d)p valueNutritional attributes11.60 ± 4.3613.49 ± 4.02− 7.240.451 ≤ 0.001Emotional influences6.72 ± 3.947.45 ± 4.29− 2.850.1770.004Food and Health Awareness9.16 ± 3.8710.48 ± 3.56− 5.670.353 ≤ 0.001Influence of Others4.64 ± 2.785.74 ± 3.05− 6.110.38 ≤ 0.001Usual eating practices7.50 ± 2.838.37 ± 2.44− 5.260.327 ≤ 0.001Weight control8.75 ± 3.9610.45 ± 3.79− 7.040.438 ≤ 0.001Food Values and Beliefs5.79 ± 2.656.17 ± 2.71− 2.260.140.024Sensory appeal7.98 ± 3.138.28 ± 2.84− 1.610.10.108Performance6.91 ± 2.898.32 ± 2.78− 7.980.497 ≤ 0.001Independent samples t-tests were conducted to compare scale scores between Exercise Addiction Risk Status. Data presented as Mean ± SD. A p-value < 0.05 was considered statistically significant Table 2Group Differences in Food Choice Motives by Exercise Addiction Risk Status (n = 1036)VariableNot at risk of EAAt risk of EAtEffect size (d)p valueNutritional attributes11.60 ± 4.3613.49 ± 4.02− 7.240.451 ≤ 0.001Emotional influences6.72 ± 3.947.45 ± 4.29− 2.850.1770.004Food and Health Awareness9.16 ± 3.8710.48 ± 3.56− 5.670.353 ≤ 0.001Influence of Others4.64 ± 2.785.74 ± 3.05− 6.110.38 ≤ 0.001Usual eating practices7.50 ± 2.838.37 ± 2.44− 5.260.327 ≤ 0.001Weight control8.75 ± 3.9610.45 ± 3.79− 7.040.438 ≤ 0.001Food Values and Beliefs5.79 ± 2.656.17 ± 2.71− 2.260.140.024Sensory appeal7.98 ± 3.138.28 ± 2.84− 1.610.10.108Performance6.91 ± 2.898.32 ± 2.78− 7.980.497 ≤ 0.001Independent samples t-tests were conducted to compare scale scores between Exercise Addiction Risk Status. Data presented as Mean ± SD. A p-value < 0.05 was considered statistically significant Group Differences in Food Choice Motives by Exercise Addiction Risk Status (n = 1036) Independent samples t-tests were conducted to compare scale scores between Exercise Addiction Risk Status. Data presented as Mean ± SD. A p-value < 0.05 was considered statistically significant Sociodemographic variables and exercise addiction risk Sociodemographic associations with EA risk are presented in Table 3. Males demonstrated higher prevalence (52.5%) than females (41.4%; p ≤ 0.001). Those with health/sports field affiliation (51.1% vs. 44.8%; p = 0.042), sports nutrition education (56.7% vs. 44.8%; p ≤ 0.001), and personal trainer supervision (54.0% vs. 45.8%; p = 0.024) showed elevated risk. Living alone was associated with lower risk (33.3% vs. 50.0% for family co-residence; p ≤ 0.001), as was chronic disease status (33.0% vs. 49.4%; p = 0.002). BMI and monthly income showed no significant associations. Table 3Association Between Sociodemographic Variables and Risk of Exercise Addiction Among Physically Active Individuals (n = 1036)VariableCategoriesNot at risk of EAAt risk of EAχ2p valueGenderMale282 (52.1%)312 (63.0%)12.565 ≤ 0.001*Female259 (47.9%)183 (37.0%)Age (years) < 25224 (41.9%)252 (50.9%)9.4570.009*25–30178 (32.9%)134 (27.1%) > 30139 (25.7%)109 (22.0%)Marital statusSingle394 (72.8%)395 (79.8%)7.1400.028*Married131 (24.2%)91 (18.4%)Other16 (3.0%)9 (1.8%)Job statusNot working59 (10.9%)41 (8.3%)10.0860.039*Student206 (38.1%)231 (46.7%)Part-time65 (12.0%)60 (12.1%)Full-time195 (36.0%)155 (31.3%)Retired16 (3.0%)8 (1.6%)Field relevance (health/sport)No303 (56.0%)246 (50.3%)4.1320.042*Yes238 (44.0%)249 (49.7%)Course in Sports NutritionNo428 (79.1%)347 (70.1%)11.1390.001*Yes113 (20.9%)148 (29.9%)Personal trainerNo425 (78.4%)359 (72.5%)5.1110.024*Yes116 (21.6%)136 (27.5%)Chronic diseaseNo472 (87.2%)461 (93.1%)10.0000.002*Yes69 (12.8%)34 (6.9%)Education level ≤ High School52 (6.9%)73 (14.7%)7.7400.021*Diploma/Bachelor426 (78.7%)378 (76.4%)Postgraduate63 (11.6%)44 (8.9%)Smoking statusNon-smoker236 (43.6%)254 (51.3%)6.2490.044*Former smoker94 (17.4%)71 (14.3%)Current smoker211 (39.0%)170 (34.3%)HousingAlone92 (17.0%)46 (9.3%)13.317 ≤ 0.001*With family449 (83.0%)449 (90.7%)Monthly income (JD)Less than 300207 (38.3%)191 (38.6%)0.7460.689300–500186 (34.4%)159 (32.1%)More than 500148 (27.4%)145 (29.3%)BMIUnderweight17 (3.1%)20 (4.0%)2.9510.399Normal326 (60.3%)273 (55.2%)Overweight151 (27.9%)153 (30.9%)Obese47 (8.7%)49 (9.9%)Chi-square test was performed to examine associations between categorical variables. A p-value ≤ 0.05 was considered statistically significant Table 3Association Between Sociodemographic Variables and Risk of Exercise Addiction Among Physically Active Individuals (n = 1036)VariableCategoriesNot at risk of EAAt risk of EAχ2p valueGenderMale282 (52.1%)312 (63.0%)12.565 ≤ 0.001*Female259 (47.9%)183 (37.0%)Age (years) < 25224 (41.9%)252 (50.9%)9.4570.009*25–30178 (32.9%)134 (27.1%) > 30139 (25.7%)109 (22.0%)Marital statusSingle394 (72.8%)395 (79.8%)7.1400.028*Married131 (24.2%)91 (18.4%)Other16 (3.0%)9 (1.8%)Job statusNot working59 (10.9%)41 (8.3%)10.0860.039*Student206 (38.1%)231 (46.7%)Part-time65 (12.0%)60 (12.1%)Full-time195 (36.0%)155 (31.3%)Retired16 (3.0%)8 (1.6%)Field relevance (health/sport)No303 (56.0%)246 (50.3%)4.1320.042*Yes238 (44.0%)249 (49.7%)Course in Sports NutritionNo428 (79.1%)347 (70.1%)11.1390.001*Yes113 (20.9%)148 (29.9%)Personal trainerNo425 (78.4%)359 (72.5%)5.1110.024*Yes116 (21.6%)136 (27.5%)Chronic diseaseNo472 (87.2%)461 (93.1%)10.0000.002*Yes69 (12.8%)34 (6.9%)Education level ≤ High School52 (6.9%)73 (14.7%)7.7400.021*Diploma/Bachelor426 (78.7%)378 (76.4%)Postgraduate63 (11.6%)44 (8.9%)Smoking statusNon-smoker236 (43.6%)254 (51.3%)6.2490.044*Former smoker94 (17.4%)71 (14.3%)Current smoker211 (39.0%)170 (34.3%)HousingAlone92 (17.0%)46 (9.3%)13.317 ≤ 0.001*With family449 (83.0%)449 (90.7%)Monthly income (JD)Less than 300207 (38.3%)191 (38.6%)0.7460.689300–500186 (34.4%)159 (32.1%)More than 500148 (27.4%)145 (29.3%)BMIUnderweight17 (3.1%)20 (4.0%)2.9510.399Normal326 (60.3%)273 (55.2%)Overweight151 (27.9%)153 (30.9%)Obese47 (8.7%)49 (9.9%)Chi-square test was performed to examine associations between categorical variables. A p-value ≤ 0.05 was considered statistically significant Association Between Sociodemographic Variables and Risk of Exercise Addiction Among Physically Active Individuals (n = 1036) Chi-square test was performed to examine associations between categorical variables. A p-value ≤ 0.05 was considered statistically significant Predictors of exercise addiction risk Binary logistic regression with backward stepwise elimination (Wald criterion, p-removal = 0.10) identified independent predictors of EA risk status from 27 candidate variables selected a priori based on theoretical relevance [7, 10]. The final model retained 12 predictors and demonstrated adequate fit: Nagelkerke R2 = 0.336; Hosmer–Lemeshow χ2(8) = 2.35, p = 0.969; classification accuracy = 72.9%. Multicollinearity diagnostics were acceptable (all VIF < 2.5). Although backward stepwise elimination is established in EA screening research, machine learning approaches (elastic net, random forests) may offer superior stability and predictive performance [7]; future studies should employ these methods with cross-validation as robustness checks. Table 4 presents the multivariable logistic regression results. Male gender (OR = 1.66, 95% CI [1.21, 2.27]), lower education (high school or less: OR = 2.03 vs. postgraduate), and living with family (OR = 0.60 for living alone) showed significant associations. Physical activity demonstrated a strong dose–response pattern, with low (OR = 0.13) and moderate (OR = 0.20) activity associated with substantially reduced odds versus high activity reference. Stress severity showed graded associations, with lower stress levels protective relative to extremely severe stress; the Normal (OR = 0.29, p = 0.004), Mild (OR = 0.35, p = 0.022), and Moderate (OR = 0.41, p = 0.049) stress categories each reached conventional significance (α = 0.05), whereas the Severe stress category (OR = 0.47, p = 0.100) did not reach this threshold but was retained in the model under the pre-specified p-removal = 0.10 criterion. Table 4Predictors of Exercise Addiction Risk Among Physically Active Individuals (n = 1036)PredictorBSEp valueOR (Exp(B))95% CI Lower95% CI UpperGender (male)0.5060.1610.0021.6591.212.274Female Reference Education level ≤ High school0.710.3210.0272.0331.0843.813Diploma/bachelor0.210.2460.3941.2330.7611.998Postgraduate Reference Housing (alone)-0.5090.2310.0270.6010.3820.944With family Reference Physical activityLow− 2.0470.26 ≤ 0.0010.1290.0780.215Moderate− 1.5980.159 ≤ 0.0010.2020.1480.276High Reference Stress levelNormal− 1.230.430.0040.2920.1260.68Mild− 1.0380.4530.0220.3540.1460.86Moderate− 0.8860.450.0490.4120.1710.995Severe− 0.760.4610.1000.4680.1891.156Extremely severe Reference Food choice motivesNutritional attributes0.0660.0230.0041.0681.0211.117Emotional influences0.0390.0210.0581.040.9991.083Influence of Others0.10.028 ≤ 0.0011.1051.0471.166Weight control0.0640.0250.0091.0661.0161.118Food values/beliefs− 0.0640.0310.0380.9380.8820.997Sensory appeal− 0.0780.0310.0130.9250.870.983Performance0.1120.0350.0011.1191.0441.198Variable(s) entered on step 1: Age, gender, BMI, marital status, Job, Field Relevance Health or Sports Connection, courses to sports nutrition, personal trainer, chronic diseases, education level, smoking, status of living, monthly income, level of education, physical activity level, levels of depression, levels of Anxiety, levels of stress, all Athlete Food Choice dimensions; B = logistic regression coefficient (log odds); SE = standard error of the coefficient; p = p-value for significance testing; OR = odds ratio (Exp(B)); 95% CI = 95% confidence interval for the odds ratio. A p-value ≤ 0.05 is considered statistically significant Table 4Predictors of Exercise Addiction Risk Among Physically Active Individuals (n = 1036)PredictorBSEp valueOR (Exp(B))95% CI Lower95% CI UpperGender (male)0.5060.1610.0021.6591.212.274Female Reference Education level ≤ High school0.710.3210.0272.0331.0843.813Diploma/bachelor0.210.2460.3941.2330.7611.998Postgraduate Reference Housing (alone)-0.5090.2310.0270.6010.3820.944With family Reference Physical activityLow− 2.0470.26 ≤ 0.0010.1290.0780.215Moderate− 1.5980.159 ≤ 0.0010.2020.1480.276High Reference Stress levelNormal− 1.230.430.0040.2920.1260.68Mild− 1.0380.4530.0220.3540.1460.86Moderate− 0.8860.450.0490.4120.1710.995Severe− 0.760.4610.1000.4680.1891.156Extremely severe Reference Food choice motivesNutritional attributes0.0660.0230.0041.0681.0211.117Emotional influences0.0390.0210.0581.040.9991.083Influence of Others0.10.028 ≤ 0.0011.1051.0471.166Weight control0.0640.0250.0091.0661.0161.118Food values/beliefs− 0.0640.0310.0380.9380.8820.997Sensory appeal− 0.0780.0310.0130.9250.870.983Performance0.1120.0350.0011.1191.0441.198Variable(s) entered on step 1: Age, gender, BMI, marital status, Job, Field Relevance Health or Sports Connection, courses to sports nutrition, personal trainer, chronic diseases, education level, smoking, status of living, monthly income, level of education, physical activity level, levels of depression, levels of Anxiety, levels of stress, all Athlete Food Choice dimensions; B = logistic regression coefficient (log odds); SE = standard error of the coefficient; p = p-value for significance testing; OR = odds ratio (Exp(B)); 95% CI = 95% confidence interval for the odds ratio. A p-value ≤ 0.05 is considered statistically significant Predictors of Exercise Addiction Risk Among Physically Active Individuals (n = 1036) Variable(s) entered on step 1: Age, gender, BMI, marital status, Job, Field Relevance Health or Sports Connection, courses to sports nutrition, personal trainer, chronic diseases, education level, smoking, status of living, monthly income, level of education, physical activity level, levels of depression, levels of Anxiety, levels of stress, all Athlete Food Choice dimensions; B = logistic regression coefficient (log odds); SE = standard error of the coefficient; p = p-value for significance testing; OR = odds ratio (Exp(B)); 95% CI = 95% confidence interval for the odds ratio. A p-value ≤ 0.05 is considered statistically significant Discussion In this fitness-oriented Jordanian recreational-athlete sample, 47.8% crossed the EAI screening threshold for EA risk, with the screening-positive rate increasing systematically across physical activity levels. At-risk individuals exhibited elevated psychological distress and distinctive food preferences emphasizing performance and weight control over sensory appeal. Male gender, high training volume, extreme stress, and performance-oriented eating emerged as independent risk factors, while living alone and hedonic food motivations showed protective associations. These patterns support considering EA risk within a broader athlete mental health continuum across recreational and elite sport contexts, while recognizing that screening-positive status does not establish clinical diagnosis. EA has significant phenomenological similarity to both overtraining syndrome (OTS) and athlete burnout—identified as mental health issues in competitive sport—in that it features excessive training even when tired or injured, mood disturbance, anxiety, depression, and unhealthy eating habits [13, 15]. The existing results, in which 66.7% of people with extremely severe stress and 66.1% of high-volume exercisers tested at the EA risk, are consistent with the hypothesis that EA, OTS, and burnout may represent manifestations of a shared syndrome of training maladaptation with psychological distress. Research on overtraining syndrome and burnout has been conducted almost entirely on elite athletes; the present study demonstrates that recreational athletes also carry substantial EA risk. For instrument-consistent comparison, Lichtenstein et al. [9] reported a 7.6% EAI-defined screening-positive rate among nationally supervised elite athletes across 15 sport disciplines—a population structurally distinct from the self-selected, unmonitored recreational fitness community studied here. This contrast demonstrates that EA risk on the EAI is not confined to elite athletes and warrants attention across all levels of competition [4]. Prevalence and demographic correlates of exercise addiction risk For context, and consistent with the established measurement principle that EA prevalence estimates are only meaningfully comparable when the same instrument and cut-off are applied [4, 44, 45], the present findings are interpreted against EAI-based benchmarks exclusively. The most directly comparable EAI study reported a 7.6% screening-positive rate among nationally supervised elite athletes across 15 sport disciplines [9], a population that is structurally distinct from self-selected recreational fitness community members. The observed 47.8% screening-positive rate requires cautious interpretation in light of the study’s recruitment frame and the non-diagnostic nature of the EAI. Three interrelated factors account for this figure. First, recruitment through fitness facilities and online fitness communities concentrated individuals already engaged in high-volume exercise (51.4% classified as high activity by IPAQ-S), creating a sampling frame enriched for EAI-defined risk phenotypes. Second, the EAI is a sensitivity-optimised screening instrument; in the absence of specificity validation in the Jordanian context, the screening-positive rate should be interpreted as an upper-bound indicator rather than a true diagnostic prevalence [3, 4]. Third, the EAI captures risk indicators rather than clinical diagnosis; accordingly, this figure reflects the proportion meeting a screening threshold, not the proportion with confirmed exercise addiction. When benchmarked against the most directly comparable EAI-based study, the 7.6% screening-positive rate reported among nationally supervised elite athletes [9] is substantially lower than the rate observed here—a difference structurally explained by a self-selected, unmonitored fitness community sample with a substantially higher proportion of high-volume exercisers. The 47.8% figure should therefore be interpreted strictly as a sample-specific EAI screening-positive rate in a fitness-oriented Jordanian recreational-athlete group, not as a diagnostic prevalence estimate and not as a population-level prevalence for Jordanian recreational athletes. The EAI functions as a sensitivity-optimised screening tool whose purpose is to identify individuals warranting further clinical assessment, not to confirm clinical diagnosis. The proportion of participants crossing the EAI threshold therefore reflects the instrument's broad capture of at-risk psychological indicators combined with the characteristics of the sampling frame, rather than the true prevalence of clinically significant exercise addiction in the Jordanian population. Clinical confirmation through structured interview would be required to establish diagnostic incidence [3, 4]. Male gender was also a notable demographic predictor (OR = 1.66), which aligns with some, but not all previous studies that indicated that men were at greater risk with regard to EA during recreational exercise [46]. The meta-analyses of the gender distinctions are inconclusive with the effects of gender differing depending on the cultural setting, exercise conditions, and measurement systems [17]. The masculine dominance in this case could be due to gender-specific exercise motivation in the Middle East fitness culture, where men might be more focused on muscle building and performance improvement and have to encounter challenges of sociocultural norms on body image and masculine values [47, 48]. In univariate analyses, younger age (< 25 years) was found to have higher prevalence of EA risks and this is consistent with the findings that compulsive exercise behaviors are clustered in student and young adult groups who might apply exercise as a means of identity formation, peer social comparison, and stress management in transitional life periods [4]. Nevertheless, age failed to be significant in adjusted models and this implied that its relationship functions via mediating factors like student status, living conditions, or stress levels. The non-linear correlation of education and EA risk- with the greatest risk among those with 0-high school education (OR = 2.03 vs. postgraduate), yet no difference between bachelor's and postgraduate levels- indicates that the educational attainment might be as a proxy of health literacy, critical thinking ability concerning exercise message or socioeconomic considerations of seeking professional advice [46]. With less education, people can be less exposed to the evidence-based information about exercise, and they are more susceptible to marketing messages in the fitness industry that promote overtraining or equate the amount of exercise with improvements in health. The fact that living alone (OR = 0.60 vs. living with family) is unexpectedly protective is in contrast to the social support theories but may be culturally appropriate given that 86.7% of the sample lived with family in Jordan: living in co-residence may augment family pressure on appearance, performance, or compliance with fitness routines, whereas living alone may allow independent regulation of exercise behaviors [49]. Alternatively, individuals living on their own can be a self-selected group that is more self-sufficient and less in need of external validation by the success of exercises. The missing BMI associations in the adjusted models are consistent with identical research of elite athletes indicating that no body mass is correlated with the EAI score, and instead, psychological and motivational factors, instead of anthropometric status, better explain EA risk variance in this group [9]. Physical activity level and exercise addiction risk The volume of physical activity proved to be the most strongly associated variable with EAI-defined risk and showed a definite dose–response trend across the IPAQ-S categories: 22% screening-positive among low-activity individuals, 30% among moderate-activity, and 66% among high-activity participants (adjusted ORs: low = 0.13, moderate = 0.20 vs. high reference). Such a graded relationship is theoretically expected, with increased exercise exposure offering more opportunity for compulsive patterns to develop, but physical activity volume is neither necessary nor sufficient for EA, as demonstrated in elite athlete studies reporting no association between training hours and EAI scores [6, 9]. The 22.0% screening-positive rate among low-activity participants is best understood through a clear distinction between the two measurement instruments. The EAI measures compulsive exercise-related attitudes and psychological symptoms, specifically salience, mood modification, tolerance, withdrawal, conflict, and relapse [5]; it does not measure physical activity volume. The IPAQ-S, by contrast, captures activity during the preceding seven days only. A participant who is habitually a recreational exerciser but whose training was reduced during the assessment week due to normal variation in training load may score low on IPAQ-S while simultaneously scoring high on EAI items, particularly the withdrawal and conflict subscales, which are activated precisely when exercise is reduced or interrupted. This discrepancy suggests that EAI-defined risk in this subgroup captures psychological preoccupation with exercise rather than evidence of high current volume. This dissociation between behavioral expression and psychological compulsivity is well established in behavioral addiction research [10] and in EA-specific literature confirming that exercise volume and psychological dependency are separable constructs [6, 9]. Furthermore, as eligibility required habitual exercise participation at screening rather than during the IPAQ-S recall window, some divergence between study entry status and IPAQ-S classification was structurally possible. This subgroup should therefore be interpreted cautiously. The cross-sectional design cannot establish cause-and-effect: high exercise volume may be associated with EAI-defined risk through mechanisms including habituation and tolerance, or risk characteristics such as obsessive thoughts, compulsions, and guilt over missed sessions may independently drive increased volume, or both processes may operate simultaneously [50]. Chronic high-volume exercise combined with insufficient energy intake-termed Relative Energy Deficiency in Sport (RED-S) is associated with hormonal responses including leptin, triiodothyronine (T3), and reproductive hormone suppression, which may contribute to physiological feedback processes hypothesized to maintain compulsive exercise [14].The convergence of high training volume (66.1% of high-activity participants at EA risk), extreme psychological stress (66.7% at risk), and instrumental eating orientations is consistent with RED-S vulnerability. RED-S may develop when inadequate energy availability—from insufficient intake or excessive expenditure—is associated with neuroendocrine adaptations (leptin and T3 suppression) that have been linked to food preoccupation and psychological distress, potentially contributing to compensatory exercise increases in a self-reinforcing cycle [14, 51]. In contrast to elite athletes who have well-organized nutrition assistance, recreational athletes who combine large training loads with eating restriction can be on high risk of developing RED-S [52]. The EAI items assessing tolerance, withdrawal, and conflict overlap substantially with RED-S warning signs, suggesting EA screening may function as indirect RED-S surveillance in populations lacking sports medicine monitoring. Future research should integrate EA screening with RED-S-specific assessments to validate this approach [51, 52]. The EA-associated physiological phenotype of the high-volume recreational athletes at risk, i.e. the occurrence of high level of physiological stress, inefficient recovery, and an adherent diet policy is a manifestation of the Clinical Sports Medicine model of overtraining syndrome, when the training workload exceeds the capacity of recovery, resulting in neuroendocrine dysfunction and psychological manifestations [10, 15, 17]. Third, excessive exercise can be normalized and peer pressure to continue or even increase excessive training can be enforced by social reinforcement in fitness communities where high training volume is glorified and recovery is stigmatized as a sign of weakness [6]. The EA risk concentration of active participants highlights the necessity of having fitness professionals apply routine screening, as well as educate the clients that more does not always equate to better with respect to dose–response relationships of exercising. Psychological distress and exercise addiction risk In univariate analyses, all three DASS-21 dimensions (depression, anxiety, stress) were significantly related to EA risk, and the strongest gradients were between stress severity. Stress was also found to maintain significance as an independent predictor in multivariable models, whereas depression and anxiety did not, indicating a high level of overlap between these phenomena, or that stress is a more generalized measure of psychological dysregulation that involves particular symptom domains. Those who reported having extremely severe stress levels reported the highest EA risk (66.7%), and a monotonically decreasing prevalence with lower levels of stress (adjusted ORs 0.29–0.47 vs. extremely severe reference). These results are consistent with meta-analytic evidence that demonstrates strong relations between the risks of EA and psychological distress, and the pooled effect size ranges in the medium range [10, 11]. Several theoretical models have been put forward to describe the co-occurrence of EA-distress, but the cross-sectional nature of data is unable to provide directional associations. The affect regulation hypothesis suggests that people can engage in excessive exercise as a maladaptive coping behavior in response to the underlying condition of depression, anxiety, or stress, they use the mood enhancement and distraction of negative thoughts as a short-term source of relief which strengthens the compulsive behavior in the long term [50, 53]. On the other hand, the addiction consequence model implies that compulsive exercise patterns may contribute to the development of psychological stress: disruption of life in the family/friends in connection with spending time on exercise, feeling guilty and anxious when not exercising due to an injury or commitments, social isolation because of non-exercise activities, physical fatigue because of insufficient recovery [10, 54]. Neurobiological hypotheses propose that dysregulation of the hypothalamic–pituitary–adrenal (HPA) axis, particularly when chronic exercise is combined with energy deficit, may be associated with alterations in cortisol patterns, inflammatory responses, and neurotransmitter balance that are linked to symptoms of depression and anxiety [55]. Beyond neuroendocrine pathways, neurobiological models of behavioral addiction implicate reward and motivational circuitry dysregulation in maintaining compulsive exercise. Repeated high-intensity exercise may alter mesolimbic dopaminergic signaling, creating tolerance effects requiring increasing exercise volume for equivalent mood enhancement—consistent with the EAI tolerance criterion [10]. Paralleling substance addictions, at-risk individuals may exhibit enhanced reward sensitivity to exercise cues coupled with diminished responsiveness to non-exercise activities, narrowing the behavioral repertoire [8]. These alterations may interact with stress mechanisms: chronic stress increases exercise incentive salience while impairing prefrontal inhibitory control, creating a neurobiological substrate for compulsivity. These hypotheses require validation through neuroimaging and longitudinal studies establishing temporal precedence and neurobiological correlates distinct from stress reactivity [10]. Lastly, bi-directional models hypothesize reciprocal influences: distress may drive excessive exercise, which provides temporary symptom relief but may promote dependency, potentially leading to increased volume, insufficient recovery, physical/social effects and worsening distress in a reinforcing cycle typical of behavioral addictions [4]. The existing evidence does not allow any of these theoretical explanations to be ruled out; however, the association pattern—in which extremely severe stress is most strongly linked to EA risk—suggests that perceived stress may function as a transdiagnostic vulnerability marker capturing broader psychological dysregulation beyond the specific symptom domains of depression and anxiety. Elevated chronic stress is associated with activation of the HPA axis and autonomic nervous system responses that may facilitate exercise-seeking as a regulatory behavior, while reward-system alterations characteristic of behavioural addictions may independently maintain compulsive exercise patterns [10]. These neurobiological pathways are hypothetical in the present cross-sectional context and require longitudinal and neuroimaging validation. Clinically, the results highlight the need to fully screen the mental health of individuals at EA risk when exposures are detected because psychological distress may be a maintaining factor that needs to be interventional or a response to the exercising behaviors that need to be modified either or both. These results support the need to conduct mental health screening regularly among all athlete groups, not only among elite athletes, and acknowledge that psychological distress, compulsive behaviors, and maladaptive coping styles (such as excessive exercise and restrictive eating) are a continuum across the recreational and elite athlete groups and must be treated with athlete-specific mental health models and not general models [56]. Food choice motives and exercise addiction risk: an instrumental eating profile At-risk individuals exhibited a unique food choice pattern which was instrumental (goal-oriented) and not hedonic (based on pleasure) in motivation of eating. The greatest between-group differences appeared in relation to performance related motives (d = 0.50), weight control motives (d = 0.44), and nutritional qualities (d = 0.45) all areas of utilitarian food choice in the favour of the maximisation of body composition, training results, or health indicators as opposed to sensory pleasures or sociocultural meaning of food. This instrumental orientation was also found in adjusted models, where performance motives (OR = 1.12 per unit increase), weight control (OR = 1.07), nutritional attributes (OR = 1.07) and social influences (OR = 1.11) independently predicted EA risk, and sensory appeal (OR = 0.93) and food values/beliefs (OR = 0.94) were found to have an inverse relationship. These tendencies are consistent with new findings that EA risk is associated with orthorexic eating patterns, or obsession with food quality, strict dietary norms, and fear of contaminated foods that overlap conceptually and empirically with EA in spite of the fact that they are two different constructs [17, 18]. Both the EA and the orthorexia nervosa include compulsive health-related behavior (excessive exercise vs. the restrictive healthy eating), perfectionistic cognitions, body dissatisfaction, and the impaired quality of life when the behaviors disrupt the social relationships or the psychological well-being [18]. The high level of nutritional values and health awareness of at-risk people in the present sample are indicative of health-purist orientations in which food becomes instrumentalized as fuel or medicine instead of a source of pleasure, social bond or cultural identity [7, 16]. It should be noted, however, that the entire food choice profile documented in Sect. 4.4—including elevated nutritional attributes, health awareness, habitual practices, and social influences, alongside reduced sensory appeal—is equally consistent with secondary EA driven by eating disorder pathology as it is with primary EA characterized by independent exercise compulsion; without concurrent administration of a validated eating disorder instrument such as the EDE-Q, no claim can be made that these food choice patterns are specific to exercise addiction rather than manifestations of comorbid disordered eating, and all interpretations in this section should be read with that interpretive boundary in mind. When combined with compulsive exercise, such dogmatic beliefs about health can represent a more general syndrome of wellness-oriented behavioral dysregulation that can cross the boundaries of conventional eating disorder typologies [17]. Elevated weight control motives among at-risk participants warrant cautious interpretation. The EA literature distinguishes primary EA (independent compulsive exercise) from secondary EA (exercise as compensatory behavior within eating disorder pathology) [4, 6]. Without eating disorder screening (e.g., Eating Disorder Examination Questionnaire (EDE-Q)), the present study cannot determine which pattern is reflected. The convergence of high weight control and nutritional attribute scores with low sensory appeal is consistent with both possibilities and should be interpreted as a potential marker of disordered eating rather than a confirmed EA-specific characteristic. Future studies incorporating eating disorder co-assessment are essential to operationalize this distinction in Arabic-speaking populations [6]. The social influence result (OR = 1.11) is the significant contribution of the peer norms, coach/trainer expectations as well as family pressures to influence exercise and eating behaviors among the communities that engage in fitness activities. In people who co-reside with their families (86.7% of the sample)- as is customary in Jordan- family member remarks on appearance, performance or what a person eats may support instrumental eating orientations, especially when family members themselves support fitness/health values or when they make social comparisons [49]. Likewise, supervision by personal trainers (24.3% of sample) can be unintentionally associated with the promotion of severe nutrient adherence in that these trainers focus on nutrient timing, macronutrient optimization, or clean eating ideologies without properly focusing on psychological flexibility and eating disorder risk factors. These results indicate that EA prevention and treatment initiatives cannot be limited to education at the individual level of analysis and should consider social and environmental components in the fitness environment that normalizes or supports extreme exercise and inflexible eating habits. Multivariable predictors: toward an integrated biopsychosocial model The final logistic regression model combined 12 predictors across demographic, behavioral, psychological, and motivational dimensions, achieving a classification accuracy of 72.9% and explaining 33.6% of the variance in EAI-defined risk within this sample. Although backward stepwise elimination was applied here because of its prior use in EA screening research, the present regression model should be interpreted as exploratory, identifying sample-specific candidate associations rather than definitive predictors, and requiring replication in independent datasets. This complex profile highlights that EA risk is not inherent to any particular factor but rather an outcome of intricate interactions between biological tendencies (possible including genetic susceptibility to behavioral addictions, not measured in this case), psychological factors (stress, perfectionistic cognitions within performance/weight motives), social factors (family co-residence, peer influence, trainer supervision), and behavioral patterns (large exercise volumes, instrumental eating) [4]. Some of the findings are worthy of integration in theory. To begin with, the intersection of high levels of physical activity, extremely intense stress, and performance/weight-obsessional eating implies a phenotype of "compulsive exercisers" who use intense training and dietary optimization in the management of psychological distress, the pursuit of appearance/performance ideals, or in imposing control during the stressful periods of life but which simply increases but does not resolve psychological distress as the exercise and eating practices become more and more inflexible and disruptive to life [50]. This profile is particularly relevant among young male students (the modal population) experiencing academic pressures, career uncertainty, and sociocultural pressures related to masculine success and physical strength [47]. Backward stepwise elimination was selected because it is well-established in EA screening research and permits transparent identification of independent predictors from theoretically justified candidates [7, 10]. This model is best understood as a preliminary explanatory analysis of EAI-defined risk in this sample, and the identified associations require replication and external validation before being treated as stable predictors. This approach carries risks of model instability and capitalization on chance with large predictor sets. Multicollinearity diagnostics confirmed acceptable collinearity (VIF < 5.0) and the Hosmer–Lemeshow test indicated adequate fit, yet machine learning approaches, including controlled feature selection and bootstrapping, achieve superior predictive performance and model stability, as demonstrated in a directly comparable exercise dependence prediction study [43]. Future research should employ these advanced methodologies alongside cross-validation to confirm the stability of the associations identified here and to validate findings across independent samples. Second, the opposite phenotype of exercise could be discovered in the inverse effects of living alone, sensory appeal and food values/beliefs that determine a consistent exercise routine in wider contexts of life including independent decision making, pleasure eating and cultural food practices-protective mechanisms, which might counter the compulsive pattern even at high levels of exercise [16]. Third, results that show no BMI impacts in adjusted model, even though trends based on univariate analyses would suggest higher risk of EA among overweight/obese participants, point to no effects of body mass per se on EA risks, given that psychological (stress) and motivational (performance/weight motives) factors result in these trends, are shown by psychometric studies that point to cognitive-affective processes rather than anthropometric status in behavioral addictions [9]. The people who support this constellation especially young men who are in high volume training should be further assessed and may be referred to sport psychology or mental health professionals to undergo a thorough assessment. On the other hand, the hedonic eating motivations as protective relationships indicate that an intervention which fosters food flexibility, cultural foods, and pleasure eating (as opposed to nutrition) can prevent [32] or mitigate EA risks by overriding compulsive eating orientations that can strengthen compulsive exercise habits [16]. Conclusion In this cross-sectional study of 1,036 Jordanian recreational athletes, 47.8% crossed the EAI screening threshold for EA risk; this figure should be interpreted as a sample-specific screening-positive rate shaped by fitness-oriented recruitment, the non-diagnostic nature of the EAI, and the high proportion of highly active participants, rather than as a population prevalence estimate. Through multivariate modeling, it was found that there was a multifactorial risk profile which includes male gender, large volumes of physical activity (dose–response pattern), very severe stress, and instrumental food choice orientations with a prioritization to performance enhancement, weight management, nutritional optimization and social factors. On the other hand, there was a protective relationship between living alone, a reduced severity of stress and hedonic food motivation (sensory appeal, cultural food values). The relationship of BMI and EA risk was found to be non-existent after the psychological and motivational variables were considered. The same patterns are related to behavioral addiction models that focus on affective dysregulation, perfectionistic cognitions, and strict health-oriented behaviors, but cross-sectional data do not allow drawing causal conclusions about directional interrelations between distress, food choice motivations, and compulsive exercise. The paper presents the first combined evaluation of EA risk and such comprehensive food choice motivation and psychological distress in an Arabic-speaking population of athletes, fulfilling critical gaps of evidence in athlete mental health research and supporting the case for culturally sensitive screening models applicable from the recreational to the elite level, multidisciplinary coordination of sport and nutritional psychological services, and longitudinal trials examining EA within the overtraining-burnout-mental health syndrome continuum in non-Western athletic populations. Practical implications Within the boundary of fitness-oriented, self-selected recreational athlete populations comparable to the present sample, findings support integrating tiered EA risk screening into athlete mental health protocols, targeting young male recreational athletes engaged in high-volume training (≥ 3,000 MET-min/week); these recommendations should not be extrapolated to the general Jordanian population, for whom representative prevalence data and population-based screening thresholds remain to be established. Fitness facilities should integrate brief, validated screening tools (Arabic EAI, DASS-21 stress subscale, modified AFCQ focusing on performance/weight motives) administered at intake and semi-annually, with positive screens triggering referral pathways to mental health professionals—specifically sport psychologists trained in exercise and eating behavior disorders—who can conduct comprehensive clinical assessment and differentiate primary from secondary EA presentations. The multidisciplinary care teams comprising of nutritionists and coaching staffs should have sports psychologists as the coordinating professionals to coordinate case management and treatment planning [6]. The cultural tailoring in nutrition interventions must utilize the respect of the Jordanian culture in relation to food tradition and the instrumental eating habits that were defined among the at-risk population. Some evidence-based interventions are: (1) Relative Energy Deficiency in Sport (RED-S) education based on the importance of sufficient energy supply in meeting training needs; (2) cognitive-behavioral interventions focused on inflexible dietary regulations and clean eating beliefs; (3) the encouragement of flexible eating in which traditional Jordanian foods (e.g., mansaf, mezze) are eaten instead of prescriptive performance-focused nutritional guidelines; and (4) screening of orthorexic tendencies, through integrated care with mental health practitioners [14, 16]. Fitness coaches and personal trainers require capacity building through continuing education modules covering: EA risk indicators (training despite injury, distress when unable to exercise, social/occupational impairment); appropriate periodization incorporating recovery; and referral protocols to sport psychology services. Given that 75.7% of the current sample lacked personal trainer supervision, broader public health messaging through university wellness programs, fitness facility orientations, and social media campaigns may reach at-risk individuals without professional guidance. Family psychoeducation addressing appearance-focused comments, performance pressure, and recognition of concerning exercise patterns is warranted given the 86.7% family co-residence rate and documented family functioning correlates of EA in collectivist cultures [49]. Strengths and limitations This investigation provides the first simultaneous examination of EA risk, psychological distress, and multidimensional food choice motives in an Arabic-speaking population, employing culturally adapted, psychometrically validated instruments (EAI, DASS-21, AFCQ) with a large sample (N = 1,036) exceeding a priori statistical power requirements for multivariable logistic regression. The multidimensional assessment framework integrating demographic, behavioral, psychological, and motivational domains enabled identification of a complex risk profile not discernible through single-domain approaches. Translation and cultural adaptation procedures followed established guidelines [38], with pilot validation samples (EAI: N = 30,AFCQ: N = 100) and post-hoc internal consistency confirmation in the full sample supporting measure reliability in this population. Strict statistical techniques such as extensive regression diagnostic, reporting of effect size, and acknowledgment of cross-sectional limitations contribute to transparency and replicability. There are a number of limitations that should be considered. First, non-probability sampling (convenience, snowball) creates a selection bias where recruiting people who are interested in fitness-related environments would overrepresent people with high exercise participation and potentially exaggerate the prevalence of EA risks compared to population-based estimates. It is thus limited to generalizability to populations that are concerned with fitness and not the general populations of Jordan. Second, all constructs were assessed by self-report without objective verification. Notably, the IPAQ-S systematically overestimates physical activity relative to accelerometry. Studies report overestimation of 36–173% with a mean of 84% [57], 165% [58], 161 min/week [59], and 128% [60]. Consequently, the classification of 51.4% of participants as highly active (≥ 3,000 MET-min/week))— and the resulting dose–response relationship with EAI-defined risk may be artificially inflated. Furthermore, because eligibility reflected participants' habitual exercise status at recruitment, whereas the IPAQ-S captured only the preceding seven days, some mismatch between eligibility classification and IPAQ-S category was structurally possible and should be interpreted cautiously. Specifically, the 22.0% screening-positive rate observed among low-activity participants most likely reflects a structural measurement mismatch rather than a true behavioral anomaly: the EAI captures psychological compulsivity — including withdrawal and conflict symptoms that are activated precisely when exercise is reduced — whereas the IPAQ-S captures only the seven-day behavioral expression preceding the assessment, such that habitual exercisers temporarily reducing their training volume may score low on IPAQ-S while simultaneously scoring high on EAI, yielding EAI-defined risk in the absence of high concurrent exercise volume. Additional self-report limitations include social desirability bias and common-method variance, which may inflate associations between questionnaire-based constructs. Third, the lack of eating disorder screening tools (e.g., EDE-Q) makes it impossible to distinguish between primary EA (independent compulsive exercise), and secondary EA (exercise as eating disorder compensatory behavior), restricting the interpretation of the weight control and nutritional attributes results. Fourth, the cross-sectional design precludes causal inference regarding temporal relationships among study variables. Fifth, the EAI functions as a screening instrument rather than a diagnostic tool; prevalence estimates reflect screening sensitivity and require clinical confirmation to establish true incidence. Sixth, univariate comparisons were conducted without formal correction for multiple testing, meaning some findings with smaller effect sizes may reflect chance variation. Findings most vulnerable to Type I error inflation in the present analyses are those with small effect sizes and p-values approaching but not substantially below 0.05, specifically the univariate associations for values/beliefs (d = 0.14), habitual practices (d = 0.33), and chronic disease status (Cramér's V = 0.09); by contrast, findings with medium-to-large effect sizes — performance motives (d = 0.50), nutritional attributes (d = 0.45), weight control (d = 0.44), and the physical activity dose–response pattern (p ≤ 0.001) — are substantially less susceptible to chance variation and should be considered the more robust observations of this study. Seventh, perfectionism—a robust predictor of EA [61, 62]—was not measured. Given that 86.7% of participants lived with family, a context where socially prescribed perfectionism is salient in collectivist cultures, the observed patterns may partly reflect unmeasured perfectionistic cognitions. Future research should incorporate perfectionism assessment in Arabic-speaking athlete populations. Future directions Priority research directions include: (1) Longitudinal designs with repeated assessments across 6–24 months to establish temporal precedence and test hypothesized causal pathways (e.g., does stress predict subsequent EA risk development? Do EA behaviors generate psychological distress over time?); (2) Eating disorder co-assessment using validated instruments (EDE-Q, Eating Disorder Inventory-3) to operationalize primary versus secondary EA and clarify the proportion of EA risk attributable to independent exercise compulsion versus eating pathology; (3) Objective measurement triangulation incorporating accelerometry/wearables for physical activity, structured clinical interviews (SCID) for psychiatric diagnoses, and targeted hormonal assays (leptin, T3, cortisol) to verify self-reported constructs and explore RED-S prevalence; (4) Representative population sampling using probability-based recruitment to generate unbiased prevalence estimates and enable comparison with international benchmarks; (5) Mediation and moderation analyses testing whether food choice motives mediate stress-EA relationships or whether perfectionism moderates physical activity-EA associations, using pre-registered analytic plans; (6) Cultural validation extending instrument adaptation to other Arab populations (Gulf States, North Africa, Levant) to establish measurement invariance and regional norms; (7) Implementation science evaluating feasibility, acceptability, and effectiveness of fitness-based screening protocols and multidisciplinary referral pathways through pragmatic trials with process evaluation; and (8) Intervention development adapting evidence-based treatments for EA (acceptance-commitment therapy, cognitive-behavioral therapy) to Jordanian cultural contexts with pilot testing and iterative refinement based on stakeholder feedback.