Changes in eating disorder symptoms and influential factors among Chinese college students: a four-wave longitudinal study Authors: Mingze Sun, Jiaojian Dai, Yukai Li, Sijie Zhu, Chenjing Wu, Dongfang Wang Published: 2026 ABSTRACT AbstractBackgroundEating disorder symptoms (EDS) have become increasingly prevalent worldwide, especially among college students. This study examines the change patterns and influential factors of EDS among college students in China.MethodsA total of 5608 students participated in assessments conducted at four time points with 6-month intervals. The Short Form of the Eating Disorder Examination Questionnaire (EDE-QS) was used to evaluate EDS across four waves. Participants’ sample characteristics, clinical variables [i.e., insomnia, depression, anxiety, and body dysmorphic disorder (BDD)], individual psychological factors (i.e., self-control) and environmental factors (i.e., family functioning, school connectedness, and cyber-victimization) were assessed. Latent growth mixture modeling was used to explore trajectories of EDS. Multivariate logistic regressions were used to examine influential factors for trajectory membership.ResultsThree distinct patterns of change in EDS were observed: low-decreasing (79.1%), moderate-stable (17.5%), and high-increasing (3.4%). Participants residing in urban or town areas were more likely to belong to the moderate-stable and high-increasing groups. Meanwhile, compared with the low-decreasing group, individuals in the moderate-stable group reported higher insomnia symptoms, anxiety symptoms, BDD symptoms, and cyber-victimization, as well as lower levels of family functioning, school connectedness, and self-control. Individuals in the high-increasing group scored higher on anxiety symptoms, BDD symptoms, and cyber-victimization, and lower on family functioning, school connectedness, and self-control, compared with the low-decreasing group.ConclusionCollege students exhibited heterogeneous trajectories of EDS, underscoring the necessity for dynamic monitoring. Furthermore, individualized interventions targeting high-risk individuals based on identified influential factors represent a promising approach.Keywords: Eating disorders symptoms, Trajectories, Influential factors, College students, Longitudinal survey Introduction Eating disorder symptoms (EDS) are characterized by marked and persistent disturbances in eating attitudes and behaviors, the core features of which are chronic, severe disordered eating accompanied by negative emotions and cognitions that ultimately compromise an individual’s overall social, psychological, and physical functioning [1]. EDS typically emerge during adolescence, and their prevalence increases from early to late adolescence [2, 3] under the intensifying social pressure that compels individuals to conform to cultural ideals [4, 5]. College students constitute the most vulnerable age group for developing EDS [6] owing to their heightened risk of developing unhealthy eating habits and their exposure to peer pressure, image-building demands, and social-media influences [7]. Recent data indicate that the global prevalence of EDS is approximately 19.7% among college students [8]. However, epidemiological studies on the prevalence of EDS among Chinese college students remain extremely limited, and the sample sizes of existing surveys are generally small. For example, Bai and colleagues investigated 271 Chinese college students and found the prevalence of EDS to be 17.0% [9]. Given that EDS may increase the risk of physical and psychological disorders [10–12], large-scale surveys investigating the prevalence of EDS among Chinese college students and identifying multisystemic influential factors are urgently warranted. As a modifiable behavior, EDS may fluctuate over time and tend to exhibit distinct change patterns. Although previous studies have examined the longitudinal trajectories of EDS, most have focused on adolescent populations. For example, Fairweather-Schmidt and Wade [13] identified three distinct trajectories for EDS (i.e., low-static group, 90%; attenuating group, 7%; and escalating group, 3%) among female adolescents aged 12–19 years. Soidla et al. [14] identified three similar trajectories in adolescents aged 11–16 years: 16% exhibited low-stable EDS, 36% showed high and progressively increasing symptoms, and 48% displayed moderate but increasing symptomatology. Furthermore, Aimé et al. [15] observed five trajectories for EDS in a cohort of 739 students aged 13–18 years: no symptoms (58%), stable-moderate (22%), high-increasing (10%), high-decreasing (4%), and chronic (6%). Although these symptom patterns often persist into the college years, evidence regarding how these trajectories evolve during university remains scarce. Previous studies have examined the influencing factors of EDS at both environmental and individual levels. For example, school bullying has been associated with an increased risk of developing EDS [16], whereas peer support has been shown to ameliorate the course of these disorders [17]. According to the family process model, individuals from dysfunctional families are more prone to EDS [18]. Moreover, risks in the online environment, such as cyber-victimization, may also serve as risk factors for EDS among college students [19]. At the individual level, EDS have been linked to lower self-control [20]; impairments in self-control can render individuals more vulnerable to impulsive behaviors, including aberrant eating [21]. Additionally, psychopathologies—particularly insomnia, depression, anxiety, and body dysmorphic disorder (BDD)—are well-established predictors of EDS [22–25]. However, it remains unclear whether these factors can distinguish different patterns of change in EDS. Mapping the course of EDS among college students offers substantial advantages over cross-sectional approaches. A nuanced understanding of the longitudinal patterns of EDS and whether these courses are uniform or different among subgroups may help in screening and assessing individuals at high risk. Equally critical is the specification of influential factors that differentiate trajectory membership, because such knowledge enables the design of interventions tailored to the unique features of each developmental pathway. Guided by these premises, the present study used a large, prospective cohort design to (1) delineate distinct developmental trajectories of EDS among college students across a two-year period and (2) identify influential factors that differentiate membership in these trajectories. Materials and methods Study population We employed a four-wave prospective cohort design to examine trajectories in the mental health of college students. Detailed sampling and data collection have been described elsewhere [26]. In brief, we administered an online questionnaire at baseline (Time 1, T1) from 17 to 29 October 2023 at three universities in Shanxi and Hunan provinces, China, using convenience sampling. Follow-up assessments occurred at 6-month intervals: April 2024 (Time 2, T2), October 2024 (Time 3, T3), and April 2025 (Time 4, T4). At each wave, respondents were screened for data quality using four exclusion criteria: (1) excessively short response time, (2) inconsistent identity details (e.g., unidentifiable student ID), (3) invalid responses (e.g., patterned responses), and (4) self-reported history of mental disorders. After each survey wave, participant data were matched with those from the previous wave. Ultimately, 5608 participants were retained for analysis, and the screening flowchart is presented in Fig. 1. Participation flow chart Every survey was administered via “Survey Star” (https://www.wjx.cn), a widely adopted online survey platform in China. Researchers distributed survey invitation letters to students via class-level WeChat groups; participants accessed the questionnaire by scanning an embedded QR code. Electronic informed consent was obtained from all participants after the study aims, confidentiality protections, and the right to withdraw without penalty had been explained. Additional guardian consent was secured for participants under 18 years of age. Participants were informed that their involvement was voluntary; each survey was conducted anonymously, and a unique student ID assigned to every participant was used solely for longitudinal data matching before full de-identification during analysis. The protocol was approved by the Health Research Ethics Committee of the School of Psychology, South China Normal University (SCNU-PSY-2023-345). Measurements Due to time constraints, it was not feasible to assess every construct at every wave. Accordingly, variables were drawn from the time points at which they were measured: EDS were assessed in all four waves; sample characteristics and other clinical variables (i.e., insomnia, depression, anxiety, and BDD symptoms) were drawn from the baseline survey (T1); individual psychological factors (i.e., self-control) and environmental factors (i.e., family functioning, school connectedness, and cyber-victimization) were measured at T4. The details are as follows: Sample characteristic variables (T1) This study also included several sample characteristic variables as covariates: age, sex, ethnicity, parental marital status, residence location, single child status, parental education, family economic status, presence of chronic physical illness, family history of mental disorders, and body mass index (BMI, kg/m2). EDS (T1–T4) EDS was assessed with the 12-item Short Form of the Eating Disorder Examination Questionnaire (EDE-QS) [27]. Items are scored from 0 to 3, yielding a total range of 0–36, with higher scores indicating greater symptom severity during the preceding seven days. A score ≥ 15 has been validated to identify clinically significant EDS cases (sensitivity = 83%, specificity = 85%) [28]. To minimize the risk of excluding participants who already manifest—or are on the threshold of—clinically meaningful pathology, a more conservative cutoff of 13 was implemented [29, 30]. The Chinese version of the EDE-QS has demonstrated satisfactory psychometric properties [31]. In the present study, the Cronbach’s α across the four time points ranged from 0.92 to 0.93 (T1 = 0.92; T2 = 0.93; T3 = 0.93; T4 = 0.93). Other clinical variables (T1) Insomnia symptoms were assessed through three core items in the Youth Self-Rating Insomnia Scale (YSIS) [32]: difficulty falling asleep, difficulty maintaining sleep, and early morning awakening. Each item ranging from 1 (never) to 5 (6–7 times/week). Higher scores indicate more severe insomnia symptoms during the past month. Participants were categorized as having significant insomnia if they endorsed ≥ 4 (i.e., 3–5 times/week) on at least one item. Item selection and score definition were informed by prior research [25, 33]. In the present study, the Cronbach’s α was 0.75. Depressive and anxiety symptoms were assessed with the 2-item Patient Health Questionnaire (PHQ-2) [34] and the 2-item Generalized Anxiety Disorder Scale (GAD-2) [35], respectively. Participants rated the frequency of each item on a 0 (not at all) to 3 (nearly every day) scale, with higher total scores indicating greater severity in the past two weeks. In the present study, Cronbach’s α was 0.79 for the PHQ-2 and 0.85 for the GAD-2. The Yale-Brown Obsessive-Compulsive Scale Modified for Body Dysmorphic Disorder (BDD-YBOCS) was used to assess BDD symptoms over the past week [36]. The BDD-YBOCS is typically administered as a clinician-rated interview. However, it has been adapted as a self-report instrument with demonstrated validity and reliability [37]. In this study, we employed a self-report format of BDD-YBOCS. The scale consists of 12 items, which are rated on a 0–4 scale, where 0 represents no symptoms and 4 suggests extreme BDD symptoms. Total scores range from 0 to 48, with higher scores indicating greater BDD severity. The Chinese self-report version of the BDD-YBOCS has demonstrated satisfactory psychometric properties in college students [38]. The Cronbach’s α was 0.90 in this sample. Self-control (T4) The Brief Self-Control Scale (BSCS) consists of seven self-report items rated on a 5-point Likert scale (1 = strongly disagree, 5 = strongly agree), with higher total scores indicating greater self-control [39]. The Chinese version has demonstrated satisfactory reliability and validity [40]; in the current sample, Cronbach’s α was 0.77. Environmental factors (T4) The family APGAR scale was used in this study [41]. The scale consists of 5 items tapping five dimensions: adaptation, partnership, growth, affection, and resolution. Each item is rated from 0 (hardly ever) to 2 (almost always). Total scores range from 0 to 10, with higher scores indicating better family functioning. Cronbach’s α in the present sample was 0.97. The School Connectedness Scale (SCS) comprises 10 items across three dimensions: peer support, school belongingness, and teacher support [42]. Responses are given on a 5-point scale ranging from 1 (definitely disagree) to 5 (completely agree), with higher scores reflecting stronger school connectedness. Cronbach’s α was 0.92. The Short Form of the Cyber-aggression and Cyber-victimization Scale (SF-CAV), originally developed by Shapka et al. [43] and culturally adapted by Xie et al. [44], includes two subscales: cyber-aggression (6 items) and cyber-victimization (6 items). Only the cyber-victimization subscale was used to assess online risk exposure. Items are rated from 0 (has never happened) to 4 (happens every day), with higher scores indicating greater cyber-victimization. Cronbach’s α was 0.97. Statistical analysis Preliminary analyses Univariate analysis was used to compare baseline EDS levels between participants who completed all four waves and those lost to follow-up. The variance inflation factor (VIF) was calculated to assess multicollinearity [45], with all variables showing values of 2.51 or below, indicating no serious multicollinearity. Harman’s single-factor analysis was conducted to examine potential common method variance [46]. All items from the study variables were included in the analysis, revealing 20 factors with eigenvalues greater than 1, with the first factor accounting for 21.53% of the variance. This suggests that common method bias was not a serious concern in the present study. Trajectory analysis We employed latent growth mixture modeling (LGMM) [47] to examine EDS among college students. We tested models ranging from 1 to 5 classes, selecting the best-fitting model according to lower Akaike information criterion (AIC), Bayesian information criterion (BIC), and sample-size-adjusted BIC (a-BIC), higher entropy values, and significant Vuong-Lo-Mendell-Rubin likelihood ratio test (VLMR-LRT) and bootstrap likelihood ratio test (BLRT) results. Analysis of influencing factors of trajectory membership Univariate analysis was performed to examine sample characteristics across EDS trajectory groups. Multivariate logistic regression was then conducted to determine the associations between sample characteristics, clinical variables, individual psychological factors, environmental factors, and trajectories of EDS. Sample characteristic variables that were significant in the univariate analysis were entered as covariates in the multivariate analyses. Odds ratios (ORs) with 95% confidence intervals (CIs) were used to quantify the strength of the associations, and two-tailed p < 0.05 was considered statistically significant. All statistical analyses were performed using SPSS 24.0 and Mplus 8.3. Results Descriptive statistics Of the 5608 participants, 20.5% were male, with a mean age of 19.44 (SD = 1.32) years. Other sample characteristics are presented in Table 1. Compared with those lost to follow-up, completers scored slightly lower on the EDE-QS (5.08 ± 6.02 vs. 5.58 ± 6.81, t = − 5.05, p < 0.001). They were significantly younger (19.44 ± 1.32 vs. 20.34 ± 1.67, t=− 39.02, p < 0.001), and a smaller proportion were male (20.5% vs. 37.0%, χ2 = 489.67, p < 0.001). Table 1Comparisons of sample characteristics among distinct trajectory memberships [N(%)]T1 sample characteristicsOverall (N = 5608)Low-decreasing (N = 4334, 79.1%)Moderate-stable (N = 980, 17.5%)High-increasing (N = 194, 3.4%)P valueSex Male1148 (20.5)919 (20.7)180 (18.4)49 (25.3)χ2 = 5.57 Female4460 (79.5)3515 (79.3)800 (81.6)145 (74.7)Age (years) M(SD)19.44 (1.32)19.46 (1.31)19.38 (1.32)19.25 (1.40)F = 3.60*Ethnicity Hana5555 (99.1)4397 (99.2)967 (98.7)191 (98.5)χ2 = 2.85 Others53 (0.9)37 (0.8)13 (1.3)3 (1.5)Parental marital status Married5178 (92.3)4101 (92.5)897 (91.5)180 (92.8)χ2 = 1.10 Not current marriedb430 (7.7)333 (7.5)83 (8.5)14 (7.2)Residence location Rural2873 (51.2)2309 (52.1)479 (48.9)85 (43.8)χ2 = 10.31* Town1478 (26.4)1143 (25.8)282 (28.8)53 (27.3) Urban1257 (22.4)982 (22.1)219 (22.3)56 (28.9)Single child status Yes1045 (18.6)821 (18.5)177 (18.1)47 (24.2)χ2 = 4.26Father’s education Primary school or below1017 (18.1)808 (18.2)178 (18.2)31 (16.0)χ2 = 3.30 Junior high school2806 (50.0)2231 (50.3)484 (49.4)91 (46.9) Senior high school1166 (20.8)915 (20.6)204 (20.8)47 (24.2) College or above619 (11.1)480 (10.8)114 (11.6)25 (12.9)Mother’s education Primary school or below1168 (20.8)932 (21.0)200 (20.4)36 (18.6)χ2 = 4.28 Junior high school2849 (50.8)2266 (51.1)491 (50.1)92 (47.4) Senior high school1052 (18.8)816 (18.4)192 (19.6)44 (22.7) College or above539 (9.6)420 (9.5)97 (9.9)22 (11.3)Family economic status Good200 (3.6)159 (3.6)36 (3.7)5 (2.6)χ2 = 2.43 Fair4288 (76.5)3401 (76.7)744 (75.9)143 (73.7) Poor1120 (20.0)874 (19.7)200 (20.4)46 (23.7)Chronic physical illness Yesc130 (2.3)96 (2.2)28 (2.9)6 (3.1)χ2 = 2.23Family history of mental disorder Yes55 (1.0)38 (0.9)16 (1.6)1 (0.5)χ2 = 5.42Body mass index (kg/m2) M (SD)22.51 (6.03)22.09 (5.90)23.79 (6.07)25.50 (6.99)F = 58.02***a Han is the major ethnic group in Chinab Not current married included separated, divorced, and widowedc Chronic physical conditions referred to having at least one of arthritis, angina, asthma, diabetes, visual impairment, or hearing problems*p < 0.05; **p < 0.01; ***p < 0.001 Comparisons of sample characteristics among distinct trajectory memberships [N(%)] a Han is the major ethnic group in China b Not current married included separated, divorced, and widowed c Chronic physical conditions referred to having at least one of arthritis, angina, asthma, diabetes, visual impairment, or hearing problems *p < 0.05; **p < 0.01; ***p < 0.001 Identification of EDS trajectories Model fit indices for the LGMM analyses are reported in Table 2. The 2- to 5-class models demonstrated adequate fit to the data. The AIC, BIC, and a-BIC decreased monotonically as the number of classes increased, and the VLMR-LRT and BLRT results remained statistically significant for all models. Among these, the three-class solution exhibited the highest entropy value (0.946), indicating clearer and more precise class separation. Additionally, compared with the two-class solution, the three-class solution identified a meaningful additional class (high-increasing) that, despite its small size, aligned with previous findings [13]. However, the four-class solution added a class that was similar to an existing class solution (moderate-stable) in the three-class solution. According to the entropy value and the interpretability of the results, the three-class model was finally selected as the optimal model. The average probability of samples belonging to each potential category ranged from 93.0 to 98.7%. Table 2Goodness of fit statistics for each class solutionModelAICBICa-BICEntropyVLMR-LRTBLRTSmallest classTotal1133766.93133826.62133798.022131548.32131627.91131589.780.927< 0.001< 0.00115.5% 3 130195.62 130295.10 130247.43 0.946 < 0.001 < 0.001 3.4% 4129215.46129334.84129277.640.9370.012< 0.0012.9%5128323.03128462.31128395.570.945< 0.001< 0.0011.3%Male128148.3628193.7728165.19227511.7327572.2827534.160.961< 0.001< 0.00116.5% 3 27182.24 27257.92 27210.28 0.961 0.032 < 0.001 4.4% 426878.8226969.6426912.470.9580.121< 0.0014.0%526617.7026723.6626656.960.9750.005< 0.0011.5%Female1105497.67105555.30105526.702103853.56103930.39103892.260.916< 0.001< 0.00115.0% 3 102861.84 102957.88 102910.22 0.942 < 0.001 < 0.001 3.2% 4102183.91102299.16102241.960.9300.040< 0.0012.7%5101543.28101677.74101611.010.948< 0.001< 0.0011.3%AIC Akaike information criterion, BIC Bayesian information criterion, a-BIC sample size-adjusted Bayesian information criterion, VLMR-LRT Vuong-Lo-Mendell-Rubin likelihood ratio test, BLRT bootstrapped likelihood ratio testBold indicates final class solution Goodness of fit statistics for each class solution AIC Akaike information criterion, BIC Bayesian information criterion, a-BIC sample size-adjusted Bayesian information criterion, VLMR-LRT Vuong-Lo-Mendell-Rubin likelihood ratio test, BLRT bootstrapped likelihood ratio test Bold indicates final class solution The three identified classes were characterized as follows (see Fig. 2). Participants in Class 1 (79.1%, low-decreasing) reported initially low levels of EDS that subsequently decreased over time (intercept = 4.11, p < 0.001; slope = − 0.93, p < 0.001). Participants in Class 2 (17.5%, moderate-stable) exhibited initially moderate EDS levels with a slight subsequent increase, yet remained relatively stable overall (intercept = 8.73, p < 0.001; slope = 0.54, p < 0.001). Participants in Class 3 (3.4%, high-increasing) reported initially high levels of EDS that subsequently increased significantly (intercept = 12.19, p < 0.001; slope = 2.65, p < 0.001). Distinct trajectories of eating disorder symptoms based on the overall sample Sensitivity analysis: To examine the robustness of the trajectory characteristics, we employed the same analytical approach to examine the developmental trajectories of EDS in male and female participants separately. Based on model fit indices, the three-class model was identified as optimal for both groups (see Table 2). As illustrated in Fig. 3, male (see Fig. 3A) and female (see Fig. 3B) college students exhibited developmental patterns of EDS similar to those observed in the overall sample. Sex-specific distinct trajectories of eating disorder symptoms. A Represents males; B represents females Factors associated with class membership The sample characteristics across different trajectory memberships are shown in Table 1. These groups differed significantly in terms of age (p = 0.028), residence location (p = 0.035), and BMI (p < 0.001). Intercept models were established to evaluate the school/college-level heterogeneity for trajectories of EDS, with no significant heterogeneity found (all p > 0.05). We established the low-decreasing group as the reference group and compared it with the other two groups. As shown in Table 3, the likelihood of being in the moderate-stable group increased if participants resided in town (OR = 1.25, 95% CI 1.04–1.49) or urban areas (OR = 1.26, 95% CI 1.04–1.53), had higher BMI (OR = 1.05, 95% CI 1.04–1.06),, increased insomnia symptoms (OR = 1.05, 95% CI 1.01–1.09), elevated anxiety symptoms (OR = 1.16, 95% CI 1.05–1.28), greater BDD symptoms (OR = 1.06, 95% CI 1.05–1.08), and cyber-victimization (OR = 1.12, 95% CI 1.09–1.16), whereas stronger family functioning (OR = 0.97, 95% CI 0.94–0.99), school connectedness (OR = 0.96, 95% CI 0.94–0.97), and higher self-control (OR = 0.87, 95% CI 0.85–0.89) were associated with decreased risk of moderate-stable EDS. Table 3Factors associated with class membershipModerate-stableHigh-increasingAge1.04 (0.98,1.10)0.99 (0.88,1.12)Residence location (Rural = Ref.) Town1.25 (1.04,1.49)**1.39 (0.95,2.03) Urban1.26 (1.04,1.53)**2.03 (1.39,2.96)***Body mass index1.05 (1.04,1.06)***1.08 (1.06,1.10)***T1 insomnia symptoms1.05 (1.01,1.09)*1.07 (0.99,1.15)T1 depressive symptoms0.94 (0.84,1.04)0.88 (0.71,1.08)T1 anxiety symptoms1.16 (1.05,1.28)**1.38 (1.13,1.67)**T1 BDD symptoms1.06 (1.05,1.08)***1.08 (1.05,1.10)***T4 self-control0.87 (0.85,0.89)***0.83 (0.79,0.88)***T4 family functioning0.97 (0.94,0.99)*0.93 (0.87,0.98)*T4 school connectedness0.96 (0.94,0.97)***0.95 (0.93,0.98)***T4 cyber-victimization1.12 (1.09,1.16)***1.23 (1.19,1.28)***The low-decreasing group as the reference groupBDD= body dysmorphic disorder*p < 0.05; **p < 0.01; ***p < 0.001 Factors associated with class membership The low-decreasing group as the reference group BDD= body dysmorphic disorder *p < 0.05; **p < 0.01; ***p < 0.001 The results also indicated that residing in urban areas (OR = 2.03, 95% CI 1.39–2.96), higher BMI (OR = 1.08, 95% CI 1.06–1.10), elevated anxiety symptoms (OR = 1.38, 95% CI 1.13–1.67), greater BDD symptoms (OR = 1.08, 95% CI 1.05–1.10), and cyber-victimization (OR = 1.23, 95% CI 1.19–1.28) were associated with increased likelihood of high-increasing EDS, whereas strong family functioning (OR = 0.93, 95% CI 0.87–0.98), school connectedness (OR = 0.95, 95% CI 0.93–0.98), and heightened self-control (OR = 0.83, 95% CI 0.79–0.88) diminished this risk. Discussion Using a large-scale cohort of college students, the current study mapped the developmental course of EDS over a two-year period. We identified three distinct trajectories of EDS among college students. The most prevalent pattern was the low-decreasing group (78.4%), characterized by low or mild EDS, consistent with previous studies on childhood and adolescence [13, 15, 48]. Meanwhile, 17.5% of participants exhibited moderate and slightly fluctuating EDS during the same period. These fluctuations underscore the vulnerability of college students to EDS, potentially shaped by peer pressure, image curation demands, and social-media influences [7]. Additionally, 3.4% of the sample reported high-increasing EDS. Although this proportion was slightly below the minimum class proportion recommended by LGMM (5%) [49], previous studies have also observed similar trajectory patterns in adolescent samples [13]. Given that their symptoms have remained unresolved over time, early intervention by both educators and clinicians is warranted to prevent deterioration. As stated in the introduction, identifying multisystem influential factors that can differentiate the developmental trajectories of EDS is crucial, as it provides more effective guidance for the prevention and intervention of EDS. This study primarily examined factors associated with EDS trajectories at the individual and environmental levels. From a sociodemographic perspective, participants from urban or town areas exhibited higher-risk EDS developmental trajectories compared with those residing in rural areas. This finding is consistent with previous research [50], possibly because urbanization heightens food resources availability, food security, and dietary intake, thereby serving as a risk factor for EDS [51]. Consistent with previous studies [52], participants with higher BMI may be more likely to be at risk for EDS. Clinically, students with higher anxiety symptoms are more likely to report moderate-stable or high-increasing trajectories of EDS. It is well-documented that anxiety symptoms may be early signs of later EDS [53]. Our findings revealed a positive association between BDD symptoms and increased odds of EDS. Schneider et al. reported that adolescents with BDD exhibited significantly higher levels of EDS compared to those without BDD [24]. Prior research has consistently supported the association between BDD and EDS in terms of clinical features, onset and course, and cognitive biases [54–56]. Consequently, it has been proposed that BDD and EDS may constitute a distinct “body image spectrum” of disorders [54, 57]. Although previous studies have shown that insomnia and depressive symptoms are associated with EDS [23, 25], insomnia symptoms only exhibited a weak effect on the moderate-stable group, while depressive symptoms did not demonstrate significant effects in the multivariate models. The reason may lie in the high correlations among insomnia, depression, anxiety, and BDD. Our findings indicate that, in college students, anxiety and BDD symptoms may be better predictors of EDS than insomnia and depressive symptoms. Furthermore, from a psychological perspective, stronger self-control was associated with reduced risk of EDS. Individuals with higher levels of self-control exhibited fewer impulse control difficulties, such as binge eating and alcohol abuse [21]. According to the dual-modes model of self-control, self-control comprises an impulsive system and a control system; either strong impulses or weak controls can lead to impulsivity [58], which in turn increases vulnerability to EDS [59]. Our findings also confirm the association between environmental factors and EDS among college students. Specifically, better family functioning was associated with a reduced risk of moderate-stable or high-increasing trajectories of EDS, which is in line with previous work [18]. Positive family functioning reflects stronger cohesion and adaptability, as well as greater family support, all of which serve as protective factors against EDS [60, 61]. Similarly, students with high levels of school connectedness are at lower risk for EDS’ levels and growth. Although direct evidence regarding the relationship between school connectedness and EDS remains limited, previous research has found that deficits in perceived peer support predict future increases in EDS [62]. In contrast, cyber-victimization may be associated with an increased risk of EDS. Based on the Diathesis-Stress Model, psychological disorders, including EDS, result from the interaction between vulnerability factors and stressors [63]; cyber-victimization exposes individuals to humiliation, isolation, and chronic, unrelenting stress. Lee et al. found that cyber-victimization may contribute to EDS through its detrimental effect on self-esteem [19]. Implications The findings carry important implications for the prevention and intervention of EDS among college students. First, the heterogeneous developmental trajectories of EDS during the college years suggest that a “one-size-fits-all” prevention strategy may be of limited effectiveness. Prevention programs should be tailored to meet the specific needs of different risk groups, with particular attention paid to students in the moderate-stable and high-increasing groups. Early identification and targeted support for students exhibiting risk factors may help shift at-risk individuals toward more favorable developmental trajectories. Clinicians and educational practitioners can reduce EDS risk among college students by enhancing self-control capacities, strengthening family communication and school support systems, addressing body image concerns, and alleviating anxiety symptoms. Given the strong association between cyber-victimization and eating disorders, anti-bullying initiatives and cyber-safety education should also be incorporated as core components of campus mental health programs. Strengths and limitations This study presents several strengths, including multi-wave assessments of EDS, a relatively large sample size, a person-centered analytical approach, and the systematic examination of influential factors. However, several limitations should be acknowledged. First, EDS and other clinical variables were assessed through self-report measures, and participants may have introduced reporting bias due to recall bias or social desirability bias. Future research could consider incorporating clinical interviews to enhance measurement validity and reduce the impact of reporting bias. Second, due to the online survey design and convenience sampling, this study experienced a relatively high attrition rate. The main reasons included senior students leaving school or graduating during subsequent surveys, as well as scheduling conflicts with academic programs in some colleges, which reduced the response rate. The data showed that participants who dropped out reported more severe EDS at baseline, which may have compromised the accuracy of the findings to some extent. Future research may benefit from implementing more rigorous retention strategies (e.g., multiple contact methods, participant incentives) to enhance follow-up rates and reduce potential bias. Third, participants in this study were recruited from three universities/colleges, which may involve contextual constraints or cultural specificity; therefore, the generalizability of the findings to the national population should be interpreted with caution. Future research could recruit more diverse and representative samples to enhance the external validity of the conclusions. Conclusion This study makes a unique contribution to the literature by examining change trajectory of EDS and their influential factors in a large sample of college students. Three patterns were identified: low-decreasing, moderate-stable, and high-increasing. Residence location, BMI, insomnia symptoms, anxiety symptoms, BDD symptoms, cyber-victimization, family functioning, school connectedness, and self-control may distinguish distinct trajectories for EDS. These findings could inform mental health professionals regarding how to provide individualized and appropriate interventions for college students with EDS. Acknowledgements The authors want to express their sincere gratitude to all participants for participating in the study. Author contributions Mingze Sun: writing-original draft, investigation, conceptualization. Jiaojian Dai: writing-review & editing, formal analysis, investigation. Yukai Li: writing-review & editing, formal analysis, conceptualization. Sijie: Zhu: writing-review & editing, formal analysis, conceptualization. Chenjing Wu: writing-review & editing, formal analysis. Dongfang Wang: conceptualization; investigation; project administration; resources; supervision; validation; writing—review & editing. Funding This research was funded by the Research Center for Brain Cognition and Human Development, Guangdong, China (2024B0303390003), the MOE (Ministry of Education in China) Major Project of Philosophy and Social Sciences Research (2025JZDZ024), Psychological services and counselling bases for the Happy Guangzhou project, which received funding from the Guangzhou government, and Striving for the First-Class, Improving Weak Links and Highlighting Features (SIH) Key Discipline for Psychology in South China Normal University. Data availability The data for this research is available from the corresponding author upon reasonable request. Declarations Ethics approval and consent to participate A detailed overview of the study and its goals, including potential risks of discomfort, was made available online to prospective participants. By choosing to complete the anonymous online survey, participants provided implied consent. Ethics approval for this study was obtained from the Health Research Ethics Committee of the School of Psychology, South China Normal University (SCNU-PSY-2023-345). Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Mingze Sun and Jiaojian Dai contributed equally to this work. References Associated Data Data Availability Statement The data for this research is available from the corresponding author upon reasonable request.