Original quantitative research – Disordered eating profiles and associations with student sociodemographic characteristics: a latent class analysis among adolescents in Canada

Abstract

Introduction: Despite its importance to health and well-being, evidence on disordered eating (DE) patterns among Canadian adolescents remains limited. This study aimed to (1) identify latent DE classes among adolescents in Canada; and (2) investigate whether students' sociodemographic characteristics predict DE class membership.

Methods: Student-level survey data were obtained from the 2021–2022 COMPASS study. The sample included 24 639 students in grades 9 to 12 from 69 secondary schools in Alberta, British Columbia, and Ontario. Latent class analysis was applied to explore DE profiles based on six items assessing weight concern, restrictive or binge eating, purging, and compensatory behaviours. Logistic regression was conducted to examine associations between DE class membership and sociodemographic characteristics (grade, gender, race and ethnicity, family affluence, and weight status).

Results: Five distinct DE classes were identified: no-to-mild DE (32.5%), changed eating with weight concern (30.7%), binge eating with weight concern (25.7%), binge eating without weight concern (7.3%), and complex DE (3.7%). Cisgender girls, transgender and gender-diverse youth, adolescents from less affluent households, and those with very high weight were at greater risk of belonging to classes with higher symptom engagement compared to the no-to-mild DE class and cisgender boys, more affluent peers, and those with healthy weight, respectively.

Conclusion: Findings indicate that Canadian adolescents exhibit heterogeneous DE patterns, with pronounced disparities by gender, family affluence, and weight status. Recognizing equity-related disparities in DE patterns highlights the importance of future longitudinal research to examine developmental trajectories of DE, and to inform tailored prevention and early intervention strategies that address the needs of diverse adolescent populations.

Keywords: adolescents, disordered eating, latent class analysis, patterns, youth

Introduction

Disordered eating (DE) refers to maladaptive eating thoughts and behaviours, including weight or shape preoccupation, restrictive or binge eating, purging, and misuse of laxatives, diuretics, diet pills, or steroids.^{Footnote 1} DE thoughts and behaviours are eating disorder (ED) symptoms that do not meet clinical diagnostic criteria but are recognized as risk factors for ED development.^{Footnote 2} A 2023 systematic review and meta-analysis reported that approximately 22% of youth globally engaged in DE.^{Footnote 3} The high prevalence of DE is a significant public health concern due to its associations with adverse health outcomes, such as psychological distress, depression, and suicide ideation.^{Footnote 4Footnote 5} In some individuals, DE behaviours may progress to clinically diagnosable EDs.^{Footnote 6Footnote 7} Without appropriate treatment, EDs can become persistent, severe, and recurrent, with the second-highest mortality rate among all mental disorders, surpassed only by opioid dependence.^{Footnote 8Footnote 9Footnote 10} Given these serious consequences, understanding DE patterns and risk factors among youth is essential for developing effective prevention and early intervention strategies.

Emerging evidence indicates that DE thoughts and behaviours form diverse, co-occurring patterns, rather than existing along a single continuum of frequency/severity.^{Footnote 11Footnote 12Footnote 13Footnote 14Footnote 15Footnote 16Footnote 17Footnote 18Footnote 19Footnote 20} Consequently, researchers have applied latent class analysis (LCA) or latent profile analysis (LPA) to identify distinct DE/ED profiles among youth, primarily in the USA^{Footnote 11Footnote 12Footnote 13Footnote 17} and Europe.^{Footnote 16Footnote 19} LCA is a statistical method used to identify latent subgroups within a population based on individuals' response patterns to observed variables (indicators).^{Footnote 21Footnote 22} Unlike traditional variable-centred methods, which assume the same effects across the population and treat DE as a continuous/unidimensional construct, potentially masking important subtypes of risk, LCA adopts a person-centred approach.^{Footnote 21Footnote 22} It enables researchers to uncover distinct behavioural patterns and gain nuanced insights into DE manifestations across populations.^{Footnote 21Footnote 22} For example, Cain et al. identified five clusters of eating, weight, and shape concerns among US young women, reflecting restrictive eating, overeating, and bulimic-like symptoms.^{Footnote 11} Similarly, another US study of male youth revealed five DE clusters, characterized by body-image disturbance, binge eating/purging, and muscularity concerns.^{Footnote 12} While several LCAs/LPAs have examined DE patterns, important gaps remain in the literature. First, many existing person-centred studies focus on young adults^{Footnote 11Footnote 13Footnote 17} or adult samples^{Footnote 20} rather than adolescents, despite adolescence being a critical developmental period for the emergence of DE behaviours.^{Footnote 6Footnote 7} Moreover, several adolescent LCAs were conducted about a decade ago,^{Footnote 14Footnote 15Footnote 16} prior to some major societal shifts—including the COVID-19 pandemic—that have altered food environments,^{Footnote 23} social media exposure,^{Footnote 24} mental health stressors,^{Footnote 25} and access to supportive resources.^{Footnote 26} Given evidence that DE behaviours are sensitive to social and contextual influences,^{Footnote 27} updated population-based analyses are needed to assess whether previously identified DE profiles remain applicable to adolescents in the contemporary context.

Second, existing LCAs/LPAs were limited by modest sample sizes^{Footnote 15Footnote 18} or restricted demographic scope,^{Footnote 11Footnote 13} constraining their ability to identify stable latent classes and to capture heterogeneity among equity-deserving populations. Growing evidence shows that transgender and gender-diverse (TGD) youth,^{Footnote 28} individuals from lower socioeconomic-status (SES) families,^{Footnote 29} racialized and ethnic minority groups,^{Footnote 30} and adolescents with larger body sizes^{Footnote 31} may face greater DE risk and increased barriers to protective resources. Thus, larger and more diverse samples are essential to adequately characterize DE profiles and to examine sociodemographic predictors of class membership with sufficient statistical power.

Third, some recent work has applied intensive longitudinal methods such as ecological momentary assessment (EMA) to examine eating-related processes and to address complementary research questions.^{Footnote 19} In contrast, survey-based LCA is uniquely suited to identifying population-level DE profiles and their sociodemographic correlates, which are critical for public health surveillance and the development of scalable prevention and early intervention strategies.

To address these critical gaps, the current study aims to (1) identify DE profiles in a large, contemporary, population-based sample of secondary school students in Canada; and (2) investigate how students' sociodemographic characteristics predict DE class membership. By identifying DE patterns and predictors, this study seeks to capture disparities in DE behaviours and risk factors to help inform more tailored and equitable prevention and intervention strategies for diverse adolescent populations.

Methods

Study design and sample

This cross-sectional study used student-level data from the 2021–22 wave of the COMPASS study (2012–2027), a prospective cohort study designed to collect hierarchical and longitudinal data from secondary school students in Canada.^{Footnote 32} Schools were intentionally selected if allowing active-information passive-consent parental permission protocols.^{Footnote 33} All adolescents in participating schools were eligible unless withdrawn by a parent/guardian and could decline participation at any time. An in-class, online questionnaire was administered to assess sociodemographic characteristics (e.g. age, gender), health behaviours (e.g. physical activity), and correlates (e.g. bullying). Ethics approval was obtained from the University of Waterloo (ORE#30118), Brock University (REB#23-087), University of Alberta (#RES0050375; Pro00129206), and participating school boards. Full COMPASS protocols are available in print^{Footnote 32} or online.^{Footnote 34}

Questions assessing DE were first introduced in the COMPASS student questionnaire (Cq) during the 2021–22 school year and administered in participating schools across Alberta, British Columbia, and Ontario. In total, 27 003 students from 69 high schools completed the survey; 2364 participants had missing responses to one or more DE items, yielding a final analytic sample of 24 639 adolescents with complete DE data.

Measures

Disordered eating

The COMPASS researchers developed a six-item DE scale adapted from existing instruments, such as the Eating Attitudes Test,^{Footnote 35} Eating Disorder Examination Questionnaire,^{Footnote 36} and Sick-Control-One-Fat-Food questionnaire,^{Footnote 37} to capture key DE thoughts and behaviours. Students were asked how often over the past 12 months they had experienced the following: 1) been preoccupied with a desire to be thinner; 2) made yourself throw up (vomit) on purpose after eating; 3) used laxatives, diet pills, diuretics (water pills), or steroids to control your weight or shape; 4) changed your eating habits in order to manage your weight or shape; 5) ate much more than most people would under the same circumstances and felt that you could not stop; and 6) had a definite fear of gaining weight or being overweight. Responses were recorded on a five-point scale: 0 = never, 1 = a few times, 2 = monthly, 3 = weekly, and 4 = daily. Total scores ranged from 0 to 24, with higher scores indicating greater DE engagement. The internal consistency of all DE items was α = 0.79. The scale's clarity, comprehensibility, and measurement invariance across sociodemographic groups have been previously demonstrated.^{Footnote 38} Preliminary data analyses indicated small cell sizes in the "weekly" to "daily" response categories, limiting further analysis. To address this issue while considering response frequency and clinical relevance, we dichotomized the five response options as: 1 = yes (combined reports of "a few times," "monthly," "weekly," and "daily"), and 2 = no (reports of "never").

Sociodemographic characteristics

Student sociodemographic characteristics were self-reported in the Cq, including grade (9–12), race and ethnicity ("White," "Black," "East Asian," "Latino," "Middle Eastern," "South Asian," "Southeast Asian," "multiethnic," "another category"), and perceived family affluence ("more comfortable," "as comfortable," "less comfortable"). Gender identity ("girl/woman," "boy/man," "nonbinary person," "Two-Spirit," "I describe my gender differently") and sex assigned at birth ("female," "male") were combined to classify participants as: "cisgender girls," "cisgender boys," or "TGD youth." Self-reported height and weight were used to calculate body mass index (BMI) categories following World Health Organization (WHO) age- and sex-specific guidelines ("underweight," "healthy weight," "overweight," and "obesity").^{Footnote 39} To avoid stigmatizing language, the weight categories were hereby referred to as "lower weight," "healthy weight," "high weight," and "very high weight," respectively.^{Footnote 40}

Statistical analysis

Descriptive statistics

Prior to latent class modelling, one-way ANOVA tests were used to descriptively compare mean ordinal DE item scores overall and across sociodemographic subgroups to characterize the distribution and engagement in DE thoughts and behaviours, and to provide contextual information for subsequent analyses.

Latent class analysis

LCA was conducted in SAS using the "PROC LCA" procedure with maximum-likelihood estimation to identify classes of students with similar DE response patterns.^{Footnote 21} We systematically fitted LCA models from one-to-six classes, comparing fit statistics at each step to determine the best-fitting solution. To address class membership instability, we ran each model with 50 random sets of starting values to ensure the global rather than local maximums were identified during parameter estimation.^{Footnote 21} Additionally, school-level clustering was accounted for by adjusting standard errors for within-school correlation, such that clustering influenced variance estimation but not the DE class identification.^{Footnote 21}

Two sets of model fit indices were used to determine the optimal number of DE classes. We started by examining primary indices. First, the corrected Akaike Information Criterion (cAIC) and adjusted Bayesian Information Criterion (aBIC) were examined, with smaller values indicating better fit.^{Footnote 21Footnote 22} Second, we assessed the likelihood ratio G-square value, considering substantial relative reductions between successive class solutions as indicative of improved model fit, rather than applying a fixed numerical threshold.^{Footnote 21Footnote 22} We also used the bootstrap likelihood ratio test (BLRT) p value (< 0.05) to determine if the model with more classes fit significantly better.^{Footnote 21Footnote 22} Third, entropy (range: 0–1) was used to evaluate classification certainty, with higher values indicating better accuracy.^{Footnote 21Footnote 22} Secondary indices included the class size, percentage of seeds associated with the best-fitting model, and number of iterations required for model convergence. Adequate class size (very small classes may indicate overfitting) ensured substantive meaning and parsimony,^{Footnote 41Footnote 42} while higher percentages of seeds and fewer iterations indicated better fit.^{Footnote 41Footnote 42} These criteria, along with model interpretability, guided the selection of the final DE classes.^{Footnote 41Footnote 42}

After identifying DE classes, we conducted multinomial logistic regression to investigate associations between five sociodemographic predictors and DE class membership.^{Footnote 21} Due to sparse cell sizes arising from overlap across DE classes, single-predictor (unadjusted) models were fitted for each variable separately within the LCA framework. Partially adjusted models including selected predictors were conducted as supplementary analyses (available from authors upon request). All analyses were conducted in SAS version 9.4 (SAS Institute Inc., Cary, NC, US), with the significance level set at α = 0.05.

Missing data handling

In the COMPASS 2021–2022 sample receiving DE items, 8.8% (n = 2364) of students had missing/incomplete DE responses. Among students with complete DE data, item non-response occurred for grade (< 1%), gender (4.3%), race and ethnicity (5.5%), family affluence (< 1%), and weight status (40.7%). Students missing DE data or the relevant sociodemographic variable were excluded from ANOVA tests on a per-analysis basis. LCA was conducted among students with complete DE responses only. In multinomial logistic regression models, students with complete DE responses were included, and missing sociodemographic values (e.g. weight status) were retained as separate categories to preserve sample size and class stability.

This approach is supported by prior analyses demonstrating that mean DE scores and subgroup differences were highly similar before and after excluding students with missing DE responses.^{Footnote 38} Additional COMPASS research shows that non-reporting of self-reported weight is systematically associated with weight perceptions and related behaviours,^{Footnote 43} suggesting that missingness may be dependent on weight values themselves, i.e. not missing at random.

Results

Sample characteristics

Mean (SD) scores of DE items overall and by sociodemographic subgroups are presented in Table 1. Overall, weight preoccupation had the highest mean score, followed by changed eating habits and fear of weight gain, whereas pills/steroid use showed the lowest score, with purging and binge eating scores also comparatively low. Significant differences in mean scores were observed across all subgroups for all items (p < 0.01), except for purging among grade subgroups. Grade 12 participants had higher means than those in lower grades for weight preoccupation, pills/steroid use, changed eating habits, and fear of weight gain. Across gender, TGD youth had the highest means for purging, pills/steroid use, and binge eating, while cisgender girls scored highest for weight preoccupation and fear of weight gain. Cisgender boys showed the lowest scores across all items. By race and ethnicity, Black students demonstrated the highest means for purging, pills/steroid use, and binge eating, while another ethnicity/multiethnic, Middle Eastern, and Latino students scored highest for weight preoccupation, changed eating habits, and fear of weight gain, respectively. In contrast, East Asian students reported the lowest scores for purging, changed eating habits, binge eating, and fear of weight gain. Students with perceived less affluent households had higher means across all items than their more affluent peers. Additionally, very-high-weight adolescents scored highest across all items, followed by adolescents with high-weight status, while lower-weight students scored lowest.

Latent DE classes

A series of LCA models ranging from one to six classes were fitted to identify the optimal number of DE classes. Primary fit statistics (Table 2) initially indicated that the six-class model provided the best fit, as supported by a substantial drop in the likelihood ratio G-square value, the lowest cAIC and aBIC values, and a significant BLRT p value. However, secondary fit indices (Table 3) suggested potential overfitting and convergence issues. Specifically, the six-class model included a class comprising only 1% of the sample, had the lowest proportion of seeds associated with the best-fitting solution, and required over 5000 iterations to converge. Given these concerns, the five-class model was ultimately selected as the most parsimonious and interpretable solution.

Figure 1 displays the class-membership probabilities and item-response probabilities for each DE item. Based on these probabilities, five distinct DE patterns were identified. The first and largest class (32.5%) was labelled "no-to-mild DE," characterized by low response probabilities (< 0.3) across all items. The second class (30.7%) was labelled "changed eating with weight concern," as students showed a high likelihood of changing their eating habits to manage weight/shape, along with weight preoccupation and fear of weight gain. The third class (25.7%) was labelled "binge eating with weight concern," as students exhibited a pattern of binge eating, accompanied by weight-related concerns. The fourth class (7.3%) was labelled "binge eating without weight concern," due to the highest response probabilities for binge-related item, but without weight preoccupation or fear of weight gain. The fifth and smallest class (3.7%) was labelled "complex DE," with students showing high probabilities (> 0.8) across all DE items.

Figure 1 - Text description

Disordered eating thought or behaviour	No-to-mild DE (32.5%)	Changed eating with weight concern (30.7%)	Binge eating with weight concern (25.7%)	Binge eating without weight concern (7.3%)	Complex DE (3.7%)
Weight preoccupation	0.1441	0.8933	0.9917	0.0844	0.9353
Purging	0.0051	0.0339	0.2891	0.0822	0.9735
Pills or steroids	0.0037	0.0119	0.092	0.0677	0.9415
Changed eating habits	0.2105	0.7505	0.9809	0.674	0.9671
Binge eating	0.0362	0.2579	0.7281	0.6997	0.8202
Fear of weight gain	0.0667	0.6603	0.9863	0.2075	0.9668

DE class membership by sociodemographic predictors

Table 4 represents the associations between DE class membership and five sociodemographic characteristics (grade, gender, race and ethnicity, family affluence, and weight status), each analyzed separately. All factors significantly predicted overall DE class membership, with gender, family affluence, and weight status showing the strongest associations. Specifically, cisgender girls were 4.74 times (95% CI: 4.15–5.41) and 15.34 times (95% CI: 12.54–18.77) more likely to belong to the "changed eating with weight concern" and "complex DE" classes, respectively, compared to cisgender boys and the "no-to-mild DE" class. TGD youth had about 8-fold (95% CI :5.72–11.16) and 15-fold (95% CI: 11.83–20.08) higher odds of belonging to the "binge eating with weight concern" and "complex DE" classes, respectively, compared to the same reference categories. Students from less affluent households were more likely to belong to the "binge eating with weight concern" (OR = 2.46; 95% CI: 2.11–2.88), "binge eating without weight concern" (OR = 2.82; 95% CI: 2.22–3.58), and "complex DE" class (OR = 5.48; 95% CI: 4.27–7.04), relative to their affluent peers and the reference class. Additionally, adolescents with very-high-weight status had increased odds of belonging to three classes with comparison to their healthy-weight peers and the reference class: 3.58, 4.87, and 6.01 times for "changed eating with weight concern" (95% CI: 2.65–4.82), "binge eating with weight concern" (95% CI: 3.70–6.41), and "complex DE" class (95% CI: 3.69–9.79), respectively.

Notably, the "binge eating without weight concern" class demonstrated a distinct sociodemographic pattern compared with other classes. In contrast to weight-concerned and "complex DE" classes, cisgender girls had lower odds of belonging to this class (OR = 0.34; 95% CI: 0.16–0.71), and adolescents with lower-weight status had higher odds of membership (OR = 1.26; 95% CI: 0.94–1.69). These divergent patterns further support the interpretation of this class as a behaviourally distinct profile rather than an extension of symptom severity.

Discussion

This study identified five distinct profiles of DE thoughts and behaviours among Canadian secondary school students and examined their associations with sociodemographic characteristics. Using LCA, we derived five classes: no-to-mild DE, changed eating with weight concern, binge eating with weight concern, binge eating without weight concern, and complex DE. Significant differences in class membership were observed across sociodemographic groups. Notably, cisgender girls and TGD youth, adolescents from less affluent households, and those with very high weight were more likely to belong to classes characterized by higher symptom engagement and multiple behaviours. These findings highlight the heterogeneity of DE experiences in youth and underscore the importance of considering sociocultural and contextual factors when interpreting DE patterns. They also emphasize the need for longitudinal research to examine developmental trajectories and to inform tailored prevention and early intervention strategies.

The five-class DE patterns align broadly with prior international research, which consistently identifies multiple, hierarchically ordered DE profiles ranging from low-risk to complex symptoms.^{Footnote 11Footnote 12Footnote 13Footnote 14Footnote 15Footnote 16Footnote 17Footnote 18Footnote 19Footnote 20} This convergence suggests that core symptom dimensions—such as weight and shape concern, binge eating, and compensatory behaviours—may reflect common underlying processes across cultural contexts. Accordingly, prevention efforts targeting shared risk factors, including body dissatisfaction and unhealthy weight-control behaviours, may have cross-national relevance. However, some differences emerged. European studies have identified more differentiated subtypes based on restraint or symptom severity,^{Footnote 14Footnote 15} while US and UK studies have highlighted gendered or muscularity-related subtypes, particularly among males.^{Footnote 12Footnote 16} In contrast, the Canadian pattern emphasized classes combining weight concern with behavioural symptoms. Some of these variations may reflect methodological differences. For example, prior studies have included items assessing muscularity-related concerns or conducted gender-stratified LCAs,^{Footnote 12Footnote 16} which may facilitate the identification of gender-specific DE patterns. Therefore, differences between studies likely reflect a combination of methodological factors and potential sociocultural variation in DE manifestation across populations.

Of the five sociodemographic factors, gender, family affluence, and weight status demonstrated the strongest associations with DE class membership. Importantly, these associations varied across specific DE profiles rather than reflecting a simple distinction between DE and non-DE. Cisgender girls were more likely than cisgender boys to belong to classes characterized by weight concerns, binge eating, and multibehaviour DE, consistent with prior research.^{Footnote 14Footnote 16Footnote 18} These gender-specific patterns likely reflect sociocultural norms emphasizing thinness for girls and muscularity for boys,^{Footnote 44} which may drive behaviours such as dieting or purging among girls and excessive exercise or supplement use among boys.^{Footnote 44} These findings support the need for gender-responsive prevention efforts that address appearance-related pressures and promote inclusive body ideals.

The inclusion of TGD youth extends the literature by demonstrating that these adolescents are at even greater risk of belonging to the complex DE class compared to cisgender girls. TGD youth may experience unique stressors related to gender incongruence, particularly when physical characteristics conflict with their gender identity.^{Footnote 45} This can contribute to psychological distress and the use of DE behaviours to modify or suppress secondary sex characteristics.^{Footnote 45} For example, transfeminine youth may restrict intake to align with feminine ideals, while transmasculine youth may pursue intensive exercise or supplement use to meet masculine standards.^{Footnote 46} Nonbinary youth may engage in restrictive eating or purging aimed at reducing gendered traits.^{Footnote 46} These findings underscore the importance of gender-affirming and tailored prevention strategies for TGD populations.

Beyond gender, socioeconomic and weight-related disparities were also evident. Adolescents who perceived their households as less affluent were more likely to belong to classes characterized by binge eating and multiple DE behaviours. Perceived lower SES may heighten vulnerability to DE through pressures for social acceptance and idealized body standards, consistent with prior research.^{Footnote 29Footnote 47} Household food insecurity may further contribute to this relationship by creating cycles of food restriction and overeating, disrupting appetite regulation and increasing risk for binge eating.^{Footnote 48} These findings highlight the importance of addressing broader structural and economic factors in DE prevention.

Our findings also revealed that adolescents with very-high-weight status were at the greatest risk of belonging to the complex DE class. This aligns with growing evidence that very-high-weight youth face elevated DE risk,^{Footnote 49} challenging longstanding assumptions that DE/ED primarily affect thin/low-weight individuals.^{Footnote 50} Such misconceptions may contribute to under-recognition of DE among very-high-weight adolescents. Very-high-weight youth are more frequently subjected to teasing and bullying from peers and family members, compared to their healthy-weight counterparts.^{Footnote 49} Experiences of weight-based stigma and discrimination may, in turn, exacerbate body dissatisfaction and promote maladaptive eating behaviours.^{Footnote 49} These results underscore the need for prevention strategies that address weight-related stigma and improve recognition of DE across diverse body sizes.

Strengths and limitations

This study has several strengths. First, it is among the largest investigations of DE patterns in Canadian adolescents, with over 24 000 participants across three provinces, allowing for robust identification of latent classes and inclusion of equity-deserving groups such as TGD and racially diverse youth. Second, the use of LCA enabled a person-centred approach to capture heterogeneity in DE patterns. Third, the inclusion of a brief DE scale, developed with youth input and demonstrating measurement invariance across subgroups, enhances methodological rigour.

Several limitations should be acknowledged. First, all measures were self-reported, introducing potential recall and social desirability bias, particularly for sensitive behaviours (purging or laxative use). Second, the adapted DE scale combined items from multiple instruments, each with known limitations in adolescent populations (e.g. brief screening tools, low-prevalence compensatory behaviours, and the absence of ARFID-related content), which may constrain the clinical interpretability of the latent DE classes while still allowing the identification of population-level symptom profiles. Third, missing data—especially for weight status—may have influenced results despite retaining missing responses as a separate category. Fourth, the sample was limited to schools in three provinces and those permitting passive consent, which may affect generalizability. Finally, regression analyses were conducted separately for each predictor due to sparse subgroup data, limiting the assessment of interactions or cumulative effects.

Conclusion

This study identified five distinct DE profiles among Canadian adolescents, revealing significant disparities in DE class membership by gender, family affluence, and weight status. Cisgender girls and TGD youth, individuals from less affluent households, and those with very high weight were at increased risk for more complex DE patterns. These findings underscore the need for gender-responsive and equity-oriented public health interventions addressing sociocultural pressures, economic disadvantage, and weight-related stigma. Future longitudinal research is warranted to elucidate developmental pathways and guide targeted prevention efforts for diverse adolescent populations.

Acknowledgements

The authors would like to thank the schools, school boards, and students that have participated in the COMPASS study, and all COMPASS team members, staff, partners, and youth engagement committee members. It takes a large team, many collaborators, and particularly, students and schools themselves, to make this study possible.

Funding

The COMPASS study has been supported by a bridge grant from the CIHR Institute of Nutrition, Metabolism and Diabetes (INMD) through the "Obesity – Interventions to Prevent or Treat" priority funding awards (OOP-110788; awarded to STL), an operating grant from the CIHR Institute of Population and Public Health (IPPH) (MOP-114875; awarded to STL), a CIHR project grant (PJT-148562; awarded to STL), a CIHR bridge grant (PJT-149092; awarded to KAP/STL), a CIHR project grant (PJT-159693; awarded to KAP), and by a research funding arrangement with Health Canada (#1617-HQ-000012; contract awarded to STL), a CIHR-Canadian Centre on Substance Abuse (CCSA) team grant (OF7 B1-PCPEGT 410-10-9633; awarded to STL), a project grant from the CIHR Institute of Population and Public Health (IPPH) (PJT-180262; awarded to STL and KAP). A SickKids Foundation New Investigator Grant, in partnership with CIHR Institute of Human Development, Child and Youth Health (IHDCYH) (Grant No. NI21-1193; awarded to KAP) funds a mixed methods study examining the impact of the COVID-19 pandemic on youth mental health, and a CIHR operating grant (UIP 178846, awarded to KAP) funds analysis of the impact of COVID-19 on health behaviours in COMPASS data. The COMPASS-Quebec project additionally benefits from funding from the Ministère de la Santé et des Services sociaux of the province of Quebec, and the Direction régionale de santé publique du CIUSSS de la Capitale-Nationale.

KAP is the Tier II Canada Research Chair in Child Health Equity and Inclusion.

Conflicts of interest

The authors declare no conflicts of interest.

Ethics approval

The current study received ethics approval from Brock University (REB#23-280). The host COMPASS study received ethics approval from the University of Waterloo (ORE#30118), Brock University (REB#18-099; 23-087), CIUSSS de la Capitale-Nationale–Université Laval (#MP-13-2017-1264) and participating school boards.

Consent to participate

All students had parental/guardian consent to participate in the study. Students attending participating schools were invited to participate using active-information passive-consent parental permission protocols. Parents were informed of the study and could withdraw their child by contacting study staff. Students not withdrawn from the study by their parents were considered eligible to participate. Under passive-consent protocols, parental informed consent is assumed for all students whose parents that did not contact COMPASS staff to withdraw their child from the study, and this pertained to all participating students.

Availability of data and material

COMPASS study data is available upon request through completion and approval of an online form. The datasets used during the current study are available from the corresponding author on reasonable request.

Authors' contributions and statement

• LZ: Conceptualization, formal analysis, writing—original draft.
• KB: Supervision, writing—review and editing.
• WP: Supervision, writing—review and editing.
• JL: Supervision, writing—review and editing.
• STL: Funding acquisition, project administration, resources, writing—review and editing.
• KAP: Conceptualization, supervision, funding acquisition, project administration, writing—review and editing.

The content and views expressed in this article are those of the authors and do not necessarily reflect those of the Government of Canada.

Previous | Table of Contents | Next