Eating Behaviors and Estimated Body Fat Percentage Among Adolescents with Type 1 Diabetes

Abstract

Aims:

Estimate associations between select eating behaviors and estimated body fat percentage (eBFP) and explore effect modification by sex among adolescents with type 1 diabetes (T1D).

Methods:

This analysis included 257 adolescents (mean age 14.9 ± 1.14 years; 49.8% female) with baseline hemoglobin A1c (HbA1c) between 8–13% (64 mmol/mol – 119 mmol/mol) from a randomized trial designed to improve glycemia. Eating behaviors and eBFP were determined from surveys and validated equations respectively. Linear mixed models were used to estimate associations. Effect modification was assessed via stratified plots, stratified associations, and interaction terms.

Results:

Disordered eating, dietary restraint, and eBFP were significantly higher among females while external eating was higher among males. Disordered eating (β: 0.49, 95%CI: 0.24, 0.73, p=0.0001) and restraint (β: 1.11, 95%CI: 0.29, 1.92, p=0.0081) were positively associated with eBFP while external eating was not (β: −0.19, 95%CI: −0.470, 0.096, p=0.20). Interactions with sex were not significant (p-value range: 0.28–0.64).

Conclusion:

Disordered eating and dietary restraint were positively associated with eBFP, highlighting the potential salience of these eating behaviors to cardiometabolic risk for both female and male adolescents. Prospective studies should investigate whether these eating behaviors predict eBFP longitudinally to inform obesity prevention strategies in T1D.

1. Introduction

Adolescence is a particularly challenging time period for type 1 diabetes (T1D) management, in terms of the transitioning glycemic and weight management responsibilities from parents to adolescents [1] and the hormonal changes that negatively affect glycemia and adiposity [2,3]. Disordered eating (i.e., symptoms such as dieting for weight loss, binge eating, and calorie purging that have not reached the diagnostic threshold for an eating disorder) often surface during this life stage and can additionally compromise T1D management and psychosocial well-being [4]. Thus, the contribution of certain eating behaviors to the cardiometabolic risk factors of glycemia and weight are important to investigate in adolescents with T1D.

Disordered eating is likely more prevalent among adolescents with T1D compared to those without due in part to the necessity of implementing day-to-day dietary strategies for glycemic management to manage a chronic disease in which insulin is administered exogenously [5–7]. In T1D, the hormonal dysregulation of insulin, glucagon, amylin, and ghrelin may disrupt hunger and satiety signaling and promote dysregulated eating patterns [5]. The heightened preoccupation with food intake resulting from glycemic management activities such as carbohydrate counting, and food restrictions may theoretically lead to food cravings and binge eating [6,8]. Furthermore, the fear of hypoglycemia and the experience of correcting hypoglycemia promote disordered eating behaviors in the context of T1D [5,6]. In a study of adults with T1D, 98% experienced disinhibited eating (i.e., eating more than intended) during perceived hypoglycemia which may lead to guilt, dietary restriction and maladaptive compensatory behaviors [9–11].

In addition to the link between disordered eating and microvascular complications due to insulin mismanagement for weight control purposes [12], those with T1D who exhibit high levels of disordered eating are more likely to have overweight and obesity [13–16], which is associated with macrovascular complications [17]. Furthermore, disordered eating and behaviors that promote positive energy balance [18] tend to persist into young adulthood due to barriers in disordered eating detection [4,8], low treatment success rate for diagnosed eating disorders [19], fear of hypoglycemia limiting physical activity [20], antagonism between glycemic management and weight loss [21], and weight-promoting external environments (i.e., obesogenic environments) [22]. Therefore, the risk of microvascular and macrovascular complications may be compounded for adolescents with T1D who experience both disordered eating and excess adiposity [17].

Several key gaps remain in our understanding of the connections between eating behaviors and adiposity in adolescents with T1D. First, body mass index (BMI) does not differentiate between lean and fat mass and is therefore prone to misclassification [23], so more direct measures of adiposity are needed. Secondly, although dietary restraint and external eating (i.e., eating in response to food stimuli, a measure of overeating tendency [24]) are prevalent in T1D [25,26] and are linked to overweight and obesity in the general population [22,27–30], no studies have extended these findings to adolescents with T1D. Thirdly, the extent to which disordered eating contributes to excess adiposity in comparison to more specific eating behaviors such as dietary restraint and external eating is unknown. Lastly, it is unclear whether the associations between these eating behaviors and adiposity differ by sex. Addressing these knowledge gaps may inform future obesity prevention efforts among high-risk groups.

Therefore, we estimated associations between three eating behaviors (disordered eating, dietary restraint, and external eating) and estimated body fat percentage (eBFP) in adolescents with T1D. Furthermore, we assessed effect modification of all associations by sex due to the sex differences in these behaviors and in adiposity (i.e., more common in females) [20,31,32]. We hypothesize that all three eating behaviors are positively associated with eBFP (disordered eating and dietary restraint more strongly associated with eBFP than external eating). Additionally, we anticipate that all associations will be stronger among females compared to males.

2. Subjects, materials, and methods

2.1. Participants

This study sample was comprised of participants from the 18-month Flexible Lifestyle Empowerment Change Trial (FLEX) which has been described elsewhere [33]. Briefly, FLEX was a multi-site randomized controlled trial (NCT01286350) designed to improve the primary outcome of hemoglobin A1C (HbA1c) through an adaptative behavioral intervention. The trial took place at the University of Colorado Barbara Davis Center for Childhood Diabetes and Cincinnati Children’s Hospital Medical Center from May 2014 to December 2017. FLEX capitalized on motivational interviewing and problem-solving skills training in addition to providing diabetes support resources (e.g., education on diet and physical activity) to promote diabetes self-management. FLEX participants (n=258) were English or Spanish speaking 13- to 16-year-olds with T1D duration of at least 1 year and baseline HbA1c between 8–13% (64 mmol/mol – 119 mmol/mol). Caregiver(s) involved in the youth’s diabetes management were also required to participate. Youth who were pregnant or had serious medical conditions (physical, developmental, or psychiatric) were excluded from the trial.

2.2. Measures

2.2.1. Participant characteristics

The FLEX study collected standardized measurements at baseline, 3-month, 6-month, 12-month, and 18-month visits. Participant characteristics were captured at the baseline, 6-month and 18-month visits. Race/ethnicity, parental income, and insulin regimen (insulin pump or injection), and tanner stage were self-reported while age, diabetes duration and sex were retrieved from medical records. Tanner stage was reported by the participant based on cartoon illustrations with corresponding text descriptions of pubic hair development from stage 1–5. Race/ethnicity was captured as Non-Hispanic White (NHW), Non-Hispanic Black, Hispanic, and other. Parental income was captured as < $25,000, $25,000 – $49,999, $50,000 – $74,999, ≥ $75,000 and declined to answer. In this analysis, race/ethnicity and income were dichotomized to NHW versus other and ≥ $75,000 versus <$75,000 respectively due to sample size limitations. We considered those who declined to report income as missing (less than 5%). Due to sparse data, tanner stages 1–5 were dichotomized to pubertal/stages 2–4 versus post-pubertal/stage 5 (no participants were in tanner stage 1).

2.2.2. Clinical characteristics

We used sex- and race- (White, Black, Mexican American, and other) specific validated equations to estimate body fat percentage [23]. The equations were developed from 1999–2006 NHANES data, validated with DXA measurements in a nationally represented sample of youth and are applicable for Americans 8 years or older [23]. The predictive model we used to estimate body fat percentage includes age, sex, race, height, weight, waist circumference and BMI. Those who identified as Hispanic in the FLEX data set were modeled with the Mexican American equation to estimate body fat percentage. Anthropometric measures were obtained by trained FLEX staff using standardized protocols at the baseline, 6-month, 18-month visits. Waist circumference was measured to the nearest 0.1 centimeter (cm) using non-tension tape measure. Height was measured to the nearest 0.1 cm using a stadiometer. Weight was measured to the nearest 0.1 kilogram (kg) with an electronic scale. BMI (kg/m²) percentile was used to determine baseline weight status (i.e., obese, overweight, normal weight). HbA1c was measured at all study visits and a central laboratory conducted all assays on specimens shipped from FLEX sites.

2.2.3. Disordered eating

Disordered eating was measured by the Diabetes Eating Problems Survey-Revised (DEPS-R, Cronbach’s α = 0.86 [34]) which includes 16 questions on a 6-point Likert scale (0 = “never” to 5 = “always”) that focuses on behaviors and feelings related to weight concerns, skipping meals or overeating, insulin restriction or other maladaptive weight loss methods, and feelings around glycemic management conflicting with weight loss or achieving thinness in the past month [34]. Responses to these questions were summed to generate a total score ranging from 0 – 80. Disordered eating was measured at the baseline, 6-month, and 18-month visit.

2.2.4. Dietary restraint and external eating

The Dutch Eating Behavior Questionnaire (DEBQ) consists of 3 subscales: dietary restraint, external eating, and emotional eating [24]. Only dietary restraint and external eating were used in this analysis. We chose to study external eating instead of emotional eating because we were more interested in externality since diabetes management may make one more susceptible to food related stimuli especially during hypoglycemia where cravings for unhealthy food and disinhibited eating (which is related to external eating [35]) is common [8,21,36]. Additionally, we had insufficient data to calculate emotional eating scores because one question was not administered during the FLEX trial.

The dietary restraint subscale was added to the FLEX protocol after the trial began as part of an ancillary study, so it was only measured at the 12-month and 18-month visits. This subscale (Cronbach’s α = 0.92) captures food restrictions arising from weight concerns [24,32]. External eating (Cronbach’s α = 0.84) was measured at the baseline, 6-month, and 18-month visits and captures the frequency of eating occasions, overeating, or having a desire to eat in response to external cues such as the sight or smell of food [24,32]. Both restraint and external eating are measured by 10 items each on a 5-point Likert scale (1 = “never” to 5 = “very often”) [24,32]. Items for each subscale were averaged to generate a mean score ranging from 1–5 [24].

2.3. Sample size and power

FLEX participants were included in this secondary analysis if they had at least one measure of disordered eating, restraint, or external eating and at least one measure of eBFP. Figure 1 shows a participant flow diagram with sample size derivations per eating behavior model. With 139 participants (our smallest analytic sample size), we had an estimated 80% power to detect a partial correlation of 0.24 between the exposures and outcome.

Figure 1:

Participant flow diagram and data availability per eating behavior.

2.4. Statistical analysis

Chi-square and student t-tests were used to compare demographic and clinical characteristics by sex for categorical and continuous variables, respectively. Since the dietary restraint model (Figure 1) used only a subset of the participants included in this secondary analysis, we conducted chi-square and student t-tests to compare demographic and clinical characteristics among adolescents with dietary restraint data versus those without.

We undertook several steps to inform the specification of the linear mixed models. We first evaluated the shape of each exposure-outcome association for potential non-linearity; no non-linearity was evident. Then we examined bivariate associations to determine which potential confounders (age, race, sex, parental income, insulin regimen and tanner stage) should be included in each model based on associations with each exposure and the outcome. Covariates associated with the exposure or outcome were included in full models for backward elimination. The study design variables of intervention group, study site and study visit number were included in all models to capture multiple visits for each participant to improve precision. We ruled out multicollinearity using the variance inflation factor (VIF) where a VIF > 10 suggests multicollinearity; VIF was < 2 in all models.

Linear mixed models (created by the MIXED procedure) were used to estimate associations between 1) disordered eating and eBFP, 2) dietary restraint and eBFP, 3) external eating and eBFP. We conducted backward elimination for each of the three exposure-outcome analyses using Akaike’s Information Criterion (AIC) and the change in the beta coefficient for the exposure. The variable was retained if the AIC increase was ≥ 20 or if the beta coefficient changed by ≥ 10% after removal from the model. All final models included study design variables and sex. In addition to those covariates, the disordered eating model controlled for income while the dietary restraint model controlled for insulin regimen and tanner stage. The external eating model also controlled for age and income.

To enable comparability of the magnitude of associations for each eating behavior, we standardized exposure variables so that all beta coefficients represent the absolute difference in eBFP based on a 1 SD increment in exposure. In an exploratory analysis, we examined effect modification by sex with stratified plots, stratified estimates of association, and by testing the statistical significance of the interaction terms between each of the three eating behaviors and sex in crude models. For this exploratory analysis, an α < 0.1 was considered statistically significant for an interaction term because we likely had insufficient power to detect significant effect modification. An α < 0.05 was considered statistically significant for all other analyses. All analyses were conducted in SAS, version 9.4 (SAS Institute, Cary, NC).

3. Results

Our sample of 257 adolescents had a mean age of 14.9 years ± 1.14, diabetes duration of 6.44 ± 3.74 years, HbA1c of 9.6 ± 1.2% (81 mmol/mol), 128 (49.8%) were female and 200 (77.8%) were Non-Hispanic White (Table 1). Nearly three-quarters of participants used insulin pumps and nearly two-thirds lived in a household with an income ≥ $75,000. Mean estimated body fat percentage (eBFP) was 20.7 ± 4.79% among males and 33.1 ± 5.36% among females at baseline (p<0.0001). Mean disordered eating, dietary restraint, and external eating scores when first measured were 13.1 ± 10.0, 1.71 ± 0.72, and 2.83 ± 0.67 respectively. Using the DEPS-R threshold of 20, which indicates a high risk for eating disorders, 22.2% (n = 57) of adolescents fall into this category with 73.7% (n = 42) of them being female. Mean dietary restraint (Females: 1.84 ± 0.78 versus Males: 1.57 ± 0.63, p = 0.02) and disordered eating scores (Females: 16.0 ± 11.3 versus Males: 10.2 ± 7.48, p<0.0001) were significantly higher among females while external eating scores (Males: 2.97 ± 0.66 versus Females: 2.69 ± 0.65, p=001) were significantly higher among males.

Table 1.

Baseline characteristics of the study sample.

Measure	All (n = 257)	Males (n = 129)	Females (n = 128)	P-values
Estimated body fat percentage %	26.9 (8.00)	20.7 (4.79)	33.1 (5.36)	<0.0001
Disordered eating	13.1 (10.0)	10.2 (7.48)	16.0 (11.3)	<0.0001
Dietary restraint n=152	1.71 (0.72)	1.57 (0.63) n=75	1.84 (0.78) n=77	0.021
External eating n=244	2.83 (0.67)	2.97 (0.66) n=122	2.69 (0.65) n=122	0.0013
Weight status	257 (100)	129 (100)	128 (100)	0.16
Normal	163 (63.4)	89 (69.0)	74 (57.8)
Overweight	59 (23.0)	26 (20.15)	33 (25.8)
Obese	35 (13.6)	14 (10.85)	21 (16.4)
Insulin regimen	255 (99.2)	128 (99.2)	127 (99.2)	0.43
Insulin pump	181 (71.0)	88 (68.75)	93 (73.2)
Injection	74 (29.0)	40 (31.25)	34 (26.8)
Age	14.9 (1.14)	14.9 (1.16)	14.8 (1.13)	0.68
Diabetes duration	6.44 (3.74)	6.71 (3.83)	6.17 (3.63)	0.24
HbA1c % [mmol/mol]	9.6 (1.2) [81]	9.64 (1.28) [82]	9.55 (1.10) [81]	0.51
Race/ethnicity	257 (100)	129 (100)	128 (100)	0.19
Non-Hispanic White	200 (77.8)	96 (74.4)	104 (81.25)
Other	57 (22.2)	33 (25.6)	24 (18.75)
Parental income	249 (96.9)	123 (95.3)	126 (98.4)	0.62
<$75,000	99 (39.8)	47 (38.2)	52 (41.3)
≥$75,000	150 (60.2)	76 (61.8)	74 (58.7)
Tanner stage	242 (94.2)	120 (93.0)	122 (95.3)	0.0007
2–4	154 (63.6)	89 (74.2)	65 (53.3)
5	88 (36.4)	31 (25.8)	57 (46.7)

Mean (standard deviation) for continuous variables; n (%) for categorical variables

P-values are from t-tests for continuous variables and chi-square tests for categorical variables

Note: Baseline mean external eating and dietary restraint values were based on the participant’s first measurement of external eating and dietary restraint (which could be at baseline, 6-, 12-, or 18-month visits) and the superscripts indicate the number of participants with these measures. Disordered eating was measured by the Diabetes eating problem survey revised. External eating and restraint were measured by the Dutch Eating Behavior Questionnaire. Body fat percentage was estimated by validated equations that included age, sex, race, height, weight, waist circumference and BMI.

3.1. Disordered eating

The unadjusted and adjusted associations between disordered eating and eBFP were both statistically significant. Each SD increment in disordered eating was associated with a 0.49 percentage point (95% CI: 0.24, 0.73, p = 0.0001) higher eBFP after adjustment (Table 2). There was no interaction between disordered eating score and sex (p = 0.29). The significant positive association detected in the full sample was also evident when stratified by sex. Each SD increment in disordered eating was associated with a 0.53 percentage point (95% CI: 0.20, 0.85, p = 0.0014) higher eBFP among males and a 0.32 percentage point (95% CI: 0.047, 0.60, p = 0.02) higher eBFP among females.

Table 2.

Overall and sex-stratified associations between eating behaviors and estimated body fat percentage.

	Unadjusted			Adjusted
	n (observations)	β (95% CI)	P-value	n (observations)	β (95% CI)	P-value
Disordered eating	257 (732)	0.40 (0.16, 0.65)	0.001	254 (698)	0.49 (0.24, 0.73)	0.0001
Male	129 (368)	0.58 (0.26, 0.90)	0.0005	127 (345)	0.53 (0.20, 0.85)	0.0014
Female	128 (364)	0.21 (−0.08, 0.50)	0.16	127 (353)	0.32 (0.05, 0.60)	0.02
Dietary restraint	152 (152)	2.37 (0.94, 3.80)	0.0013	139 (139)	1.11 (0.29, 1.92)	0.0081
Male	75 (75)	1.00 (−0.01, 2.01)	0.052	69 (69)	1.07 (0.11, 2.04)	0.03
Female	77 (77)	1.46 (0.27, 2.65)	0.017	70 (70)	0.94 (−0.37, 2.26)	0.16
External eating	244 (389)	−0.22 (−0.54, 0.10)	0.17	238 (372)	−0.19 (−0.47, 0.10)	0.20
Male	122 (190)	−0.27 (−0.61, 0.07)	0.11	117 (177)	−0.24 (−0.56, 0.08)	0.14
Female	122 (199)	−0.03 (−0.48, 0.42)	0.89	121 (195)	−0.12 (−0.55, 0.31)	0.57

Note: Beta coefficients of eating behaviors were standardized to 1 SD and should be interpreted as such (i.e., a 1 SD difference in eating behavior corresponds to a difference in estimated body fat percentage).

3.2. Dietary restraint

Dietary restraint data came only from the 18-month visit, so it’s not surprising that observations included in this model tended to be from older adolescents (16.4 ± 1.11 vs. 15.3 ± 1.26, p<0.0001) with more advanced tanner stage (70.5% of those included were in tanner stage 5 while only 36.6% of those excluded were in tanner stage 5, p<0.0001). Other demographic and clinical characteristics were similar among those included vs. not included.

The unadjusted and adjusted association between dietary restraint and eBFP were both statistically significant. Each SD increment in dietary restraint was associated with a 1.11 percentage point (95% CI: 0.29, 1.92, p = 0.0081) higher eBFP after adjustment (Table 2). The adjusted association was statistically significant among males (β = 1.07%, 95%CI: 0.11, 2.04, p = 0.03) but not among females (β = 0.94%, 95%CI: −0.37, 2.26, p = 0.16), and no significant interaction by sex was detected (p=0.64).

3.3. External eating

The association between external eating and eBFP was not statistically significant in either the unadjusted or adjusted model. There was no interaction between external eating and sex (p = 0.28), and sex-stratified associations were not statistically significant (Table 2).

4. Discussion

Disordered eating and dietary restraint, but not external eating were significantly and positively associated with eBFP among adolescents with T1D and high HbA1c. The magnitude of association for dietary restraint was roughly twice that of disordered eating (~1% versus ~0.5%). Given that dietary restraint is related to both disordered eating and obesity [32,37], it may be important to monitor dietary restraint in early adolescence to facilitate the management of blood glucose and adiposity.

Our results are consistent with other studies of adolescents with T1D reporting positive associations of disordered eating with BMI and weight status [13–16,31]. Disordered eating is likely associated positively with weight status and body fat percentage in this unique population due to a combination of factors ranging from body image dissatisfaction, negative affect, and unsuccessful dietary restraint and loss of control over eating [5], to those that are more unique to T1D such as insulin mismanagement [6,8]. Considering that a 10-point higher DEPS-R score was associated with a ~0.5 percentage point higher eBFP, the magnitude of the association was quite small. The small estimate could be explained by the heterogeneous behaviors captured in the DEPS-R that can have opposing effects on body adiposity (e.g., cyclical over- and undereating, insulin restriction for weight loss, and binge eating).

The positive association between dietary restraint and eBFP in this sample agrees with previous findings on dietary restraint and BMI among samples without diabetes and varying age groups [29,32,38,39]. According to restraint theory, long-term dietary restriction eventually leads to excessive food intake due to feelings of deprivation, cravings, and the breakdown of self-control, ultimately leading to increased adiposity [40–42]. However, it is worth noting that longitudinally studies have reported the reverse relationship where higher BMI leads to higher dietary restraint [28,29]. Prospective studies among people with T1D are needed to confirm our results and assess the directionality of the dietary restraint and eBFP association.

The null association between external eating and eBFP found in this sample contradicts our initial hypothesis that external eating would be associated with greater eBFP. Prior studies conducted in adolescents and adults without T1D have reported mixed results, including those with null [39,43], positive [22,30], and negative associations between external eating and BMI [32,44]. These conflicting results may arise from variability in participant age group (external eating tends to decrease with age [39]), country of origin, and other confounders such as parental influence on food intake [43]. Importantly, the DEBQ has not been validated among people with T1D. This instrument may therefore inaccurately capture external eating in adolescents with T1D in whom certain aspects of glycemic management could drive these behaviors (e.g., alarms that signal hypoglycemia serving as external cues for eating). Therefore, it is possible that external eating was underestimated.

4.1. Sex differences

Similar to previous studies among adolescents with T1D, disordered eating scores were higher among females than males in our sample [45,46]. Although the estimate of association between disordered eating and eBFP was greater among males compared to females in this study, the 95% confidence intervals for males and females were overlapping. The difference in the point estimate may be due to the different distributions of disordered eating between males and females; females had sparse data points above a DEPS-R score of 40, which corresponds to the maximum score among males (Figure 2a). Since the interaction between disordered eating and the female sex was not significant, we do not have evidence to suggest that the magnitude of association differed by sex.

Figure 2.

Scatterplots showing unadjusted associations of eating behaviors and estimated body fat percentage among males and females with 95% confidence limits.

Sex differences in mean dietary restraint scores were consistent with results from Snoek et al. in which female Dutch adolescents without T1D reported greater dietary restraint compared to males [32]. We detected a significant positive association between dietary restraint and eBFP in males only, but the direction of the association was similar in females. Since the interaction term was not significant, there is no evidence of a difference in association by sex. The non-significant association among females may be attributed to the greater imprecision in the effect estimate and small sample size (n = 70). This is in line with the wider range and greater variability of dietary restraint and eBFP among females compared to males (Figure 2b).

Although clear sex differences in eating behaviors and eBFP exist, there was no evidence of effect modification by sex in this sample potentially due to the truncated distribution of disordered eating and dietary restraint scores among males and small sample size. However, our disordered eating and dietary restraint results suggest that associations with eBFP are similar in direction for both males and females. Since males and females with T1D are both vulnerable to the health impacts of disordered eating, general screening and other preventative measures should be considered over sex-specific methods.

4.2. Strengths

This is the first study to examine dietary restraint and external eating from the DEBQ in relation to eBFP among adolescents with T1D. A strength of this study is the use of eBFP which is a more direct measure of adiposity and should better capture diabetes-related cardiometabolic risk compared to BMI [47]. Although BMI is commonly used to screen for obesity, it cannot distinguish lean mass from fat mass and is therefore a poor predictor of adiposity in some youth [48]. Additionally, this study assessed effect modification by sex which is a known modifier of both disordered eating and body composition. The consistent directionality of estimates in the overall and sex-stratified samples warrants further investigation in studies specifically designed to examine the association between these eating behaviors and body fat percentage.

4.3. Limitations

Since FLEX was not originally designed to answer our specific research questions, the main limitations lie in the cross-sectional interpretation of the analysis and sample size. Therefore, it is unclear whether these eating behaviors contribute to higher eBFP or the reverse. Although the effect modification analyses were likely underpowered, the direction of associations were similar between males and females. Selection bias may be a concern for the dietary restraint model due to differences in mean age and post-pubertal prevalence among FLEX participants included in this model compared to those who were not. However, the age difference likely results from the different time points in which data was collected. Since age was not a confounder in the association between dietary restraint and eBFP and there is not a large difference in body fat percentage between tanner stage 4 (i.e., last stage of puberty) and 5 (i.e., post-puberty) [49], substantial bias resulting from these differences is unlikely. As with all observational research, our analysis is subject to residual confounding. We did not control for body dissatisfaction and caloric intake as data was unavailable or very sparse.

Due to the inclusion criteria of the FLEX trial (adolescents aged 13–16 years with suboptimal glycemic control), the generalizability of our results to adolescents with T1D who meet glycemic targets may be limited. Therefore, future studies should investigate these eating behaviors and adiposity among groups with good glycemic control to confirm whether these associations hold true regardless of glycemic status. Although disordered eating behaviors typically emerge during adolescence, they are also prevalent and may be more severe among young adults, to whom these analyses should be extended [4]. Participants were mainly non-Hispanic White, of higher socioeconomic status, and willing to participate in an 18-month trial to improve diabetes self-management—factors that may limit generalizability to adolescents with diverse racial and ethnic backgrounds and who have a lower socioeconomic status. It is possible that the intervention affected eating behaviors and eBFP but unlikely as the intervention did not target eating behaviors and did not affect BMI z-score [50].

While our study evaluated the independent associations between three eating behaviors and eBFP, it is likely that high degrees of relatedness amongst these behaviors exist, suggesting that their contribution to eBFP should be jointly evaluated [31,38,40]. However, multicollinearity amongst the eating behaviors prevented us from doing so. Different analysis approaches such as cluster analysis could provide more information regarding the effect of co-existing eating behaviors (e.g., high externality and dietary restraint) on eBFP [38].

Since all exposures variables were self-reported, social desirability bias is another limitation to this analysis. Individuals with disordered eating may be more reluctant to report behaviors perceived as being maladaptive, which could lead to differential misclassification [8]. The use of self-reported race in percent body fat calculations is also a potential limitation to our analysis. Recall bias is also a common concern with questionnaire data (e.g., DEPS-R assess behaviors over a 1-month period). Since the DEPS-R does not include behaviors and attitudes that may be more relevant to males (e.g., a focus on exercise or muscularity), it is possible that disordered eating was underestimated in males compared to females. The aforementioned limitations may contribute to an underestimate of associations. Since the observed magnitudes of associations were small in this analysis, future studies are needed to assess clinical significance.

5. Conclusion

Disordered eating and dietary restraint were positively associated with eBFP in both male and female adolescents with T1D and high HbA1c, with a greater magnitude for dietary restraint. However external eating was not associated with eBFP. Prospective studies in individuals with T1D should replicate our analyses to determine the directionality of associations and explore the impacts of co-occurring eating behaviors on eBFP over time. Clinicians should consider the incorporation of early prevention measures (i.e., routine screening, monitoring, and education) addressing maladaptive eating behaviors into routine diabetes management visits given their high prevalence in T1D and their significant associations with adiposity [7]. This strategy could be an important component of preventing overt eating disorders, supporting glycemic and weight management, and minimizing the risk of downstream diabetes complications among growing adolescents soon to transition to independent diabetes management.

Data availability

Data will be shared as appropriate following the submission of a formal application to the FLEX Publications and Presentations Committee. If necessary, a data use or other data sharing agreement will be established.