Mindfulness, disordered eating, and impulsivity in relation to glycemia among adolescents with type 1 diabetes and suboptimal glycemia from the Flexible Lifestyles Empowering Change (FLEX) Intervention Trial

Abstract

Objective:

To assess the relationship between mindfulness and glycemia among adolescents with type 1 diabetes (T1D) with suboptimal glycemia, and evaluate the potential mediation by ingestive behaviors, including disordered eating, and impulsivity.

Methods:

We used linear mixed models for hemoglobin A1c (HbA1c) and linear regression for continuous glucose monitoring (CGM) to study the relationship of mindfulness [Child and Adolescent Mindfulness Measure (CAMM)] and glycemia in adolescents with T1D from the 18-month Flexible Lifestyles Empowering Change (FLEX) trial. We tested for mediation of the mindfulness-glycemia relationship by ingestive behaviors, including disordered eating (Diabetes Eating Problem Survey – Revised), restrained eating, and emotional eating (Dutch Eating Behavior Questionnaire); and impulsivity (total, attentional, and motor, Barrett Impulsiveness Scale).

Results:

At baseline, participants (n=152) had a mean age of 14.9 ± 1.1 years and HbA1c of 9.4 ± 1.2% [79±13 mmol/mol]. The majority of adolescents were non-Hispanic white (83.6%), 50.7% were female, and 73.0% used insulin pumps. From adjusted mixed models, a 5-point increase in mindfulness scores was associated with a −0.19% (95%CI −0.29, −0.08, p=0.0006) reduction in HbA1c. We did not find statistically significant associations between mindfulness and CGM metrics. Mediation of the relationship between mindfulness and HbA1c by ingestive behaviors and impulsivity was not found to be statistically significant.

Conclusions:

Among adolescents with T1D and suboptimal glycemia, increased mindfulness was associated with lower HbA1c levels. Future studies may consider mindfulness-based interventions as a component of treatment for improving glycemia among adolescents with T1D, though more data are needed to assess feasibility and efficacy.

Introduction

At the population level, measures of glycemia among US adolescents with type 1 diabetes (T1D) have not improved in the past two decades, despite substantial treatment and technological advances.^1–3 Suboptimal glycemia among adolescents with T1D (defined in various studies as a hemoglobin A1c [HbA1c] ≥9%)^4,5 is a major risk factor for both acute, life-threatening complications,⁶ and future micro-⁷ and macrovascular disease.^5,8 The burden of diabetes management is associated with an increased prevalence of disordered eating behaviors (DEBs),⁹ diabetes distress,¹⁰ and other mental health comorbidities,^11,12 which may contribute to adolescents with T1D not meeting glycemic targets.^9–12 Thus, novel and comprehensive treatment approaches that address challenges related to mental health and eating behaviors are needed to address the challenges associated with glycemic management among adolescents with T1D.

Mindfulness is a trait that involves focusing one’s attention on the present moment with a sense of non-reactivity and acceptance^13,14 and may be a promising component of interventions that aim to improve glycemic outcomes among adolescents with T1D.^15,16 Mindfulness-based interventions, such as Acceptance and Commitment Therapy (ACT),^17,18 have been shown to reduce diabetes-related psychological distress,^19,20 reduce maladaptive ingestive behaviors including DEB (e.g., external and emotional eating),^21,22 improve eating disorder (ED) symptoms,^14,23 and augment diabetes self-managment²³ among adults with diabetes. ACT may also benefit adolescents with T1D through decreased DEBs and increased impulse control ^23–26 both of which behaviors have been shown to be associated with improved diabetes treatment adherence and diabetes self-management.^24,27 However, ACT has not yet been tested and replicated in adolescents with T1D, whose unique behavioral and metabolic profiles may modify effectiveness of ACT for improving DEB, impulse control, and glycemia.^15,28,29

The high prevalence of both subclinical DEBs^9,30 and certain clinical EDs,^31,32 as well as impulsive behavior,³³ complicate glycemic management among adolescents with T1D. Preliminary evidence suggests that mindfulness-based therapies may be particularly useful for helping to manage DEB^18,22 and impulsivity³³ in adolescents with T1D for whom these behaviors are a challenge.¹⁵ DEBs unique to adolescents with T1D involve omitting or restricting insulin doses to purge calories in the form of glucose, and over-delivering insulin to permit binge-eating.³⁴ DEB and insulin omission in particular contribute to elevated HbA1c levels,³⁰ more frequent episodes of diabetic ketoacidosis,^32,35 and increased levels of neuropathy, nephropathy, retinopathy, and all-cause mortality among adolescents with T1D and DEB compared to adolescents with T1D and no DEBs.^32,36,37 Impulsivity may also contribute to difficulty with components of diabetes self-management (dietary management, glucose monitoring, insulin administration, and exercise)³³ and elevated HbA1c levels.²⁴

To address knowledge gaps relating mindfulness to DEBs, impulsivity, and glycemia among adolescents with T1D, we examined these relationships among adolescents with T1D in the Flexible Lifestyles Empowering Change (FLEX) intervention trial.³⁸ FLEX was an 18-month randomized controlled trial comparing an adaptive intervention for adolescents with T1D (intervention arm) to usual care (control arm) among adolescents with T1D and suboptimal glycemia (UC4DK101132, clinicaltrials.gov, NCT01286350). Briefly, the FLEX intervention aimed to improve diabetes self-management through motivational interviewing and problem-solving skills training supplemented with a “toolbox” comprised of T1D education, Behavioral Family Systems Therapy – diabetes, social support, and communication technology.³⁸

Our objective for this post-hoc statistical analysis was to test the hypothesis that increased mindfulness is associated with improved glycemia as assessed by both long-term (i.e., HbA1c) and short-term measures (i.e., continuous glucose monitoring [CGM] indices) and to test for mediation of the mindfulness-glycemia relationship by ingestive behaviors, including disordered eating, and by impulsivity. We aimed to generate insights that may inform future interventional studies among adolescents with T1D.

Research Design and Methods

Study sample

FLEX participants were adolescents aged 13–16 years old with a T1D diagnosis duration greater than one year, HbA1c between 8–13% [64–119 mmol/mol] and at least one participating caregiver. Individuals were excluded from the FLEX study if they were pregnant or had a serious medical condition that was physical, developmental, or psychiatric in nature. Thus, none of the participants in the study had a known, clinically diagnosed ED. The primary outcome of FLEX was HbA1c, and secondary outcomes included metrics of glycemia derived from CGM, cardiovascular disease risk factors, health-related quality of life, and cost effectiveness. The FLEX trial had a 93.4% retention rate at the 18-month follow-up visit. The FLEX intervention was found to have no significant effect on HbA1c compared to the control group; however, the intervention improved the secondary outcomes of motivation, problem-solving skills, barriers to adherence and diabetes self-management compared to the control group. ^38,39 The present analysis used data from the baseline, 12-, and 18-month visits of the FLEX trial in a subset of adolescents who participated in the ancillary study of mindfulness and impulsivity.

Ancillary study of mindfulness and impulsivity

An ancillary study of mindfulness and impulsivity was implemented partway through the FLEX trial (Figure 1). The ancillary study collected additional measures of mindfulness and impulsivity data via questionnaires at the 12- and 18-month visits for a subset of the participants. The purpose of the ancillary study was to understand the potential impact of mindfulness and impulsivity on diabetes self-management in adolescents with T1D. Of the 258 adolescents enrolled in FLEX, 80 and 152 adolescents from the FLEX study completed the ancillary study measures of mindfulness and impulsivity at 12- and 18-months, respectively.

Figure 1.

Timeline of available measures of glycemia, mindfulness, and potential mediators in the FLEX trial.

Continuous Glucose Monitoring (CGM)

FLEX collected CGM data at baseline, 6-, and 18-months of the trial. Given that the Child and Adolescent Mindfulness Measure (CAMM) was only collected at the 12- and 18-month visits, the present analyses used only 18-month CGM data. Consistent with previous FLEX analyses,^39,40 participants were included in the present analysis if they had greater than 24 total hours of CGM data over the seven-day wear period. Of the 198 participants with CGM data at 18-months, 131 also had CAMM data at 18-months. We considered using 12-month CAMM data to predict 18-month CGM data to establish temporality; however, the sample size of adolescents who had both CAMM at 12-months and CGM at 18-months was small (n=67). Therefore, we elected to use cross-sectional data for analysis of CAMM and CGM at 18-months, which was available for a larger subsample (n=131).

A blinded CGM [iPro^®2 Professional CGM; Medtronic Diabetes, Northridge, CA; mean absolute relative difference 11.1% (average error between CGM and reference values)]⁴¹ was worn for up to a seven-day period to measure interstitial glucose levels in real time. At the baseline, 6- and 18-month visits, participants inserted the iPro2^® CGM system with Enlite™ sensor into abdominal subcutaneous adipose tissue. Participants were instructed on proper use and maintenance of CGM and advised to calibrate the sensor before eating and before bed with a compatible glucometer (OneTouch Ultra 2). The Enlite™ sensor measured interstitial glucose level every five minutes within the 40–400 mg/dL range. On the last day of the CGM wear week, participants were reminded to send the devices back using the provided pre-paid envelope. CGM data were downloaded with the CareLink^™ iPro System and uploaded to the coordinating center for processing. As part of the blinding process, no communication from the CGM device was available to participants.^38,39

Measures

Laboratory data

HbA1c was measured in whole blood using an automated nonporous ion exchange High Performance Liquid Chromatography system (model G-7; Tosoh Bioscience) at a central laboratory (Northwest Lipid Metabolism and Diabetes Research Laboratories, Seattle, WA, USA).^38,39

Demographic data

Standardized questionnaires were used to collect self-reported data including age, sex, race/ethnicity, and insulin regimen. Race and ethnicity were defined based on self-report and were classified as non-Hispanic white, Hispanic, non-Hispanic Black, and other including Asian/Pacific Islander, Native American, or unknown.^38,42

Questionnaires

Mindfulness was measured with the CAMM⁴³ as part of the ancillary study of mindfulness and impulsivity. CAMM (Cronbach’s α=0.81)⁴³ is a validated 10-item self-report questionnaire designed to measure present-moment awareness and mindful acceptance in youth over the age of nine. Questionnaire items are presented on a 5-point Likert scale (0 = “never true” to 4 = “always true”). CAMM items are reverse-scored and total scores range from 0–40, with higher scores indicating increased dispositional, or trait, mindfulness.

Ingestive behaviors measured in FLEX included disordered eating (Diabetes Eating Problem Survey, Revised (DEPS-R, Cronbach’s α=0.86)⁴⁴ and the three subscales generated by the Dutch Eating Behavior Questionnaire (DEBQ).⁴⁵ The DEPS-R is a validated 16-item self-report questionnaire that measures both general and diabetes-specific disordered eating behaviors and attitudes in the past month. DEPS-R items are scored on a 6-point Likert scale (0 = “never” to 5 = “always”). Total DEPS-R scores range from 0–80 with higher scores indicating greater disordered eating. The DEBQ is a 33-item instrument comprised of three subscales used to measure present-moment emotional, external, and restrained eating (Cronbach’s α=0.92, 0.84, and 0.92, respectively).⁴⁶ The items are scored on a five-point Likert scale (1 = “never” to 5 = “always”). Mean scores are calculated for each subscale. Higher scores indicate greater levels of their respective eating behavior construct.

Impulsivity was measured with the Barratt Impulsiveness Scale (BIS-15, Cronbach’s α=0.79)⁴⁷ as part of the ancillary study of mindfulness and impulsivity. BIS-15 is a 15-item self-report questionnaire used to measure three facets of trait impulsivity (non-planning, motor, and attention impulsivity).⁴⁷ Items are scored using a 4-point Likert scale (1 = “rarely/never” to 4 = “almost/always”) and then summed to get total scores ranging from 15–60 points and individual facets from 5–20, with higher scores indicating increased impulsivity.

Outcomes

The primary outcome of this post-hoc analysis was long-term glycemia measured using HbA1c, which was available at both the 12- and 18-month study timepoints.⁴⁸ CGM data collected at 18-months was used to derive the secondary outcomes of short-term glycemia over up to seven days of wear time, including mean glucose, the percent time in range (70 –180 mg/dL), percent time in hypoglycemia (<70 mg/dL), and percent time in hyperglycemia (>180 mg/dL). The seven-day CGM collection period was determined for FLEX prior to the publication of the consensus statement recommending a 14-day wear period for optimal data sufficiency.⁴¹

Figure 1 presents a timeline of the measures of mindfulness, glycemia, and potential mediators in the FLEX trial and indicates the measures used for the present analysis. Participants provided informed consent and the study was approved by the Institutional Review Boards at each clinical site (Colorado and Ohio, USA) and the coordinating center in North Carolina.³⁸

Statistical Analysis

This post-hoc analysis used mindfulness (CAMM) as the exposure variable and glycemia (HbA1c and CGM) as the outcomes, with ingestive behaviors (DEPS-R and DEBQ) and impulsivity (BIS-15) as the potential mediators.

For descriptive analysis of the relationship of CAMM score with HbA1c (at 12- and 18-months) and CGM metrics (at 18-months), we computed mean HbA1c (%), mean glucose, percent time in range, percent time in hypoglycemia, and percent time in hyperglycemia across quartiles of CAMM scores at the relevant time points. In bivariate analyses, we tested for potential differences in each CGM metric across CAMM score quartiles at 18-months using ANOVA for normally distributed variables (mean glucose, time in range, and time in hyperglycemia) and a non-parametric Kruskal Wallis test for the non-normally distributed variable of percent time in hypoglycemia.

For multivariable analyses of both HbA1c and CGM metrics, we confirmed linearity of relationships between CAMM and each outcome by including a squared term and assessing the shape of trends using a smoothing function (SAS PROC LOESS). In presenting results, we multiplied all beta coefficients from modeled results for HbA1c and CGM metrics by a factor of 5 to represent the change in glycemia associated with a 5-point change in CAMM score, which we deemed appropriate based on the body of literature measuring mean change in CAMM scores in adolescents following mindfulness interventions.^49–56

Linear mixed models were used to test the primary hypothesis that increased dispositional mindfulness scores from CAMM are associated with improved glycemia from HbA1c. We included a repeated statement for participant ID nested within clinical site in the mixed models to account for non-independence of observations as some participants had HbA1c and CAMM data at both the 12- and 18-month visits and were therefore included in statistical models up to two times. For the CGM-based glycemia outcomes, we used linear regression on the 18-month cross-sectional data. Additionally, we conducted a sensitivity analysis of CGM data limiting the analysis to participants with greater than six days of CGM data available given that recent consensus statements encourage use of longer CGM wear periods for more accurate approximation of mean glycemia measures⁴² (n=70).

For each outcome, sequential models were constructed as follows: Model 1, unadjusted; Model 2, adjusted for potential confounders (randomization assignment, timepoint (12- or 18-months), age, race/ethnicity, sex, and insulin regimen). We created a binary variable for race/ethnicity (white and non-white) in order to increase precision to detect statistically significant effects, due to considerations of statistical power. Additionally, the covariate of insulin regimen was missing in four instances. Consistent with previous FLEX analyses,⁴² we imputed missing insulin regimen data from the nearest timepoint at which data were available.

We tested for mediation by estimating the indirect effects of potential mediators on HbA1c along with bootstrapped 95% confidence intervals.^57,58 This process required using linear mixed models (for HbA1c at 12- and 18-months) or multiple regression (for CGM metrics at 18-months) to estimate the total and indirect (i.e., mediation) effect estimates. We used the same set of adjustment covariates as for main effects models. The difference in coefficients was computed by subtracting the effect of the exposure on the outcome in the mediated model from the unmediated model. Confidence intervals were generated using bootstrapping methods. Potential mediators included ingestive behaviors (disordered eating using the DEPS-R and restrained and emotional eating using the DEBQ), and impulsivity (BIS-15 total and subscale scores). We used a significance level of 0.05 for each test; when the confidence interval contains 0, the indirect effect is not statistically significant.

Note that not all mediators—specifically DEPS-R scores and the DEBQ external eating subscale—were available at 12-months. In order to include 12-month outcome data (for HbA1c) in the mediation models, we investigated the time-stability of the potential mediators using Spearman’s correlation to determine whether missing 12-month values could be reasonably imputed from the 18-month measures. To assess time-stability, we examined the within-participant Spearman correlations between baseline and 6-months. If the correlation was above 0.7, we concluded that there was enough time-stability to impute the 12-month value with the 18-month value.

A p-value of < 0.05 was considered statistically significant for all analyses. Statistical analyses were conducted using SAS, version 9.4.

Results

Characteristics of the study sample at baseline are displayed in Table 1. Participants (n=152) had a mean age of 14.9 ± 1.1 years, 83.6% of adolescents were non-Hispanic White, 50.7% were female, and 73.0% were treated with insulin pump therapy. Mean HbA1c was 9.4 ± 1.2% [79 ± 13 mmol/mol]. At enrollment, participants included in the study sample did not differ from those who did not participate with respect to most characteristics, including intervention assignment (50.7% of those included in the analytic sample and 50.0% of those excluded were assigned to intervention; p=0.92, Supplementary Table 1), but did differ with respect to race and ethnicity (83.6% of included participants and 68.9% of excluded participants were non-Hispanic White, p=0.005). Shown in Table 2, mean mindfulness scores were 30.6 ± 7.6 and 29.2 ± 8.4, and mean HbA1c was 9.2 ± 1.2% [77 ± 13 mmol/mol] and 9.5 ± 1.4% [80 ± 15 mmol/mol] at 12- and 18-months, respectively. Mean glucose was 232 ± 43 mg/dL, mean percent time in range was 31.3 ± 14.5%, median (IQR) percent time in hypoglycemia was 0.94% (4.1), and mean percent time in hyperglycemia was 65.4 ± 16.2% at 18-months.

Table 1.

Baseline characteristics of the mindfulness and impulsivity subgroup of FLEX adolescents with T1D (n=152).

Participant demographics and clinical characteristics, mean (±SD) or n (%)
Age (years)	14.9 (±1.1)
Sex:
Male	75 (49.3%)
Female	77 (50.7%)
Race/ethnicity:
Non-Hispanic White	127 (83.6%)
Other †	25 (16.5%)
Site:
Colorado	82 (54.0%)
Ohio	70 (46.0%)
Study Group Assignment:
Intervention	77 (50.7%)
Control	75 (49.3%)
Insulin Pump Users	111 (73.0%)
HbA1c (%) [mmol/mol]	9.4 (1.2) [79 ± 13]

Abbreviations: HbA1c = Hemoglobin A1c

^†Non-white race/ethnicity includes Non-Hispanic Black/African American (3.3%), Hispanic (9.9%), and Multiracial (3.3%)

Table 2.

Descriptive statistics for exposure, outcomes, and potential mediators among FLEX adolescents at 12 and 18 months.

Mindfulness, Glycemia, Ingestive Behaviors, Disordered Eating and Impulsivity, mean (SD) or median (IQR)	12 Month (n=80)	18 Months (n=152)
Exposure:
Mindfulness (CAMM)	30.6 (7.6)	29.2 (8.4)
Outcomes:
HbA1c (%) [mmol/mol]	9.2 (1.2) [77 ± 13]	9.5 (1.4) [80 ± 15]
CGM metrics (n=131)	--	131
Mean glucose (mg/dL)		232.0 (42.7)
Time in Range (% of time)†		31.3 (14.5)
Hypoglycemia (% of time)‡		0.94 (4.1)
Hyperglycemia (% of time)§		65.4 (16.2)
Potential Mediators:
DEBQ:
Restrained eating	1.7 (0.72)	1.7 (0.78)
Emotional eating	1.7 (0.90)	1.8 (0.95)
DEPS-R	--	11.8 (10.4)
BIS-15 (total score)	30.2 (7.1)	31.6 (7.9)
Attentional impulsivity	9.8 (3.0)	10.5 (3.3)
Motor impulsivity	8.8 (3.1)	9.1 (3.3)
Non-planning impulsivity	11.6 (3.4)	12.0 (3.8)

Note: n= 152 at 18 months for all measures except CGM metrics (n=131)

Abbreviations: HbA1c, hemoglobin A1c; SD, standard deviation; CGM, continuous glucose monitoring; DEBQ, Dutch Eating Behavior Questionnaire; DEPS-R, Diabetes Eating Problem Survey – Revised; BIS-15, Barratt Impulsiveness Scale

^†Time in range - % of time spent between 70 –180 mg/dL

^‡Hypoglycemia - % of time spent below 70 mg/dL, presented as median (IQR)

^§Hyperglycemia - % of time spent above 180 mg/dL

Quartiles of CAMM were calculated separately for those with HbA1c data present (n=152) and for those with CGM data present (n=131) at 18-months of the FLEX intervention. As shown in Table 3, unadjusted HbA1c was 10.1 ± 1.6% [87 ± 18 mmol/mol], 9.5 ± 1.4% [80 ± 15 mmol/mol], 9.1 ± 1.2% [76 ± 13 mmol/mol] and 9.1 ± 1.3% [76 ± 14 mmol/mol] from the lowest to highest quartiles of CAMM, respectively (p=0.01). There were no significant differences in any CGM metrics across CAMM quartiles.

Table 3.

Measures of glycemia presented across quartiles of mindfulness (CAMM) at 18 months of the FLEX intervention.

	CAMM Quartiles				P value
CAMM quartiles among adolescents with HbA1c (n=152)	<1 (0–22.5) (n=38)	2–3 (22.6–31) (n=40)	3–4 (31–37) (n=37)	>4 (37–40) (n=37)
HbA1c (%) [mmol/mol]	10.1 (1.6) [87 ± 18]	9.5 (1.4) [80 ± 15]	9.1 (1.2) [76 ± 13]	9.1 (1.3) [76 ± 14]	0.01*
CAMM quartiles among adolescents with CGM (n=131)	<1 (0–23) (n=35)	2–3 (23–31) (n=33)	3–4 (31–37) (n=31)	>4 (37–40) (n=32)
Mean glucose (mg/dL)	240 (48)	230 (42)	227 (32)	229 (46)	0.61
Time in Range (% of time)†	30 (16)	32 (16)	31 (10)	32 (15)	0.91
Hypoglycemia (% of time): Mean (SD)‡	3 (5)	4 (5)	4 (6)	3 (4)	0.43
Hyperglycemia (% of time)§	67 (19)	64 (17)	65 (13)	65 (17)	0.90

Data are mean (SD)

Data are from FLEX adolescents who participated in the ancillary study of mindfulness. Hypothesis tests were conducted using ANOVA or Kruskal Wallis following tests of normality.

^†Time in range - % of time spent between 70–180mg/dL

^‡Hypoglycemia - % of time spent below 70 mg/dL

^§Hyperglycemia - % of time spent above 180 mg/dL

^*P<0.05

The results of models testing the primary hypothesis that increased mindfulness scores are associated with reduced HbA1c and improved CGM-based glycemia are displayed in Table 4. We did not find any evidence of non-linearity in the analyses of the relationships among both HbA1c and CGM with CAMM. In the linear mixed models using HbA1c as an outcome, the unadjusted model (Model 1) estimated a −0.22% (95% CI −0.33, −0.12, p<0.0001) change in HbA1c per five-point increase in CAMM. After adjusting for age, sex, study site, timepoint, intervention group, and insulin regimen (Model 2), the relationship remained statistically significant, with an estimated −0.19% (95% CI −0.29, −0.081, p=0.0006) change in HbA1c.

Table 4.

Modeled estimates of predicted change (Δ) in glycemia from a 5-unit increase in mindfulness; and exploration of potential mediation by ingestive behaviors, including disordered eating, and impulsivity.

Model and outcome	Mindfulness
	β (95% CI)	P value
Δ in HbA1c (%):
Model 1	−0.22 (−0.33, −0.12)	<0.0001
Model 2	−0.19 (−0.29, −0.081)	0.0006
Estimates of indirect effect
Disordered eating: DEPS-R	0.018 (−0.014, 0.0046)	NS
Emotional eating: DEBQ	0.021 (−0.012, 0.0078)	NS
Restraint: DEBQ	0.0092 (−0.0052, 0.012)	NS
Impulsivity: BIS-15	0.025 (−0.014, 0.0074)	NS
Motor impulsivity: BIS-15	0.022 (−0.015, 0.0093)	NS
Non-planning impulsivity: BIS-15	0.0065 (−0.0047, 0.0049)	NS
Attention impulsivity: BIS-15	0.0099 (−0.0097, 0.0089)	NS
Δ in Mean Glucose (mg/dL):
Model 1	−3.0 (−7.5, 1.5)	0.19
Model 2	−1.5 (−6.2, 3.2)	0.52
Δ in Time in Range (% of time) ‡ :
Model 1	0.63 (−0.92, 2.2)	0.42
Model 2	0.16 (−1.5, 1.8)	0.85
Δ in Hypoglycemia (% of time) § :
Model 1	0.026 (−0.51, 0.56)	0.92
Model 2	0.15 (−0.42, 0.73)	0.60
Δ in Hyperglycemia (% of time) ¶ :
Model 1	−0.66 (−2.4, 1.1)	0.46
Model 2	−0.31 (−2.1, 1.5)	0.73

HbA1c estimates derived from linear mixed models using participant data from 12 and 18 months (n=232 study visits from 152 participants). CGM estimates derived from linear regression models at 18 months (n=131 participants).

Model 1 unadjusted estimates (for HbA1c, includes repeated statement for subject nested within site)

Model 2 adjusted for age, sex, site, insulin regimen, timepoint (for HbA1c), and intervention group.

Indirect effects were computed by subtracting estimates with each potential mediator from the Model 2 estimate. 95% confidence intervals were computed using bootstrapping methods.

^‡Time in range:% of time spent between 70–180mg/dL

^§Hypoglycemia:% of time spent below 70 mg/dL

^¶Hyperglycemia: % of time spent above 180 mg/dL

Abbreviations: DEPS-R = Diabetes Eating Problem Survey, Revised; DEBQ = Dutch Eating Behavior Questionnaire; BIS = Barrett Impulsivity Scale; NS = Non-significant.

Additionally, Table 4 presents the results of linear regression models in which no significant relationships between CAMM and any of the CGM metrics were observed (Models 1 and 2 for each CGM metric).

Table 4 further presents the results of mediation model to test our secondary hypothesis that ingestive behaviors, including disordered eating, and impulsivity mediate the observed relationship between CAMM and HbA1c. Estimates of indirect effects for all potential mediators were not statistically significant.

Neither DEPS-R nor DEBQ was measured at 12-months, so we assessed whether imputation with the 18-month value would be appropriate. The baseline and 6-month DEPS-R scores were highly correlated (Spearman’s r=0.75); therefore, we deemed it appropriate to impute DEPS-R scores at 12-months with the 18-month value. DEBQ external eating was only available at 18-months, so we were unable to assess temporal stability, but the other DEBQ subscales were only weakly correlated six months apart (r=0.56 for emotional eating and r=0.59 for restrained eating at baseline vs. six months). We therefore excluded analysis of DEBQ external eating as a potential mediator for the HbA1c outcome.

The sensitivity analyses limited to participants with greater than six days of CGM wear time are shown in Supplementary Table 2. All results are presented per five-point increase in CAMM score. The unadjusted and adjusted models for mean glucose at 18-months revealed a change of −5.1 mg/dL (95% CI −9.4, −0.85, p=0.02) and −4.3 mg/dL (95% CI −8.9, 0.22, p=0.06), respectively. Percent time in hypoglycemia increased by 0.45% (95% CI-0.016, 0.93, p=0.058) in the unadjusted and 0.66% (95% CI 0.17, 1.2, p=0.009) in the fully adjusted model. The estimates for percent time in range and percent time in hyperglycemia were not statistically significant, but the direction of effects was consistent with those of the unrestricted analyses, which were also not statistically significant (increased percent time in range and decreased percent time in hyperglycemia).

Discussion

In this post-hoc analysis of an ancillary mindfulness study among adolescents with T1D with suboptimal glycemia, we found evidence to support that increased mindfulness is associated with lower HbA1c. The relationship between increased mindfulness and decreased HbA1c was found across quartiles of mindfulness as measured by CAMM as well as in modeled estimates. However, we did not detect a parallel association between mindfulness and CGM-based metrics (mean glucose, time in range, time in hyperglycemia or hypoglycemia) in the total study sample. Sensitivity analyses restricted to participants with six or more days of CGM data revealed that increased mindfulness was associated with reduced mean glucose and with a small increase in percent time in hypoglycemia. The non-significant degree of mediation by ingestive behaviors and impulsivity—behavioral factors posited to influence glycemia in T1D^23,24,27,33—suggests that the mindfulness-glycemia relationship is independent of these characteristics and that mindfulness may have utility as a component of diabetes self-management interventions for adolescents with T1D who have suboptimal glycemia.

Although the FLEX intervention was not a mindfulness-based intervention, our observational analyses support a limited, yet growing body of evidence which suggests that mindfulness-based interventions may offer opportunity to positively influence diabetes management in the setting of T1D.^{22,23,59–63} In one study, increased mindfulness was associated with less emotional eating in adults with both T1D and type 2 diabetes, which suggests that mindfulness may improve maladaptive eating behaviors–a potential source of dysglycemia–for individuals with diabetes. ²² Another study found that mindfulness as measured by CAMM was positively associated with diabetes-specific measures of psychological flexibility and acceptance in adolescents with T1D, but not with HbA1c.⁶⁴ Two randomized controlled trials testing mindfulness-based interventions in adolescents with T1D found reduced levels of psychological stress²⁸ and increased levels of self-efficacy (an individual’s ability to control thoughts and actions in stressful situations).²⁹ However these two small trials (n=48 across three study arms, and n=40 across two study arms) of short duration (9- and 10-weeks) either did not measure glycemia,²⁹ or showed no effect of mindfulness on glycemic levels in adolescents with suboptimal HbA1c at baseline.²⁸ Thus, our results add to the existing evidence on mindfulness and psychosocial outcomes by demonstrating an association between mindfulness and glycemic levels over a 6-month period.

Contrary to our hypothesis, increased mindfulness scores were not associated with improvements in CGM-based glycemia in our study sample of adolescents with suboptimal glycemic levels. One possible explanation for this is that up to seven days of CGM data were collected in the present study and participant data were included in the initial analysis if at least 24 total hours of CGM data were available. However, a consensus statement that was published subsequent to FLEX demonstrated that 70% of CGM data over a 14-day wear period is necessary to accurately estimate HbA1c.^41,65 The sensitivity analysis that restricted CGM data to those with greater than six days of CGM wear time (53.4% of the original analysis sample) resulted in findings for mean glucose that were reflective of the results for HbA1c. Thus, it is possible that in order to improve precision for CGM metrics, a longer data collection period for CGM data would be necessary to reveal effects of mindfulness on typical glucose excursions throughout the day.

We did not find evidence that impulsivity or disordered eating mediated the mindfulness-glycemia relationship. This was unanticipated considering previous research has observed relations between glycemia with impulsivity ^27,33 and disordered eating⁹ and has indicated a potential beneficial role of mindfulness for management of impulsive behaviors ^24,26 and disordered eating.^18,22,23 Mechanistically, mindfulness may reduce impulsivity by shifting the processing of body signals away from the dorsomedial prefrontal cortex or by promoting development of the left anterior insula both of which are brain structures involved in impulse control.^66–69 This may be particularly relevant to some adolescents with T1D who have low impulse control which may complicate diabetes self-management.³³ Prior studies have also linked mindfulness-based interventions to reduced disordered eating symptoms and improved glycemia among older adolescents and adults with T1D.²³ It is conceivable that impulsivity or disordered eating may influence glycemia independent of mindfulness. Indeed, the association of disordered eating with glycemia is well established in T1D and may operate through a multitude of behaviors, including insulin restriction relative to carbohydrate intake to promote weight loss and binge eating in response to hypoglycemia.^44,70,71 Future studies may therefore evaluate whether impulse control is an independent predictor of diabetes management outcomes.^27,33,72

Moreover, the null findings for mediation of the mindfulness-glycemia relationship by disordered eating could be due to a bi-directional relationship between mindfulness and disordered eating. This notion is consistent with a recent meta-analysis that found a similar magnitude of effect in both directions between mindfulness and eating disorder psychopathology.¹⁴ Translation of these findings to the setting of subclinical disordered eating behaviors in T1D is necessary to better understand the relationship between mindfulness and disordered eating behaviors in this unique disease state.

The present study should be interpreted in the context of its limitations and strengths. The analysis was limited to available data from the previously conducted FLEX randomized controlled trial which was not specifically designed to study the relationship of mindfulness and glycemia. Thus, our analysis does not imply causality, but rather infers an existing relationship between mindfulness and glycemia that warrants further investigation through carefully designed intervention studies, especially with respect to CGM-based metrics. FLEX participants—particularly those included in the study sample—were predominantly non-Hispanic White; therefore, our ability to extrapolate results to adolescents of color is limited. Additionally, as a diabetes self-management trial, the FLEX study enrolled adolescents with suboptimal glycemic levels, and the majority of participants were treated with insulin pump therapy. Thus, the generalizability of results to adolescents with T1D at large may be somewhat limited. The ancillary study of mindfulness and impulsivity was initiated at month 12 of the FLEX trial; therefore, the sample size of the current analysis was small relative to that of the main FLEX trial. The relatively small sample size limited the ability to capture changes to the main effect of mindfulness over time and limited the ability to assess mediation by ingestive behaviors and impulsivity. CGM-based metrics were utilized only in cross-sectional analyses, so temporality could not be established. The questionnaires that measured mindfulness (CAMM), ingestive behaviors (DEBQ) and impulsivity (BIS-15) are not diabetes-specific measures and may therefore over- or underestimate these behaviors in adolescents with T1D.

This study has several strengths. To our knowledge, this study is among the first to investigate the relationship between mindfulness and glycemia in adolescents with T1D. The study participants in the FLEX trial had suboptimal glycemic levels at baseline and multiple measurements over time---study design elements which lent themselves well to our post-hoc research questions. The analyses were adjusted for multiple potential confounding variables, including randomization assignment to the FLEX intervention group, to reduce confounding bias in our estimation of the mindfulness-glycemia relationship. The use of mixed models with data from multiple timepoints for the outcome of HbA1c allowed us to establish temporality in the relationship between mindfulness and HbA1c.

It is especially critical to design novel intervention studies targeting glycemia, in light of recent findings from the SEARCH for Diabetes in Youth study which highlight a lack of population-level improvements to HbA1c over the past two decades among youth with T1D.^2,73 The present study suggests that heightened mindfulness in the context of T1D may be an important component of improving glycemic levels in adolescents. Furthermore, the results of the present study support evidence of a relationship between increased mindfulness and improved HbA1c among adolescents with T1D with suboptimal glycemic levels that is robust to adjustment for potential confounders. Although increased mindfulness did not independently produce a clinically significant reduction in HbA1c (commonly defined as 0.5% change in HbA1c)⁷⁴ our findings imply that future research studies may integrate mindfulness into a comprehensive intervention to improve glycemia among adolescents with T1D. Additional studies of longer duration are needed to establish causality in the relationship between mindfulness and glycemia among adolescents with T1D, and to determine whether mindfulness-based interventions can feasibly and effectively be implemented into clinical settings as part of an integrative care plan for improving glycemia among adolescents with T1D.