Weight Management in Young Adults with Type 1 Diabetes: The Advancing Care for Type 1 Diabetes and Obesity Network Sequential Multiple Assignment Randomized Trial Pilot Results

Abstract

Aims

Co-management of weight and glycemia is critical yet challenging in type 1 diabetes (T1D). We evaluated the effect of a hypocaloric low carbohydrate, hypocaloric moderate low fat, and Mediterranean diet without calorie restriction on weight and glycemia in young adults with T1D and overweight or obesity.

Materials and Methods

We implemented a nine-month Sequential, Multiple Assignment, Randomized Trial pilot among adults aged 19–30 years with T1D for ≥1 year and BMI 27–39.9 kg/m². Re-randomization occurred at 3- and 6-months if the assigned diet was not acceptable or not effective. We report results from the initial three-month diet period and rerandomization statistics prior to shutdowns due to COVID-19 for primary (weight, hemoglobin A1c [HbA1c], percent of time below range [%TBR] <70 mg/dL) and secondary outcomes (body fat percentage [BFP], percent of time in range [70–180 mg/dL], and %TBR <54 mg/dL). Models adjusted for design, demographic, and clinical covariates tested changes in outcomes and diet differences.

Results

Adjusted weight and HbA1c (n=38) changed by −2.7 kg (95%CI −3.8, −1.5, p<0.0001) and −0.91 percentage points (95%CI −1.5, −0.30, p=0.005), respectively, while adjusted BFP remained stable, on average (p=0.21). Hypoglycemia indices remained unchanged, on average, following adjustment (n=28, p>0.05). Variability in all outcomes, including weight change, was considerable (57.9% were re-randomized primarily due to loss of <2% body weight). No outcomes varied by diet.

Conclusions

Three months of a diet, irrespective of macronutrient distribution or caloric restriction, resulted in weight loss while improving or maintaining HbA1c levels without increasing hypoglycemia in adults with T1D.

Plain Language Summary

Young adults with type 1 diabetes (T1D) face unique metabolic and behavioral challenges with managing their weight and blood glucose—potentially competing outcomes—but very few clinical trials have tested which diets are safe, feasible, and acceptable for people with T1D. Because the effectiveness of a given diet may vary across individuals, we implemented an adaptive trial which allowed young adults with T1D and overweight or obesity to try up to three randomized diets (low-calorie low carbohydrate, low-calorie low fat, and Mediterranean without calorie restriction) to help manage their weight and blood glucose over nine months. COVID-19 interrupted our study, so we report results from the first three-month diet period pre-COVID-19. On average, participants lost weight and lowered their hemoglobin A1c without increasing hypoglycemia, assessed by continuous glucose monitoring. There were no differences in the three-month change in study outcomes by diet assignment. However, there was wide variability in outcomes, overall and within each diet group. We will use these results to help power an adaptive efficacy trial, which can identify individual-level predictors of dietary response in adults with T1D.

Introduction

Over two-thirds of adults with type 1 diabetes (T1D) now have overweight or obesity (1, 2). This proportion has increased by two-fold in the past three decades, and nearly seven-fold for the prevalence of obesity specifically—from 3.4% in 1986–1988 to 22.7% in 2007 (3) and to 36.8% in 2018 (1). When contextualized against the background of a three-to-eight times greater baseline risk of cardiovascular disease among people with T1D compared to those without diabetes (4), this raises serious concern about the added contribution of obesity to cardiometabolic disease in this at-risk population (4–7).

The literature is lacking in T1D-specific weight management trials that assess the effectiveness of diets that vary markedly in macronutrient composition for weight loss. Critically, weight loss strategies created specifically for people with T1D must demonstrate safety in terms of limiting one’s exposure to hypoglycemia, hyperglycemia, and diabetic ketoacidosis in the challenging setting of exogenous insulin therapy. Furthermore, T1D-specific challenges must be addressed: the need to ingest rescue carbohydrates to treat iatrogenic hypoglycemia, barriers to safe exercise (e.g., the fear of hypoglycemia), and the anabolic nature of insulin may all counter weight loss efforts (8, 9). Accordingly, our main objective was to implement a pilot feasibility study using a rigorous Sequential, Multiple Assignment, Randomized Trial (SMART) design to identify acceptable and effective dietary strategies to co-optimize weight and glycemia in young adults with T1D.

SMARTs are vital tools in biomedical research: they efficiently address practical treatment comparison questions and adapt dynamically based on participant responses (10, 11). Using a SMART pilot design, we delivered three evidence-based dietary approaches that produce weight loss in adults with type 2 diabetes (12–14) and vary substantially in macronutrient composition and in whether calories are explicitly restricted: a hypocaloric low carbohydrate diet, a hypocaloric Look AHEAD (i.e., moderate low fat) diet, and a healthy Mediterranean diet (not calorically restricted). We hypothesized that each diet would lead to weight loss and glycemic maintenance, but that variability in outcomes would be evident, creating the need to tailor the treatment to the individual. The results of this SMART pilot alongside ongoing and future studies will be used to develop a fully powered efficacy trial to co-optimize weight and glycemia and identify predictors of dietary response among adults with T1D.

Materials and Methods

The design of the Advancing Care for Type 1 Diabetes and Obesity Network (ACT1ON) study has been described elsewhere (15). Briefly, ACT1ON was a nine-month pilot feasibility and acceptability study conducted at the University of North Carolina at Chapel Hill (UNC) and Stanford University to identify acceptable and effective dietary strategies (hypocaloric low carbohydrate, hypocaloric Look AHEAD, or Mediterranean diet without calorie restriction) to co-optimize weight and glycemia among adults with T1D aged 19–30 (1DP3DK113358, NCT03651622). Primary outcomes were change in weight, hemoglobin A1c (HbA1c) level, and percent of time below range (<70 mg/dL) (16) as assessed by continuous glucose monitoring (CGM) at baseline and following each of three, three-month, dietary periods. Secondary outcomes were change in percent body fat as assessed by dual-energy x-ray absorptiometry (DXA), percent of time in target glucose range (70–180 mg/dL), and percent of time below range (<54 mg/dL) (16). Additional measurements that were analyzed descriptively included the change in kg of body fat, and, per CGM consensus guidelines, the changes in percent of time above range (181–250 mg/dL and, separately, ≥250 mg/dL), as well as the coefficient of variation (CV) (16).

Young adults meeting all eligibility criteria according to medical record data were enrolled between November 12^th, 2018, and March 6^th, 2020. On March 27^th, 2020, in response to COVID-19, the study transitioned to a virtual format via a HIPAA-secure video conferencing platform (Zoom Video Communications Inc., San Jose, CA), including dietary counseling and data collection, and recruitment ceased (68 of the anticipated 72 participants were enrolled). Given the need to alter the measurement methods for the primary outcomes of weight and glycemia during COVID-19, results of the COVID-19-influenced periods of the study will be reported in a separate manuscript. Here, we report results from the first three-month diet period, for visits that were completed pre-COVID-19. The final study visit was completed on February 22^nd, 2021.

Study Sample

This study was approved by the UNC and Stanford University IRBs and study participation began after participants signed informed consent. Participants were adults aged 19–30 years at baseline with T1D for ≥1 year, literate in English, an HbA1c <13.0% (<119 mmol/mol), and a body mass index (BMI) of 27.0–39.9 kg/m². Participants were excluded if they had a history of a diagnosed eating disorder, gastrointestinal or bowel disorder, dietary restrictions that precluded following the study diets, were pregnant or lactating, had any episode of diabetic ketoacidosis or hypoglycemia requiring third-party assistance in the prior six months, or were weight unstable (change of ±10 lbs in the prior six months).

Sequential Randomization

Permuted block randomization stratified by site and confidentially generated by study statisticians was used to assign 12 potential treatment regimens (i.e., diet sequences, R statistical software). Registered Dietitian interventionists revealed the diet assignment to participants at the time of each randomization. Following an initial randomization at enrollment, the SMART design adapted dynamically to participant responses by re-randomizing those for whom the assigned diet was not acceptable or not effective to an alternate diet based on a priori decision rules at months ~3 and ~6 of the intervention (10). We emphasize that here, we report the results of the initial randomization and ~3-month re-randomization for participants who completed the first diet period pre-COVID-19. The re-randomization criteria included clinical outcomes (<2% weight reduction [unless weight loss resulted in a BMI <25 kg/m²], HbA1c increase ≥0.5%, and self-reported increased or problematic hypoglycemia) and self-reported diet unacceptability.

Measurements

Study measurements were taken at baseline and following each three-month diet period (Supplemental Figure 1). We report the baseline and ~3-month change in each measure for visits that were completed prior to COVID-19.

Anthropometrics

Height (nearest 0.1 cm), weight (nearest 0.1 kg using a calibrated clinic scale), and waist circumference (nearest 0.1 cm) were measured following standard procedures.

Glycemia

Blood was drawn at each measurement visit and sent to a central laboratory (Northwest Lipid Research Laboratory, WA) for determination of HbA1c using high-performance liquid chromatography.

Participants wore a blinded CGM (Freestyle Libre Pro, Abbott Diabetes Care Inc., CA) for up to 14 days (range 1–14 days, median 13.3 days [IQR 8.5, 14.0]) at baseline and each subsequent measurement visit. Study staff inserted CGM sensors, which were used to compute percent of time below range (TBR: <70 mg/dL, <54 mg/dL), percent of time in target range (TIR, 70–180 mg/dL), percent of time above range (TAR: 181–250 mg/dL, >250 mg/dL), and CV, according to consensus guidelines (16). A comparison of CGM metrics among participants with <8.5 days (i.e., less than quartile 1) of CGM wear time and all included participants is shown in Supplemental Table 1.

Body composition

Body fat percentage was measured via DXA (UNC: GE Lunar iDXA, GE Medical Systems Ultrasound & Primary Care Diagnostics, WI; Stanford University: Horizon Model A, Hologic, MA).

Dietary intake

At baseline and each measurement visit, 24-hour dietary recalls were administered via telephone by trained UNC NIH/NIDDK Nutrition Obesity Research Center (NORC, P30DK056350) staff using a multi-pass method (17). Given that the objective is to estimate usual intake (18), recalls deemed to be unreliable by NORC staff or for which participants indicated the amount consumed was “a lot more” or “a lot less” than usual were excluded from analysis. Nutrition Data System for Research Version 2019 (University of Minnesota, MN) (19) was used to derive nutrients associated with recalled foods and beverages.

Demographics and health history

Participants reported demographic (age, gender, race, ethnicity) and clinical data (diagnosis date, insulin regimen [once, twice, three times, or more than three times daily; or insulin pump], and medical history) using standardized questionnaires. Insulin regimen was captured at baseline and updated at each measurement visit. Missing data for insulin regimen (n=3 visits) were imputed from the closest visit in time.

Dietary interventions

Registered Dietitians provided guidelines about general healthy eating and specific dietary recommendations, employing motivational interviewing and problem-solving skills training aimed at overcoming barriers to adherence (20, 21). The intervention consisted of eight full-length counseling and education sessions, and 15 shorter “check-in” sessions. Check-in sessions focused on goal setting, weight monitoring, self-efficacy, and readiness to change; and Registered Dietitians queried participants about hypoglycemia events. Per Look AHEAD study procedures, the energy goals for individuals who weighed <114 kg (250 lbs) and ≥114 kg were 1200–1500 kcal/day and 1500–1800 kcal/day, respectively (13), for the low carbohydrate and Look AHEAD diets. Registered Dietitians re-calculated calorie prescriptions at each measurement visit to incorporate current body weight. Counseling strategies related to carbohydrate counting for insulin dosing and encouragement of usual physical activity were consistent across diets.

To promote adherence (22) and inform dietary counseling sessions with Registered Dietitians, participants voluntarily tracked dietary intake and body weight using the MyFitnessPal smartphone app (San Francisco, CA) and BodyTrace© scale (Palo Alto, CA), respectively.

Low carbohydrate diet

Although we planned to implement a diet with <14% of calories from carbohydrate (12), concerns about safety (i.e., the potential for excessive hypoglycemia if insulin was not reduced accordingly or ketosis if too little insulin was administered) and adherence led us to increase the carbohydrate goal prior to study implementation to 15–20% of calories from carbohydrate (45–75 grams/day depending on energy requirements) and <10% of fat from saturated fat (12).

Look AHEAD diet

Recommended macronutrient distributions were <30% of calories from fat and <10% of fat from saturated fat, as per the original Look AHEAD study procedures. In contrast to the Look AHEAD study, there was flexibility related to percent of calories from carbohydrates and protein (13).

Mediterranean diet

Dietary guidance mirrored the PREDIMED trial. Dietary guidance emphasized abundant use of olive oil; increased consumption of plant-based foods, whole grains, and fish; reduced red meat consumption; avoidance of high-sugar baked goods and sugar-sweetened beverages; and moderate consumption of red wine among those who drink alcohol (14).

Statistical Analysis

Based on prior studies (23, 24) and our originally planned sample size of n=72, we estimated with 80% power that we could detect a ~3.0 kg weight change, and a 1.3 percentage point change in HbA1c and in percentage points of time <70 mg/dL.

We compared the three-month responses in primary and secondary outcomes overall and to each of the three initial diets at the end of diet period 1 (pre-COVID-19), following the original study protocol. Of major concern leading to this decision was the need to change our measurement methods for the co-primary outcomes of weight and HbA1c, and the substantially disrupted cadence of diet period intervals caused by the several months required to establish the virtual protocol including IRB approvals at both clinical sites. The sample sizes for the second and third diet periods (n=14 and n=5, respectively) were extremely limited by the time of clinic shut-down due to COVID-19. Using standard descriptive statistics (categorical variables: Chi-squared or Fisher’s exact test; continuous variables: ANOVA or Kruskal Wallis), we compared the baseline characteristics of participants included (n=38) or excluded (n=30) from the analysis.

We tested for 3-month within-group change in outcomes using paired t tests for normally distributed variables or the Wilcoxon signed rank test for non-normally distributed variables. We applied ANOVA or the Kruskal Wallis test to evaluate differences across the three diets in the 3-month change in outcomes for normally or non-normally distributed variables, respectively. Crude and adjusted general linear models were used to analyze changes in primary and secondary outcomes. Model 1 estimated changes in outcomes irrespective of diet and included study site, the baseline level of each outcome, and diet duration (months). To facilitate parsimony of Model 1, for each outcome, we included gender, age, and race and ethnicity when the p-value was <0.1 for each term. Based on this criterion, gender was included in Model 1 for all outcomes. Model 1 for HbA1c and percent of time in range was additionally adjusted for race and ethnicity (dichotomized due to sample size limitations). The terms for gender and race and ethnicity in Model 1 assessed whether outcomes were influenced by these demographic covariates, as we lacked power to formally test for interaction. Model 2 further included diet assignment and tested for diet differences when the term for diet assignment was statistically significant.

We evaluated adherence using 24-hour dietary recalls. We assessed change in percent fat, percent carbohydrate, or grams of fiber intake on the low carbohydrate, Look AHEAD, and Mediterranean diets, respectively, and the change in total calories on the low carbohydrate and Look AHEAD diets.

We truncated (winsorized) outlier observations that had undue influence on the results but were not the result of measurement error to 10 percent greater than the next greatest absolute value (25). We did this for one observation for 3-month change in percentage points of TIR (crude change of 56.6% was winsorized to 34.8%).

All analyses were conducted using SAS version 9.4 (SAS Institute, Cary, NC). An alpha level of p<0.05 was considered to be statistically significant.

Results

Baseline characteristics

Of the n=38 participants who completed the first three-month diet period pre-COVID-19, n=16 were initially randomized to the low carbohydrate diet, n=12 to the Look AHEAD diet, and n=10 to the Mediterranean diet. All baseline characteristics were balanced across treatment arms (Table 1) and did not differ according to whether participants were included or excluded from the pre-COVID-19 diet period 1 analysis (Supplemental Table 2).

Table 1.

Baseline Characteristics of ACT1ON Participants Who Completed Diet Period 1 Pre-COVID-19

	All (n=38)	Low carb (n=16)	Look AHEAD (n=12)	Mediterranean (n=10)	p-value
Demographics
Age (years)	26.1 (23.6, 27.2)	24.7 (23.2, 27.9)	26.3 (24.0, 26.6)	26.8 (24.4, 29.5)	0.61
Female gender	30 (79.0)	11 (68.8)	9 (75.0)	10 (100.0)	0.20
Other race and ethnicity (collapsed)†	13 (34.2)	4 (25.0)	3 (25.0)	6 (60.0)	0.16
Clinical characteristics
Insulin pump use	23 (60.5)	10 (62.5)	8 (66.7)	5 (50.0)	0.7
BMI (kg/m2)	30.5 (28.2, 34.4)	30.0 (28.4, 33.4)	31.9 (28.7, 35.0)	29.7 (27.4, 31.9)	0.66
CGM-Related Outcomes
	All (n=35)	Low carb (n=14)	Look AHEAD (n=12)	Mediterranean (n=9)	p-value
%TBR (<70 mg/dL)	5.1 (1.7, 9.2)	6.0 (2.4, 10.4)	4.2 (1.6, 6.0)	4.8 (1.1, 8.1)	0.42
%TBR (<54 mg/dL)	0.88 (0.26, 3.0)	1.6 (0.73, 3.1)	0.59 (0.09, 1.9)	0.69 (0.0, 2.3)	0.24
%TIR (70–180 mg/dL)	51.4 ± 24.5	55.3 ± 24.6	51.7 ± 22.6	44.7 ± 27.9	0.61
% TAR (181–250 mg/dL)	19.7 ± 8.7	18.7 ± 11.5	19.0 ± 5.2	21.8 ± 7.3	0.73
% TAR (>250 mg/dL)	9.0 (2.2, 4.3)	5.2 (0.19, 38.9)	8.3 (2.5, 19.7)	29.9 (4.7, 51.6)	0.28
CV (%)	37.3 ± 6.5	37.2 ± 6.8	38.2 ± 6.0	36.4 ± 7.3	0.86
24-hr recall diet data
	All (n=33)	Low carb (n=14)	Look AHEAD (n=10)	Mediterranean (n=9)	p-value
Calories	1773.4 ± 502.2	1820.4 ± 534.2	1928.4 ± 510.5	1528.3 ± 388.1	0.20
Calories from carbohydrate (%)	37.4 ± 12.6	33.9 ± 9.9	39.0 ± 12.5	41.1 ± 16.1	0.38
Calories from fat (%)	39.1 ± 10.6	37.8 ± 10.0	41.7 ± 10.7	38.4 ± 12.0	0.66
Calories from protein (%)	19.3 ± 6.6	21.2 ± 7.9	18.2 ± 4.8	17.4 ± 6.0	0.33
Fiber (g)	16.2 ± 8.4	16.9 ±8.7	15.5 ± 7.6	15.9 ± 9.6	0.92
Healthy Eating Index Score	54.8 ± 13.6	55.3 ± 13.0	52.7 ± 13.3	56.5 ± 16.2	0.83

^†Race: 13.2% African American, 2.6% Asian, 2.6% Native Hawaiian/Pacific Islander, 2.6% More than one, 2.6% Other, 68.4% White, 7.9% Missing. Hispanic ethnicity: 18.4%. Continuous variables presented as mean ± standard deviation or as median (quartile 1, quartile 3). Categorical variables are presented as N (%). Diet groups compared using ANOVA or Kruskal Wallis for continuous variables, and the Chi-squared or Fisher’s exact test for categorical variables.

Abbreviations: BMI—body mass index; CV—coefficient of variation; kg—kilograms; HbA1c—hemoglobin A1c; TAR—time above range; TBR—time below range; TIR—time in range; g—grams

Study retention, fidelity, and safety

Thirteen of the 51 participants scheduled to complete diet period 1 pre-COVID-19 dropped out (seven were withdrawn and six were lost to follow-up, retention: 38/51=74.5%). The number of dropouts did not vary by diet (p=0.067 for Fisher’s exact test) but was numerically smaller for the low carbohydrate diet (n=1) than the Look AHEAD or Mediterranean diets (n=6 each).

Fidelity of intervention delivery (the proportion of standardized intervention components delivered during each reviewed intervention session) was 97.0%.

One hypoglycemic episode was reported but was reportedly related to the participant’s acute use of a steroidal medication and therefore deemed to be unrelated to the study. No other adverse events were related to the study.

Re-randomization statistics

Twenty two of 38 participants (57.9%) were identified for re-randomization at three-months according to the study decision rules. The proportion of participants re-randomized did not vary by diet assignment (Chi-squared p=0.59), nor did the reasons for re-randomization (Fisher’s exact p=0.16–1.0). The most common reason for re-randomization was not losing ≥2% of body weight (n=17, 77.3% of those re-randomized), followed by the diet not being acceptable from a personal preference or adherence perspective (n=13, 59.1%), an HbA1c increase ≥0.5% (n=4, 18.2%), and self-reported hypoglycemia (n=2, 9.1%) (categories not mutually exclusive).

Primary outcomes

Weight

Crude three-month weight change was −2.0 kg (95% CI −3.0, −1.1, p<0.0001) (n=38, Table 2, variability shown in Figure 1A). Model 1 adjusted weight change was −2.7 kg (95% CI −3.8, −1.5, p<0.0001, Table 3) and −3.0% (95%CI −4.3, −1.7, p<0.0001), respectively, and did not differ across diet assignments (Model 2 p=0.34 for diet difference). According to the Model 1 term for gender, women lost 2.6 (95%CI 0.41, 4.9, p=0.02) fewer kilograms of weight than men.

Table 2:

Three-Month Pre-COVID-19 Diet Period 1 Intervention Effects—Overall and by Randomized Diet Assignment

Primary Outcomes
	All (n=38)	Diet Assignment (Low carbohydrate as P-value Reference)
		Low carb (n=16)	Look AHEAD (n=12)	Mediterranean (n=10)	p-value†
Weight (kg)
Baseline	87.4 ± 12.9	87.2 ± 10.8	90.5 ± 16.5	84.1 ± 11.5	0.53
~3-months	85.4 ± 12.7	84.8 ± 11.8	89.2 ± 15.5	81.7 ± 10.2	0.39
~3-month change	−2.0 (−3.0, −1.1), p<0.0001	−2.4 (−3.9, −0.93)	−1.3 (−3.4, 0.82)	−2.4 (−4.3, −0.52)	0.55
HbA1c (%)
Baseline	7.9 ± 1.5	7.3 ± 1.6	8.3 ± 1.3	8.3 ± 1.4	0.14
~3-months	7.8 ± 1.5	7.3 ± 1.6	7.7 ± 1.0	8.6 ± 1.7	0.09
~3-month percentage point change	−0.13 (−0.39, 0.12), p=0.29	−0.04 (−0.25, 0.17)	−0.65 (−1.1, −0.18), p=0.027	0.34 (−0.34, 1.0), p=0.18	0.006
HbA1c (mmol/mol)
Baseline	63 ± 16.4	56 ± 17.5	67 ± 14.2	67 ± 15.3	0.14
~3-months	62 ± 16.4	56 ± 17.5	61 ± 10.9	70 ± 18.6	0.09
~3-month change	−1.4 (−4.3, 1.3), p=0.29	−0.4 (−2.7, 1.9)	−7.1 (−12.0, −2.0), p=0.027	3.7 (−3.7, 10.9), p=0.18	0.006
	All (n=28)	Low carb (n=12)	Look AHEAD (n=8)	Mediterranean (n=8)	p-value†
% Time below range (<70 mg/dL)
Baseline	5.0 (1.9, 9.8)	6.0 (2.9, 12.3)	4.3 (1.6, 8.2)	3.5 (0.73, 6.9)	0.29
~3-months	5.3 (2.2, 9.8)	9.1 (4.5, 11.4)	3.1 (1.5, 7.6)	3.3 (0.43, 6.5)	0.04
~3-month percentage point change	0.42 (−1.7, 2.6), p=0.69	1.3 (−2.1, 4.6)	0.23 (−3.9, 4.4)	−0.64 (4.8, 3.5)	0.76
Secondary outcomes
	All (n=38)	Low carb (n=16)	Look AHEAD (n=12)	Mediterranean (n=10)	p-value†
Body fat (%)
Baseline	42.2 (34.7, 45.2)	42.1 (29.4, 45.2)	42.8 (37.7, 45.3)	40.7 (34.8, 45.2)	0.83
~3-months	40.6 (33.8, 44.7)	40.5 (27.9, 44.2)	43.7 (37.3, 45.5)	40.1 (35.9, 43.2)	0.52
~3-month percentage point change	−0.53 (−1.0, −0.03), p=0.04	−0.87 (−1.7, −0.05)	−0.058 (−0.83, 0.72)	−0.57 (−1.9, 0.75)	0.39
Body fat (kg)
Baseline	34.4 ± 9.8	32.6 ± 10.6	37.1 ± 9.6	34.1 ± 8.8	0.49
~3-months	33.3 ± 10.0	31.0 ± 11.2	36.7 ± 9.9	32.8 ± 7.6	0.33
~3-month change	−1.1 (−1.9, −0.40), p=0.003	−1.6 (−2.8, −0.38)	−0.36 (−1.6, 0.83), p=0.76	−1.3 (−3.1, 0.44), p=0.15	0.34
	All (n=28)	Low carb (n=12)	Look AHEAD (n=8)	Mediterranean (n=8)	p-value†
% Time in range (70–180 mg/dL)
Baseline	52.6 ± 26.0	55.8 ± 25.5	58.2 ± 24.2	42.3 ± 28.8	0.42
~3-months	54.8 ± 22.9	58.0 ± 24.8	63.6 ± 17.0	41.4 ± 21.1	0.12
~3-month percentage point change	2.2 (−3.4, 7.8), p=0.43	2.1 (−6.7, 11.0)	5.4 (−8.7, 19.6)	−0.88 (−12.6, 10.8)	0.70
% Time below range (<54 mg/dL)
Baseline	0.82 (0.19, 2.8)	1.6 (0.73, 3.1)	0.21 (0.04, 1.9)	0.71 (0.0, 2.9)	0.63
~3-months	1.2 (0.19, 3.3)	1.6 (0.63, 3.8)	0.37 (0.0, 3.7)	1.2 (0.0, 2.9)	0.45
~3-month percentage point change	0.00 (−1.2, 1.5), p=0.68	0.67 (−1.1, 2.5)	0.57 (−1.6, 2.8)	0.18 (−2.0, 2.4)	0.93
Additional CGM indices
% Time above range (181–250 mg/dL)
Baseline	19.7 ± 8.7	18.7 ± 11.5	19.0 ± 5.2	21.8 ± 7.3	0.73
~3-months	21.3 ± 12.0	18.0 ± 11.8	17.1 ± 8.2	30.3 ± 11.7	0.03
~3-month percentage point change	0.32 (−2.9, 6.5), p=0.61	−0.53 (−7.4, 5.9)	−1.1 (−2.9, 0.68)	6.6 (1.2, 13.8)	0.07
% Time above range (>250 mg/dL)
Baseline	9.0 (2.2, 4.3)	5.2 (0.19, 38.9)	8.3 (2.5, 19.7)	29.9 (4.7, 51.6)	0.28
~3-months	9.8 (2.5, 27.5)	7.1 (0.57, 18.9)	8.5 (4.3, 23.5)	22.8 (8.1, 35.5)	0.26
~3-month percentage point change	−4.2 (−11.5, 3.2), p=0.25	−2.7 (−12.6, 7.2)	−3.6 (−18.1, 10.9)	−7.0 (−28.9, 14.9)	0.89
CV (%)
Baseline	37.3 ± 6.5	37.2 ± 6.8	38.2 ± 6.0	36.4 ± 7.3	0.69
~3-months	38.1 ± 6.9	37.6 ± 8.2	40.4 ± 6.6	36.4 ± 4.9	0.50
~3-month change	0.46 (−2.4, 3.1), p=0.67	0.27 (−1.9, 3.2)	−0.14 (−2.8, 5.5)	1.8 (−3.0, 2.3)	0.99

Baseline values are mean ±SD or median (IQR). Change values are mean (95% confidence interval) or median (IQR). Within-group p-values are from paired t tests for normally distributed variables or the Wilcoxon signed rank test for non-normally distributed variables.

^†P-values for diet difference in the rightmost column were calculated using ANOVA for normally distributed variables or the Kruskal Wallis test for non-normally distributed variables.

Abbreviations: CV—coefficient of variation; Low carb—Low carbohydrate; mg/dL—milligrams per deciliter; HbA1c—hemoglobin A1c.

Figure 1.

Variability of ~3-month changes in the co-primary outcomes of weight (kg, A) and hemoglobin A1c (percentage points, B) (n=38 for both outcomes), displayed by diet assignment. Abbreviations: kg—kilograms; Low carb—Low carbohydrate

Table 3:

Modeled changes in primary and secondary study outcomes for pre-COVID-19 diet period 1

	Model 1 Adjusted β (95% CI)	Model 2 p-value for diet difference
Primary Outcomes (~3-month change)
Weight (kg) (n=38)	−2.7 (−3.8, −1.5) p<0.0001	0.34
HbA1c (n=38): percentage point	−0.91 (−1.5, −0.30) p=0.005	0.09
HbA1c (n=38): mmol/mol	−9.9 (−16.4, −3.3) p=0.005	0.09
Percentage points of time below range (<70 mg/dL) (n=28)	1.4 (−1.0, 3.7) p=0.25	0.27
Secondary Outcomes (~3-month change)
Body fat percentage points (n=38)	−0.59 (−1.5, 0.35) p=0.21	0.51
Percentage points of time in range (70–180) (n=28)	12.4 (−2.3, 27.2) p=0.09	0.46
Percentage points of time below range (<54 mg/dL) (n=28)	0.84 (−0.55, 2.2) p=0.23	0.93

All Model 1 β estimates were adjusted for gender, study site, the baseline level of the outcome, and diet duration (months). Model 1 for HbA1c and percent of time in range was additionally adjusted for race and ethnicity. Paired diet differences are not shown given that the term for diet assignment from the adjusted general linear Model 2 was not statistically significant for any outcome.

Hemoglobin A1c

Although crude HbA1c change was not statistically significant (p=0.29), there was considerable inter-individual variability (Figure 1B) and a statistically significant difference by diet (p=0.006), whereby crude HbA1c decreased only among participants randomized to the Look AHEAD diet (−0.65 percentage points [95% CI −1.1, −0.18]; −7.1 mmol/mol [95%CI −12.0, −2.0], p=0.027). After adjustment for baseline HbA1c, diet duration, study site, gender, and race and ethnicity (Model 1), adjusted HbA1c decreased by −0.91 percentage points (95%CI −1.5, −0.30, p=0.005; −9.9 mmol/mol [95%CI −16.4, −3.3]), and did not vary by diet assignment (Model 2 p=0.09 for diet difference). Women reduced their HbA1c by 0.62 fewer percentage points (95%CI 0.02, 1.2, p=0.04) than men; and individuals belonging to a racial or ethnic subgroup other than non-Hispanic White reduced their HbA1c by 0.28 fewer percentage points (95%CI 0.10, 1.23, p=0.02) than those with a non-Hispanic White race and ethnicity.

Percent of Time Below Range (<70 mg/dL)

The three-month change in crude and adjusted percentage points of time <70 mg/dL was not statistically significant overall (Model 1 change 1.4 percentage points [95% CI −1.0, 3.7], p=0.25, n=28; Model 2 p=0.39 for diet difference]) or in any randomized diet group (variability shown in Supplemental Figure 2). However, the absolute percentage of time below range at 3-months was higher among participants assigned to the low carbohydrate diet (9.1% [IQR 4.5, 11.4%]) than among participants assigned to the Look AHEAD (3.1% [IQR 1.5, 7.6%]) or the Mediterranean (3.3% [0.43, 6.5%]) diets (p=0.04).

Secondary outcomes

Percent of Time in Range (70–180 mg/dL)

The Model 1 adjusted three-month increase in the percentage points of TIR (12.4 percentage points [95%CI −2.3, 27.2], p=0.09, n=28) was consistent with the lower HbA1c but did not reach statistical significance and was not different across diet groups (Model 2 p=0.46 for diet difference). According to the Model 1 term for race and ethnicity, participants belonging to a racial or ethnic subgroup other than non-Hispanic White reduced their time in range by 10.4 fewer percentage points (95%CI −21.3, 0.21) than their non-Hispanic White counterparts.

Percent of Time Below Range (<54 mg/dL)

Percent of time <54 mg/dL was 0.82% (IQR 0.19, 2.8%) at baseline. The three-month change was not statistically significant according to crude or Model 1 adjusted estimates (0.84 percentage points [95%CI −0.55, 2.2], p=0.23; Model 2 p=0.93 for diet difference).

Body Fat

Body fat decreased by −0.53 percentage point (95%CI −1.0, −0.03, p=0.04, n=38; p=0.39 for diet difference from Model 2) and −1.1 kg (95%CI −1.9, −0.40 kg, p=0.003; p=0.34 for diet difference) over three months, respectively. After adjustment for baseline body fat percentage, diet duration, study site, and gender in Model 1, the change in percentage points of body fat was attenuated to non-significance (−0.59 [95% CI −1.5, 0.35] p=0.21; Model 2 p=0.51 for diet difference).

Additional CGM Indices

The ~3-month change in percentage points of TAR (181–250 mg/dL and >250 mg/dL) remained stable overall. Although change in TAR (181–250 mg/dL) was numerically higher among participants assigned to the Mediterranean diet (6.6 percentage points [IQR 1.2, 13.8]) compared to those assigned to the low carbohydrate diet (−0.53 percentage points [IQR −7.4, 5.9]) or the Look AHEAD diet (−1.1 percentage points [IQR −2.9, 0.68]), the diet difference was not statistically significant (p=0.07). Nonetheless, the absolute 3-month percent of time in TAR (181–250 mg/dL) was higher among participants assigned to the Mediterranean diet (30.3 ± 11.7%) than among participants assigned to the low carbohydrate (18.0 ± 11.8%) or the Look AHEAD (17.1 ± 8.2%) diets (p=0.03). CV remained unchanged and was not different across diets.

Adherence to randomized diet assignments

Paired 24-hour dietary recall data were available for 27 participants. Participants assigned to the low carbohydrate (n=13), Look AHEAD (n=7), and Mediterranean diet (n=7) reported mean ~3-month changes in energy intake of −40.9±320.3 kcal, −804.2±662.2 kcal, and 462.0±619.9 kcal, respectively (p=0.0004 for diet difference, Supplemental Table 3). Change in percentage points of carbohydrate intake was −2.8±15.4, 8.6±7.1, and 6.0±17.4, respectively, among participants assigned to the low carbohydrate, Look AHEAD, and Mediterranean diets, respectively (p=0.20 for diet difference). Baseline carbohydrate intake was 35.2±11.3% overall and 33.6±10.2% in the low carbohydrate diet group (p=0.78 for diet difference). Change in percentage points of fat intake was 4.9±12.6, −11.9±7.0, and 6.0±17.4 among participants assigned to the low carbohydrate, Look AHEAD, and Mediterranean diets, respectively (p=0.01 for diet difference). Change in fiber intake was 0.68±6.9 g, −4.6±8.0 g, and 11.1±10.7 g among participants assigned to the low carbohydrate, Look AHEAD, and Mediterranean diets, respectively (p=0.005 for diet difference).

Discussion

Our study addressed the key gap in scientific knowledge related to the feasibility, acceptability, and effectiveness of dietary approaches for weight management tailored to adults with T1D. Our results demonstrate the feasibility of achieving modest short term weight loss (~2kg over 3 months) while sustaining or improving HbA1c levels and without increasing hypoglycemia exposure among young adults with T1D. Echoing the results of numerous weight management studies conducted in people without diabetes (26), our study demonstrates that there is not one single dietary approach that facilitates superior weight loss in people with T1D. Furthermore, assignment to a Mediterranean diet that does not require caloric restriction can lead to equivalent three-month weight loss compared to a hypocaloric low carbohydrate or Look AHEAD diet in this study population. It is important to acknowledge that only the Look AHEAD diet resulted in a statistically significant reduction in HbA1c levels, albeit the diet difference was attenuated following adjustment.

This study reveals the heterogeneity of weight loss responses to diets in people with T1D that has characterized virtually all diet interventions conducted in people without T1D (27). This heterogeneity is likely explained by a complex web of genetic, metabolic, behavioral, and environmental factors (28) yet to be fully elucidated, particularly in the metabolically unique setting of T1D. The gap in knowledge around factors that predict safe weight loss and maintenance in response to experimental diets among adults with T1D persists and requires further study alongside identification of barriers if T1D-specific recommendations are to be established. Although not statistically significant, there was a subset of participants in whom TAR increased on the Mediterranean diet; future studies with larger sample sizes should replicate this finding and investigate potential causes.

To our knowledge, this pilot study is the first randomized trial to indicate that explicit restriction of carbohydrates is not required to co-manage weight and glycemia effectively in T1D. This is an important point given the growing interest in this dietary approach for T1D management (29), which may not be feasible or sustainable for all, as demonstrated by the relatively low adherence to the carbohydrate goal in our low carbohydrate study arm. On the other hand, a low carbohydrate diet and more generally, weight loss, did not increase the incidence of hypoglycemia in our young adult study participants, which is speculated to be a substantial safety issue for patients on insulin therapy—although additional studies are needed to draw firm conclusions. It is worth noting that although the 3-month change in percentage points of time below range (<70 mg/dL) and time above range (181–250 mg/dL) did not differ by diet assignment, the absolute amount of time below range at 3-months was highest among participants assigned to the low carbohydrate diet, while the absolute 3-month percent of time above range was highest in the Mediterranean diet group. Thus, issues related to hypoglycemia may, in part, underlie the lower adherence to the low carbohydrate diet; and future, larger studies should confirm whether a Mediterranean diet without calorie restriction induces hyperglycemia among adults with type 1 diabetes—and in which subsets of participants. One limitation of this study is the heterogeneity of CGM wear time, which ranged from 1–14 days; although our sensitivity analysis indicates that CGM metrics were not statistically significantly different between participants with or without a minimum wear time of 8.5 days, future studies should develop strategies to ensure a two-week wear time for accurate estimation of CGM indices.

We detected disparities in study outcomes: men lost nearly 3kg more weight than women; and men and adults who identified with a non-Hispanic White race and ethnicity had a clinically significantly (30) greater reduction in HbA1c than women and adults who did not identify as non-Hispanic White, respectively. A growing body of evidence acknowledges the need to address social determinants of health to truly bridge disparities (31). The modest sample size of this pilot study and the scope of the aims precluded us from identifying solutions to these disparities. A future fully powered trial can investigate whether tailoring of diets by cultural eating preferences helps to mitigate disparities.

This study includes several limitations in addition to those already delineated. Most notably, we restricted our analysis to the pre-COVID-19 diet period 1 given changes to the primary outcome measurement methods and intervention delivery due to COVID-19-related study adaptations and limited pre-COVID-19 sample sizes in diet periods 2 and 3. Therefore, our results should be interpreted with caution, particularly for the comparison of diets. Nonetheless, we confirmed that participants included and excluded from analysis were not substantively different according to baseline characteristics—limiting the possibility that restricting our sample biased the results. A future paper will report findings following the temporary close of our research clinics due to COVID-19. A challenge of collecting 24-hour dietary recalls is that they are time-consuming to complete and rely on mutual availability between dietary interviewers and study participants; thus, a limitation of assessing adherence in the present study is that we were missing dietary intake data from 11 of the 38 completers. Furthermore, only one diet recall was available for some participants at a given time point, whereas it is ideal to collect two or more diet recalls per time point. Nonetheless, missingness did not vary by diet assignment, and seminal literature in nutritional epidemiology suggests that one or two dietary recalls are sufficient to obtain group means, although standard deviation will be overestimated (32). The day of the week on which diet recalls were collected did not vary by time point—which prevents further inflation of the standard deviation (32).

Future studies should estimate the concomitant changes in insulin dose that accompany weight loss and dietary changes among adults with T1D. Generalizability of study findings to middle-aged or older adults with T1D, and those who may benefit from glycemic management but do not have overweight or obesity, may be limited. It should be noted that study retention was likely impacted by the unique challenges faced by young adults. As cited by our study participants, these included financial considerations related to making dietary changes, balancing the demands of parenthood, and the ability to make a long-term commitment in a life stage of high flux. Inherently, this pilot work was designed to inform a future, fully powered trial. As such, we note several lessons learned: there was variability in CGM wear time, stemming largely from sensor displacement, which can be addressed in a future trial through the provision of moisture-resistant sensor patches. Additionally, future studies enrolling young adults might improve retention and data completeness by building in flexibility (e.g., offering virtual counseling sessions) and collecting 24-hour dietary recall data using the self-administered ASA-24, which has high acceptability and has been validated against interviewer-administered 24-hour recalls (33).

This study includes several strengths. Virtually no randomized trials have tested diet approaches for weight management in adults with T1D—which has precluded the development of T1D-specific weight management guidelines. We deployed a diet intervention designed to improve not one, but two key outcomes (weight and glycemia) which are strongly predictive of long-term cardiovascular risk (7). We considered three dietary strategies within the same trial, each implemented using equivalent, established theories of health behavior change (34). We utilized an innovative SMART pilot design to address heterogeneity in participant response and achieved safe weight loss in a population at high risk of dysglycemia and weight gain (35).

Although our interrupted study requires replication in another SMART design to inform personalized diet tailoring, we conclude that weight loss and glycemic management are not inexorably at odds with one another in the clinical management of T1D, and that a variety of diet approaches—including those that do not require explicit calorie restriction—can be effective to serve these equally important goals. Much work remains to be done to tailor weight management guidelines to individuals with T1D, for whom co-optimization of weight and glycemia is critical for cardiovascular risk reduction. In a future analysis, we will consider predictors of weight loss, including adherence, physical activity, insulin regimen, CGM utilization, and ingestive behaviors (36). The ultimate goal of this pilot study was to inform the design of a fully powered efficacy trial to divulge individual-level predictors of dietary response, so that T1D-specific weight management guidelines may be established, and healthcare professionals may be able to personalize recommendations more optimally.

Abbreviations:

ACT1ON

Advancing Care for Type 1 Diabetes and Obesity Network

CGM

continuous glucose monitoring

DXA

dual energy x-ray absorptiometry

HbA1c

hemoglobin A1c

IQR

interquartile range