Time-Restricted Eating as a Nutrition Strategy for Individuals with Type 2 Diabetes: A Feasibility Study

This study was a feasibility study conducted at Australian Catholic University’s Melbourne campus between July 2018 and November 2019. Ethical approval was obtained from the ACU Human Research Ethics Committee (2018-75H) and was prospectively registered on the Australia New Zealand Clinical Trial Registry (ACTRN12618000938202). The study was conducted in accordance with the Declaration of Helsinki and all participants provided written informed consent prior to participation.

The study was a pre-post non-randomized design whereby participants completed a 2-week “Habitual” baseline period of dietary recordings followed by a 4-week TRE intervention period where they were instructed to limit all eating occasions to between 10:00 and 19:00 h on as many days each week as possible. No dietary advice around food type, quality, or portion sizes was provided. Participants were requested to take photos at every eating occasion and complete daily food diaries throughout both the Habitual and TRE intervention periods. The primary outcomes for the study are reported here, including the feasibility of recruitment (number of persons screened compared to final enrolments); retention rate (number of participants enrolled compared to the number of participants who finish the final study measures); safety (participant report of adverse events, i.e., self-report of any gastrointestinal, nausea, dizziness, or other events as a result of the dietary intervention); compliance to the dietary intervention assessed using a combination of participant self-report and photo images of meal times (with time stamps); and acceptability (attitudes/opinions/barriers assessed in a qualitative interview at the completion of the intervention (i.e., after 4 weeks of TRE)). We also explored glycaemic control (via mixed meal tolerance test (MMTT), fasting blood profiles (e.g., glycated haemoglobin (HbA1c), glucose, insulin and lipids), physiological (e.g., body mass, composition, blood pressure (BP), and resting metabolic rate (RMR)) and psychological outcomes (e.g., quality of life, depression, anxiety and stress, eating behaviours, sleep quality and cognitive function, assessed using validated self-report questionnaires) pre and post the TRE intervention period.

Individuals with T2D (aged 35–65 years, body mass index (BMI) between 25 and 45 kg/m²) were recruited using targeted campaigns of flyers/posters, social media advertisements, databases from previous studies and via a research participant recruitment company (Trialfacts, Melbourne, Australia). Initial eligibility screening was conducted by telephone and comprised of medical history, age, and estimates of height and weight (to calculate BMI). Respondents were eligible if they met the following criteria: diagnosed (by a GP/endocrinologist) with T2D, with an HbA1c between 6.5 and 9%, and either diet-controlled or taking a maximum of two oral hypoglycaemic agents (excluding sulphonylureas, insulin, and GLP-1 agonists); and currently consuming food (i.e., dietary intake) over a period of 12 h or more, habitually (i.e., self-reported on 5 of 7 days/week).

Potential participants were excluded based upon the following criteria: not a regular (≥5 days/week) breakfast consumer; could not operate the camera function on a smart phone; taking glucose lowering medications which are contraindicatory for fasting [28] (i.e., sulphonylureas, or insulin) or requiring injecting (i.e., GLP-1 agonists); reported to be following a strict diet (i.e., vegan, coeliac/gluten free, or ketogenic); had participated in regular fasting (defined as fasting ≤16 h/day or having completed twelve 24-h fasts within the past year); participating in shift work (i.e., >3 h between 22:00 and 05:00 h for 1 day/week (>50 days per year)); not weight stable (>5 kg change over last 3 months); on prescribed medications required to be taken with food in the early morning or late evening or taking other prescribed medications for <3 months; current smoker (tobacco, nicotine, or marijuana) or within 3 months of quitting; women who were pregnant, breastfeeding (within 24 weeks); history of psychotic disorder, current diagnosis of other major psychiatric illness (e.g., mood disorder, eating disorder, substance use disorder; does not include depression); psychopharmacological treatment that has not been stable for more than 3 months; taking medications known to promote weight gain, weight loss or interact with glucose metabolism (i.e., corticosteroids); diagnosed gastrointestinal conditions, surgery (i.e., bariatric) or impaired nutrient absorption; or antibiotic use in previous 3 months.

Potentially-eligible participants were invited to meet with the principal researcher to provide informed consent and complete questionnaires including morning-eveningness questionnaire self-assessment (MEQ-SA) [29,30] and Eating Attitudes Test-26 EAT-26; [31] to identify circadian preferences and tendencies for eating disorders, respectively. The two participants demonstrating an elevated EAT-26 score (i.e., ≥20) met with a dietitian to clarify these scores. In both cases they were attributed to diabetes-related dietary stress rather than eating disorders, so the participants remained in the study.

Following informed consent and enrolment, participants attended the laboratory between 07:00 and 09:00 h after a >10 h fast for a baseline MMTT. A fasting blood sample (4 mL, EDTA) was collected from an indwelling cannula inserted into an antecubital vein prior to consumption of the MMTT drink (50% CHO, 30% fat, 20% protein; Sustagen Hospital Chocolate with Fibre with full fat (3.5%) milk) which was 20% of estimated total daily energy requirements, calculated using the Schofield equation (activity factor 1.3) [32]. Following the MMTT drink consumption, blood samples (4 mL, EDTA) were obtained every 30 min for 2 h. During the MTTT, psychological and cognitive testing was performed between blood sampling points. At the end of the MMTT, participants were given a handbook for dietary recordings and instructions for taking photos of all meals and snacks and were asked to complete a 2-week “Habitual” period of baseline monitoring. After the 2-week, at Visit 3 (Figure 1), participants reported to the laboratory after a 10-h fast for measures of body composition (dual-energy x-ray absorptiometry (DXA); GE Lunar iDXA Pro, enCORE software Version 16) and resting energy expenditure (REE; TrueOneRMR, Parvo Medic, Sandy, UT, USA), as previously reported [33], and seated blood pressure (BP; ProBP 3400, Welch Allan Inc, Skaneateles Falls, NY, USA), prior to a fasting blood sample via venepuncture. Dietary entries where checked for completeness with questions asked to clarify any information from written records, EDD, or photos. Participants then reported to the laboratory once per week for 4-weeks for ~30 min for a fasting blood sample, to check dietary entries and were queried with regards to any medical issues that may have occurred (Visits 4–6). The fasted physiological measurements (DXA, RMR, and BP), MMTT, psychological, and cognitive testing were repeated at the end of the 4-weeks of TRE (Visit 7). Within 0–4 days of the end of the fourth week (Visit 8), a qualitative interview was conducted by a researcher (EBP) to understand attitudes, barriers and adherence, with written notations and voice recordings (VoiceRecorder app for iOS, Tapmedia Ltd., London, UK). At the conclusion of the study, participants were gifted a $250 supermarket voucher to thank them for their participation.

For the entire study period (6 weeks), participants were asked to record all food and drink consumed using their choice of paper-based handbook or using the publicly available phone app EasyDietDiary (EDD; Xyris Software, Brisbane, Australia). The paper- or app-recorded intakes were entered into FoodWorks 9 software (Xyris, Brisbane, Australia) using the AUSNUT 2011-13 nutrient database (https://www.foodstandards.gov.au/science/monitoringnutrients/ausnut/pages/default.aspx↗) and all entries were reviewed/screened by one researcher (KHCL). In conjunction with recording their diet, participants were asked to photograph each eating/drinking occasion using their own mobile device. The photos were then stored in a shared folder with the primary investigator or were transferred at each measurement visit. On occasions where participants missed taking a photo, they were encouraged to manually record the time of food or drink consumption in the paper-based handbook or in the notes section of the EDD app (as the app does not allow for time data or time-stamped photos). Photos were encouraged over written records as time-stamp data generated by mobile devices were more accurate than the estimated written times (e.g., writing 13:00 h vs. time-stamp 13:11 h). Time-stamp data were extracted from raw photo files using Directory Lister Pro v2.36 (KRKsoft.com, Krakow, Poland) and researchers viewed and recorded the contents of each photo to match time-stamps with the corresponding meals/snacks recorded in the EDD app or paper diary. These time-stamps and participant-written times were also entered into FoodWorks 9. At face-to-face appointments, where necessary, participants were asked by one of the research team to clarify any ambiguous food entries or food photos for clarity and accuracy of dietary recordings. Regardless of method of recording dietary entry, participants were asked to self-report whether they were adherent on each day of the TRE intervention by circling Yes or No to the question, “Were you able to adhere to the 10 am to 7 pm time-restriction for eating and caffeine intake today?”.

Food and drink intake were grouped into eating occasions using the definition that best predicted variance in energy intake [34], which is food/drink consumed within a 15-min window with energy content ≥210 kJ. In the paper diaries, participants were asked to report foods eaten during “Early morning”, “Breakfast”, “Morning tea”, “Lunch”, “Afternoon tea”, “Dinner” or “Supper”. The EDD app asked participants to report foods as “Breakfast”, “Lunch”, “Dinner” or “Snacks & Drinks”. With the exception of “Snacks & Drinks”, median time of these meals/snacks by weekend/weekday and Habitual/TRE was calculated. For example, median times for breakfast on a Habitual weekday, Habitual weekend, TRE weekday, and TRE weekend were calculated for each participant. Where participants had recorded the composition of a type of meal/snack but timing was not available, the corresponding median time was applied (e.g., a participant’s median time of habitual weekday breakfast was used if the actual time of such a breakfast was unavailable). Where time was available but not meal type (i.e., a photo was available but food was categorised as “Snacks & Drinks’’) and the 15-min and ≥210 kJ criteria were not met, the food was classified based on the adjacent meals/snacks (e.g., a food entry with a photo taken at 15:20 h would be classified as ‘Afternoon tea’ if it fell between the participant’s lunch and dinner that day). Further, to allow for photos taken just prior or following consumption, a 15-min buffer was applied, so that time-stamps between 09:45 and 09:59 h and 19:01 and 19:15 h were considered as falling within the TRE time window. To account for days where food consumption continued past midnight, one ‘day’ of dietary data was defined as starting from the first eating occasion until the last eating occasion of the day (e.g., “Supper” at 01:30 h on a Tuesday was grouped with Monday’s intake). In this group of 19 participants, the earliest first eating occasions of the day occurred at 05:15 h and the latest occurred at 03:36 h.

Upon collection, blood samples (EDTA) were inverted and blood lipids (triglycerides, total, HDL, and LDL cholesterol) were analyzed in whole blood at the time of collection using a Cobas b 101 instrument (Roche Diagnostics Ltd., Basel, Switzerland). At Visits 3 and 7, glycated haemoglobin (HbA1c) was also measured from whole blood using the Cobas b 101 instrument prior to centrifugation. The remaining samples were spun at 1800× g for 10 min at 4 °C to obtain plasma samples which were then frozen at −80 °C for later analysis. Glucose concentrations were measured in duplicate from thawed plasma samples using YSI 2900 analyzer (YSI Life Sciences, Yellow Springs, OH, USA; coefficient of variation (CV) < 1.0%). Insulin concentrations were measured using thawed plasma with a commercially available ELISA (Alpco Ltd., Windham, NH, USA), in duplicate with an intra-assay CV of 3.8%. Total area under the curve (AUC) was calculated for venous glucose and insulin concentrations using the trapezoid method using GraphPad Prism (Version 7.01, GraphPad Software Inc., San Diego, CA, USA).

Psychological Wellbeing: The Depressive Anxiety Stress Scale 21 (DASS) was used to measure participants negative emotional states of depression, anxiety, and stress [35]. The AQoL-8D was used to measure participants health-related quality of life [36]. The Pittsburgh Sleep Quality Index (PSQI) was used to measure participants sleep quality and disturbances [37]. The Eating Disorders Examination Questionnaire (EDE-Q) was used to measure participants behaviour and attitude towards eating [38]. The Clinical Impairment Assessment questionnaire (CIA) was used to measure participants psychological impairment resulting from eating disorder symptoms [39]. All psychological measures assessed variables over the previous 7 (i.e., DASS, AQoL-8D) to 28 days (CIA, EDE-Q, PSQ) and have been validated and used in both community and clinically-indicated samples.

Cognitive Functioning: The Cogstate Brief Battery (CBB) was used to measure participants cognitive functioning across four domains including processing speed (detection task—measuring speed of performance), attention (identification task—measuring speed of performance), visual learning (one card learning task—measuring accuracy of performance) working memory (two back task—measuring accuracy of performance) [40], and executive functioning (Groton Maze Learning Task—measuring errors in performance). Lower scores on the Groton Maze Learning, identification and detection tasks indicate improved cognitive performance. Higher scores on the one card learning and two back tasks indicate improved cognitive performance. These constructs were chosen as they have been shown to be impacted by metabolic disorders, and/or lifestyle interventions [41]. These measures were selected as they have demonstrated validity, reliability, stability and sensitivity in healthy and disordered populations. Each task has been developed to be brief, simple and able to be repeated without giving rise to practice effects [42].

All psychological data was checked for accuracy and completeness through routine data cleaning and management procedures. Intervention efficacy was assessed through comparison of the outcomes pre- and post-intervention using a series of repeated measures analyses. AQoL-8D results were calculated as weighted psychometric scores to allow comparison before and after intervention [43]. Cognitive data was transformed according to Cogstate data handling procedures prior to analysis [42].

Semi-structured interviews consisting of 13 open-ended questions were conducted by one researcher (EBP) to gain insight and understanding of the attitudes of participants after the 4-weeks TRE intervention period. The questions within the interviews were developed based on a previous intervention ([44]; see Supplementary Material: Qualitative Interview Questions). Written and recorded responses were collated and analysed by one researcher (CG). The audio-recorded interviews were imported verbatim into NVivo (NVivo qualitative data analysis software, Version 10, QSR International 14 Pty Ltd., VIC, Australia) and supplemented with the interviewer’s written notes when required. Data were then explored using thematic analysis [45]. Data from the interviews were first read and key preliminary findings were highlighted and noted. Participants’ responses were then coded in an inductive, exploratory manner. These codes were then grouped into common themes and reviewed and adjusted where necessary. Post-analysis verification of themes was conducted between three investigators (CG, LB, and EBP). Quotations from original responses provide support for the findings.

As this was a feasibility study, it was not possible to conduct power analyses to determine sample size requirements. However, due to the varied tolerability to an intermittent fasting intervention (equivalent to 4–6 h TRE) in individuals with T2D reported in a previous 2-week intervention with an n = 10 sample size [20], we set out to recruit 24 individuals, with an expected dropout rate of ~15%, to complete 20 individuals. All data was analysed using the IBM Statistical Package for Social Sciences (SPSS) version 25. Significance was set at p < 0.05 and all data are presented as mean ± SD. One-way ANOVA tests were used to compare differences between intervention (Habitual vs. TRE) periods, with mean differences and 95% confidence intervals (CI) reported for significant differences and with effect sizes (Cohen’s d) where appropriate. Where more than one factor was investigated (i.e., intervention period and time or proportion of adherence), linear mixed models were used with post hoc tests using Bonferroni corrections, where appropriate. For analyses of nutrition and physiological data, visual inspection of outcome variable residual plots was conducted to assess their distributions.

Through multiple forms of advertising, 594 participants either contacted the research team directly (via phone or email; 25%) or answered the online screening questionnaire promoted by Trialfacts (75%) (see Figure 2). Trialfacts referred 449 respondents, of whom 417 were not eligible (93%) to participate largely due to time between first and last meal (eating window) of <12 h (40%), BMI outside of 25–45 kg/m² (26%), not a regular breakfast consumer (24%), or HbA1c not between 6.5 and 9% (21%). The phone screening of the pre-screened Trialfacts referrals led to 15 of 32 participants being eligible (where 3 were uncontactable, 7 were not interested, and 7 were ineligible) and 13 (87%) were then enrolled into the study (with 2 with HbA1c too high (≥9.0%)). Of those who directly contacted the research team via phone or email (145), 133 were phone screened for further eligibility, with 121 not eligible (91%; due to medications (37%), could not recontact (16%), HbA1c not in range (15%), not interested (13%), eating window (10%) or BMI out of range (9%)), and therefore 12 participants attended the laboratory for pre-screening with 11 (92%) enrolling into the study. Thus, 24 participants were enrolled in the study and began the Habitual baseline period, which was 4% of those who expressed initial interested in the study.

Although 20 participants completed the 4-week TRE intervention period, there was insufficient dietary data (7/14 days in Habitual, 50%; and 13/28 days in TRE, 46%) for one participant. Therefore, all analysis is for the final participant cohort of 19 and participant characteristics of the cohort is presented in Table 1. Of the 19 participants, nine were not taking any hypoglycaemic medications, ten were taking biguanides (i.e., metformin), three taking SGLT-2 inhibitors and two taking DPP-4 inhibitors, with three participants taking two hypoglycaemic medications. No incidences of hypoglycaemia were reported by participants nor were any adverse events recorded during the duration of the study.

Of the 19 participants, six completed written recordings with the remainder using the electronic EDD app. Dietary data for each day were considered valid if participants recorded two or more meals/snacks. Days with a single meal/snack were valid if these records were verbally confirmed at the weekly visits. Table 2 outlines the average dietary intake throughout the Habitual and TRE periods for all participants, with individual data presented as box plots for each period in Supplementary Materials Figure S1. There were no differences in daily energy intake, macronutrient contribution to daily energy intake (relative, %), or absolute macronutrient or micronutrient intakes between Habitual and TRE period (p > 0.05).

Participants reported a total of 3427 energy-containing meals/snacks over the 6 week recording period, of which ~92% met the ≥210 kJ energy content threshold (1089/1234 in Habitual period and 2051/2233 in TRE period, Figure 3A) to be defined as an eating occasion (EO; [34]). Participants complied with providing photos or self-reported timing for 90 ± 12% of meals/snacks (range: 63–100%; 87 ± 20% of Habitual period days and 91 ± 12% of TRE period days). When participants did not provide photos or timing of EO’s, calculations of total energy intake were reduced in the TRE period (with time data: 9013 ± 2687 kJ/d; without time data: 7590 ± 3057 kJ/d, p < 0.001) but was not altered in the Habitual period (with time data: 8757 ± 2764 kJ/d; without time data: 8245 ± 2682 kJ/d, p = 0.07).

Participants reported an average of 3.7 ± 1.2 EO’s per day during the 6 weeks. On average, TRE reduced the time window of energy consumption by 22% from 10 h 42 min to 8 h 20 min (Figure 3A). Using a previous definition of the energy consumption window derived from 95% of the times a participant consumed a meal/snack containing >5 kcal (21 kJ) [3,4], the median timespan of eating was 14 h 19 min and reduced to 11 h 30 min in the TRE period (−2 h 53 min; 95%CI: 1 h 52 min–3 h 45 min; p < 0.001). Mean time of first EO shifted from 09:19 h in the Habitual period to 10:32 h in the TRE period and from 20:03 h to 18:43 h for the last EO for Habitual and TRE periods, respectively. Despite the reduction in duration over which energy was consumed, the number of EOs was similar between Habitual (3.9 ± 1.3 EOs) and TRE (3.7 ± 1.1 EOs) periods. The distribution of EOs, and thus the accumulation of total energy intake, differed between the Habitual and TRE periods (Figure 3B; main interaction effect, p < 0.001). Delaying the start time of eating reduced the accumulated energy intake in the TRE period between 07:00 and 09:00 h and at 12:00 h (p < 0.05). The earlier finishing eating time corresponded with a greater proportion of energy consumption occurred between 18:00 and 19:00 h in the TRE period (p < 0.05).

Prior to the prescribed TRE period, participants did not habitually restrict energy consumption to the 10:00–19:00 h window, with 0.9 ± 1.2 days meeting the TRE intake window. Adherence to the 10:00–19:00 h per day TRE window (±15 min either side) during the TRE period, as assessed by photo metadata or self-reported time of eating, was 20 ± 7 of the 28 days (range: 1–28 days; 72 ± 24%, range: 4–100% compliance), equivalent to ~5 days/week. During TRE, a greater proportion of weekdays (i.e., Monday–Friday) were deemed adherent to the prescribed TRE period (77% (288/376 weekdays)) compared to Saturdays and Sundays (61% (90/148 weekend days)). A greater proportion of daily non-adherence was due to food/drink being consumed after 19:00 h (60%; 87/146 non-adherent days), while 23% (33/146 days) was due to food/drink being consumed before 10:00 h and 17% due to non-adherent in the morning and evening (before 10:00 h and after 19:00 h).

Participants failed to report whether they were adherent or not in their written handbook on 14% of occasions (Table 3). On the 453 instances where adherence was self-reported (Yes/No), 17% of self-report was inaccurate compared to time-stamped adherence (09:45–19:15 h). Participants reported non-adherence on 23/325 occasions (8%) when the time-stamped dietary data indicated they were adherent. In contrast, a larger proportion of participants reported they were adherent on 55/128 occasions (43%) when the time-stamped dietary data indicated they did not adhere.

Considering the variability in self-report responses, adherence data for dietary analysis has been calculated using the time-stamped dietary data. When comparing the days where participants did adhere to the time-restriction of 09:45–19:15 h, total energy intake was significantly lower (−1033 kJ, 95%CI: −1544 to −523 kJ, p < 0.001; Table 4). That is, when participants were adherent to the prescribed TRE period they consumed lower absolute intakes of carbohydrate (−49 g/d, 95%CI: −64 to −33 g/d), sugar (−10 g/d, 95%CI: −17 to −4 g/d), fibre (−3 g/d, 95%CI: −5 to −1 g/d), alcohol (−7 g/d, 95%CI: −10 to −5 g/d) and potassium (−201 mg/d, 95%CI: −389 to −14 mg/d), which coincided with higher relative intakes of protein (+2%, 95%CI: 1–3%) and fat (+4%, 95%CI: 2–7%) and lower relative intakes of carbohydrate (−5%, 95%CI: −7 to −2%) and alcohol (−2%, 95%CI: −3 to −1%). To assess how adherence to the 10:00–19:00 h TRE period affected dietary intake over the 4-w in comparison to habitual intakes, the proportion of time spent adherent in the TRE period (from time-stamped data; median adherence 77% (interquartile range: 54–93%) was modelled against nutrient intakes in the habitual vs. TRE periods. Only carbohydrate intake was associated with TRE adherence, with lower intakes reported by the more adherent participants. Specifically, during the TRE phase, participants in the lower quartile of adherence (54% adherence) consumed 204 g/d (95%CI: 175–232 g/d) carbohydrate compared to 151 g/d (95%CI: 120–181 g/d) carbohydrate in the upper quartile (93% adherence), a difference of 53 g/d (p = 0.005; 95%CI: −91 to −16 g/d).

From fasting blood samples obtained throughout the intervention (Table 5, Table S1), HbA1c was not significantly lower after the 4-week TRE intervention (−0.2 ± 0.4%; p = 0.053). No differences were measured in fasting glucose, insulin, or lipid concentrations between Habitual and TRE periods.

Blood samples were unable to be obtained throughout the MMTT for one participant, and thus data for this measure is from n = 18 participants. Mean glucose concentrations over the MMTT were significantly lower between Habitual and TRE (main effect of condition: p < 0.001) but glucose concentrations did not change differently over time between Habitual and TRE (Interaction effect; p = 0.97; Figure 4A). Fasting glucose levels at the beginning of the MMTT was not significantly lower after the TRE intervention (Habitual: 8.8 ± 2.2 mmol/L; TRE: 8.2 ± 1.7 mmol/L; −0.6 mmol/L, 95%CI: 0.1 to −1.3 mmol/L, p = 0.075), as did peak glucose during the MMTT (Habitual: 12.0 ± 2.8 mmol/L; TRE: 11.4 ± 2.4 mmol/L; −0.7 mmol/L, 95%CI: 0.1 to −1.4 mmol/L, p = 0.075). Total AUC for glucose was not significantly lower after the TRE intervention (p = 0.056; Figure 4B). There were no differences between Habitual and TRE periods in response to the MMTT for insulin concentrations (main effect of condition: p = 0.38) or total AUC insulin (p = 0.20; Figure 4C,D, respectively). Consequently, neither fasting nor peak insulin concentrations from the MMTT were not different between Habitual and TRE periods (p = 0.45 and p = 0.37, respectively).

There were no differences between body mass, composition resting energy expenditure, or blood pressure measures between the Habitual period and the end of the TRE intervention period (Table 6).

Sixteen of the 19 participants completed post-intervention qualitative interviews (~14 ± 3 min). Thematic analysis revealed four themes related to participant experiences over the 4-week TRE intervention period: (1) difficulties with the morning or evening, (2) impact on eating behaviours and food choices, (3) emotional reactions, and 4) monitoring and accountability. A summary of the themes and sample quotes identified from participant interviews are outlined in Table 8.