Study design, setting and population

The study is a two-arm, parallel group, single-masked randomised controlled trial. Adults with pre-diabetes or early-stage T2DM who are not treated with glucose-lowering medications are randomised, with equal allocation, to either the PNP or LFHF diet arms, in both arms participants are commenced on metformin Extended Release (XR) 1500 mg/day treatment for 6 months. All the study visits are performed at the Clinical Research Facility at the Garvan Institute of Medical Research (Sydney). Metagenomics and data processing for the personalised dietary interventions are performed at the Weizmann Institute of Science (Rehovot).

Patient and public involvement

Patients or the public were not involved in the design or other aspects of the research.

Eligibility

Adults (20–70 years old) with pre-diabetes or recently (in the last 6 months) diagnosed T2DM with (HbA1c ≤8.0% (64 mmol/mol)), not pregnant or planning to become pregnant during, and for at least 3 months after the study, are recruited (box 1). A wide age range was selected to encompass different populations of individuals managing their pre-diabetes for short or long durations and to increase the likelihood to recruit the sample size in a timely manner. The HbA1c cap at 8.0% (64 mmol/mol) was selected to ensure that individuals with T2DM are relatively well controlled. Blood tests indicative of normo-glycaemia, HbA1c above 8.0% (64 mmol/mol), liver enzymes (alanine aminotransferase and/or aspartate aminotransferase) over three times the normal range limit and estimated glomerular filtration rate (eGFR)19 lower than 45 mL/min/1.73 m² are grounds for exclusion. Individuals treated with glucose-lowering medication other than metformin in the last 24 months, or metformin in the last 3 months, will be excluded. Individuals with conditions or treatments that affect glycaemia (eg, oral steroids), impact weight (eg, bariatric surgery, weight loss medications) or the gut microbiome (eg, inflammatory bowel disease, coeliac, frequent antibiotic treatment) will be excluded. Participants who have had a cardiovascular event in the previous 6 months or received an investigational new drug within the last 6 months will be excluded. Participants with conditions that may interfere with the ability to understand the requirements of the study and those who refuse treatment with metformin or refuse to use the smartphone application will be excluded (box 1).

Recruitment

The study is advertised in general practitioners (GPs), endocrinologists and dieticians’ practices in the Sydney metropolitan area, and through targeted social media campaigns. A collaboration with Blacktown Mt Druitt Hospital (Western Sydney) has been established in 2019 for the purpose of recruitment. During a hospital screening programme ran between 2016 and 2018, 17.3% and 30.2% of individuals visiting the emergency department (ED) at Blacktown Mt Druitt Hospital have had HbA1c values indicative of diabetes and pre-diabetes, respectively.20 Since September 2019, individuals visiting the ED who have had a blood test indicative of pre-diabetes receive a letter prompting them to contact the PREDICT team and encouraging they share the result with a GP.

To date (February 2020), 38 participants were enrolled. Of the 38 participants enrolled, 20 completed, 13 are ongoing, and 5 withdrew before the end of the treatment (13% dropout rate). Recruitment of participants is expected to complete in December 2021.

Participants contacting the team receive the participant information sheet (online supplemental material) via email or post. Experienced clinical research nurse/associate provides details about the study over the phone. Willing participants are referred to a commercial pathology to perform an oral glucose tolerance test (OGTT, 75 g) and HbA1c test. They are asked to sign a consent form after reading the participant information sheet explaining the possible risks of undergoing the OGTT and HbA1c tests prior to performing the blood tests. If the blood tests indicate either T2DM (with HbA1c ≤8.0% (64 mmol/mol)), or impaired fasting glucose or impaired glucose tolerance or HbA1c ≥5.7 (39 mmol/mol), they are invited to a screening and enrolment visit at the Garvan Clinical Research Facility.

A stool collection kit for metagenomics OMNIgene GUT (OMR-200; DNA Genotek) is mailed to participants prior to the screening/enrolment visit. Participants collect the sample according to the manufacturer’s instructions the day before the visit and keep the sample at room temperature. At the Clinical Research Facility, the sample is vortexed, centrifuged for a few seconds and material aliquoted into cryo vials and kept in −70°C freezer. One vial is transferred to facilities at the Weizmann Institute of Science and stored at −20°C until DNA extraction.

The pretreatment data are collected across the screening/enrolment and the baseline visits.

Screening/enrolment procedures and measurements

During the screening/enrolment visit, participants sign the study informed consent form (online supplemental material) and undergo medical examination by a physician. Participants have their weight, height, waist and hip circumference and blood pressure measured. Basal metabolic rate (BMR) is estimated using bioelectrical impedance analysis (BIA, used for calculating the energy requirement, see the Energy target section). Blood samples are collected to evaluate liver (liver enzymes) and kidney (creatinine and eGFR) function and full blood count. Glucose monitor (FreeStyle Libre Pro, Abbott, Germany) is attached for a period of 14 days.

A link to download the PNP smartphone application is sent to the participants prior to the screening/enrolment visit along with a short video demonstrating the app use. They are asked to log-in to the app with a personal (re-identifiable) code provided by email, and to familiarise with the app in preparation for a training session with the dietician. During the screening/enrolment visit, the dietitian practices with the participants browsing the food database, selecting food and beverage items and indicating the amount consumed. Participants are taught to add frequently consumed foods to a favourites list, which makes future search of food items easier. When the CGM is on, participants are asked to carry on with their usual routine and to record all meals, snacks and drinks using the app. The period between the screening/enrolment and the baseline visits (4–6 weeks) serves as the ‘run-in’ period.

Randomisation

Randomisation is performed between the screening/enrolment and baseline visits. Individuals are randomised with 1:1 allocation into the two arms in rounds of four to six individuals each, with randomisation performed within each round using the minimisation programme for allocation of subjects to parallel groups, modified from Saghaei et al.21 They are stratified by gender, age (20–49 or 50–70 years), BMI (<25.0 or >25.1 kg/m²) and HbA1c (<5.7 (39 mmol/mol) or >5.8% (40 mmol/mol)). To avoid bias, the randomisation is performed by a study investigator located at the Weizmann Institute who does not interact with the study participants. The study nurses and physicians who have direct contact with the participants are blinded to the randomisation order, however due to the nature of the intervention, the study dietician is not blinded to the treatment allocation.

Baseline visit: measurements

The baseline visit is performed approximately 4–6 weeks after the screening/enrolment visit (table 1). Participants attend the Clinical Research Facility following an overnight fast. Blood is drawn for serum lipids measurement and anthropometric measures and blood pressure are taken. Arterial stiffness (pulse wave analysis, AtCor Medical, Australia) is measured twice and average recorded, as described.22 This is followed by measurement of resting energy expenditure, carbohydrate and fat oxidation over 30 min by indirect calorimetry (Quark, Cosmed, Italy).23 Body composition is assessed using dual-energy X-ray absorptiometry (DXA, Lunar Prodigy, GE Healthcare). Specifically, total body fat mass and fat-free mass (enCORE software), the android and gynoid region, and visceral fat (CoreScan software, GE Healthcare) are recorded.24 Liver steatosis (controlled attenuation parameter, CAP) and liver fibrosis (liver stiffness measurements, LSMs) are assessed using FibroScan (Touch 502 by Echosens) by a trained technician. CAP and LSMs have been reported to correlate closely with steatosis and fibrosis assessed using the gold standard liver biopsy.25 A physical activity monitor (ActivePal, Pal Technologies) is applied on the thigh for a period of 14 days. At the end of the baseline visit, the participants practice using the app with the dietician.

Prediction of PPGR using the algorithm

The prediction of PPGR in PREDICT follows the modelling framework described in Zeevi et al16 and is performed between the screening/enrolment and baseline visits. Time-stamped food records from the app, CGM and other data collected during the enrolment visit at the Garvan are shared with a mathematician at the Weizmann Institute of Science where data processing occurs, on an Institutional secured server. The data, together with the stool metagenome sequencing data are integrated with the Weizmann Institute’s database to develop personalised algorithms for predicting each individual’s PPGR. A database of recipes of meals (n=233) and smaller meals (‘snacks’, n=249) varying in macronutrient composition to generate feedback on the PPGR to pre-consumed meals has been created. Using the participants' features, personalised PPGRs are calculated for every meal and snack in the database based on nutrient composition, and energy-adjusted quintile cut-offs of PPGR are used to create personalised meal ratings ranging from 1 to 5 (corresponding with ‘excellent’, ‘good’, ‘medium’, ‘bad’ and ‘very bad’). The predictive model, originally trained on data collected in an Israeli adult cohort,16 has been shown to be predictive of PPGR in a US cohort of healthy adults (n=327) consuming a Western-style diet.26

Interventions

Both arms use the PNP mobile app to select meals/foods. In 2018, the app, developed at the Weizmann Institute of Science, was adapted to Australian consumers including the Australian food database (AUSNUT 2011–2013)27 of approximately 5700 food items.

Personalised diet (PNP diet arm)

Participants in the PNP diet arm receive personalised feedback on each of their food item/meal choice and are asked to consult with the app in real-time to select the recommended meal for them. The feedback is color-coded with a traffic light system; green (‘good’ and ‘excellent’), yellow (‘medium’) and red (‘bad’ or ‘very bad’) PPGR (figure 1A,B). Participants are advised to aim for as many ‘good’ and ‘excellent’ scores, occasional ‘medium’, and to avoid ‘bad’ and ‘very bad’ scores. When receiving bad scores, they are advised to trial substituting, adding or removing ingredients from the meal to improve the score. In individuals with hypercholesterolemia, a special set of recipes containing reduced saturated fat are uploaded into the smartphone app. Special recipes are available for individuals practising vegetarianism or avoiding dairy, eggs and fish/seafood.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Screenshots of the smartphone application used daily by participants in the Personalised Medicine in Pre-diabetes—Towards Preventing Diabetes in Individuals at Risk Study. Participants randomised to the personalised diet arm receive scores for each meal. Panels (A) and (B) depict two meal options selected by an individual in the study where two isocaloric breakfasts are predicted to result in modest (A) or exaggerated (B) postprandial glycaemic responses. The daily energy intake and macronutrient breakdown are provided to each of the study participants (C).

Monitoring and adherence evaluation (relevant to both study arms)

The dietician reviews scores calculated programmatically based on the frequency of using the app and on meeting the daily energy target, along with the time-stamped meals consumed by the participants daily. In the PNP arm, the score incorporates the proportion of meals achieving the desired (‘excellent’ and ‘good’) scores, while in the LFHF arm, the proportion of days in which dietary fat ≤35%, saturated fat ≤10%, carbohydrates ≥45% and dietary fibre ≥15 g. The dietician contacts individuals who need encouragement to achieve better scores. Participants of the two treatment groups receive the same attention according to their adherence. Time devoted to each individual by the dietician is recorded for later analysis purposes. Satisfaction with the diet is assessed using the Diet Satisfaction Questionnaire30 at 6 months (table 1).

Adherence to the metformin is based on pill counting at the 6 months’ visit and on logs of daily dose using the app or logbooks.

Physical activity and other confounders

Participants are asked to maintain the same level of physical activity throughout the study. Physical activity is monitored at two time points during the study using ActivePal (table 1). The device records time (start and duration) and type (quiet, standing and steps) of activity and 14 days’ worth of data, stored in the device, downloaded on return. Information about background medications and nutritional supplements is collected before the start of the treatment using questionnaires (table 1). Participants are asked to report any change in medications at each of the study visits and use the medication screen in the smartphone app.

Primary and secondary outcomes: measurements

Participants attend the Clinical Research Facility five times during the study, over approximately 7 months. Primary and secondary endpoint measures are collected before the start of the intervention (across two visits: screening/enrolment and baseline) and at 3 and 6 months of treatment. Table 1 outlines the measurements obtained at each of the study visits/events.

Study outcomes

The primary outcome measure is change in HbA1c from baseline to 6 months of treatment. Furthermore, a comprehensive glycaemia assessment is enabled through continuous glucose monitoring. Interstitial glucose concentrations are recorded every 15 min using CGM for 14 days before the start of the intervention (in the run-in period) and after 3 and 6 months of intervention. The sensor stores the data for the duration of the recording, while the participants are blinded to the glucose readings. The data are downloaded on return of the sensor. Time of the day with glucose readings below 7.8 mmol/L before versus after the treatment will be compared. Glycaemic variability,31 including (1) mean amplitude of glucose excursion (a measure of the variation of glucose concentrations from the mean), (2) the SD and the (3) mean postprandial area under the curve will also be assessed. Furthermore, fasting plasma glucose, 1-hour and 2-hour plasma glucose post 75 g glucose assessment are repeated after 6 months and will be compared with the baseline values. HbA1c test is repeated 6 months after treatment cessation (at 12 months, table 1), along with a short questionnaire, including weight, diet and medication status.

Weight, waist and hip circumferences are recorded at 3 months and 6 months of treatment and compared with baseline. Fat, fat-free mass and android/gynoid fat distribution and visceral fat measurements by DXA are repeated at 6 months of treatment.24 Similarly, resting energy expenditure and fat/carbohydrate oxidation is measured after 6 months of treatment.

Hepatic steatosis is common in pre-diabetes and T2DM1 32 and prevention of liver steatohepatitis is key target in individuals with pre-diabetes or T2DM. Metformin primarily targets the liver, inhibiting lipogenesis and increasing fatty acid oxidation; therefore, a beneficial effect on liver lipid and fibrosis with metformin has been assumed.33 However, comprehensive meta-analyses of randomised clinical trials concluded that reduction in both steatosis and fibrosis with metformin were underwhelming.34 35 Liver fibroscan measure is repeated after 6 months of treatment.

Blood pressure and pulse wave analysis measurements are repeated at 3 and 6 months of treatment and serum lipids measured after 6 months of treatment.

Safety/adverse events monitoring

Gastrointestinal side effects are the most common adverse effects of metformin and may occur in 20%–30% of individuals.36 37 Specifically, abdominal discomfort, nausea, diarrhoea and anorexia are common.36 While the gastrointestinal adverse effects are transient, in approximately 5% of individuals the symptoms may persist and result in cessation of metformin.37 Vitamin B₁₂ concentrations may be lower with metformin, if metformin is administered for a long duration.38 39 The mechanism(s) responsible for the lower plasma B₁₂ concentrations are unclear. A very rare, but potentially fatal complication of metformin use is lactic acidosis, mainly in patients with renal impairment.36 40 In PREDICT, individuals with severe renal impairment are excluded. Metformin is titrated over 3 weeks to negate the potential gastrointestinal side effects. Adverse events are recorded and monitored over the phone and during the study visits.

Statistical analysis

Laboratory testing

HbA1c is analysed using high performance liquid chromatography (Bio-Rad D-100, Bio-Rad Laboratories), plasma glucose using the Cobas 8000 (Roche), and liver and renal function tests using the Atellica platform (Siemens). Serum lipid profile is analysed by a spectrophotometric assay (Advia 2400 Chemistry System (Siemens Medical Solutions Diagnostics)), with LDL calculated using the Friedewald equation. Metagenomic DNA from the stool samples is purified using DNeasy PowerMag Soil DNA extraction kit (Qiagen) optimised for Tecan automated platform. Next-Generation Sequencing libraries are prepared using Nextera DNA library prep (Illumina) and sequenced on a NovaSeq sequencing platform (Illumina). Sequencing is performed with 100 bp single end reads with the depth of 10 million reads per sample. Host DNA is detected by mapping reads to the human genome with inclusive parameters, and those reads removed. Bacterial relative abundance estimation is performed by mapping bacterial reads to species-level genome bins representative genomes.44 Mapping is performed using Bowtie45 and abundance estimated by calculating the mean coverage of unique genomic regions across the 50% most densely covered areas, as previously described.46

Confidentiality and data storage

Each participant is associated with an individual program-generated code used to identify their study documents, data and specimens collected during the study. The re-identifiable code is documented in the participant’s record and on all study documents. Study data are collected and managed using REDCap electronic data capture tools47 48 hosted at the Garvan Institute of Medical Research. Some coded data are shared with essential personnel at the Weizmann Institute of Science on institutional Dropbox. Data collected in the form of paper hard copies are kept in locked cabinets and electronic files on a password-protected folder with access granted to the Garvan study team. Re-identifiable blood, stool, plasma and serum samples will be kept at the Garvan Institute’s freezer facility. All the study questionnaires are disseminated using REDCap.

Dissemination of results

The results of the study will be disseminated to healthcare professionals via open access publications in medical journals, without any restrictions. On completion of data analysis, the participants will be invited to an information session at the Garvan Institute of Medical Research with the study investigator(s) where the findings of the study will be shared and discussed. Individual letters are disseminated to the study participants after the 6 months’ treatment visit (approximately 7–8 months from study enrolment) summarising individual results (eg, baseline and post-treatment weight, body fat, liver fat, HbA1c, fasting, 1-hour and 2-hour plasma glucose concentrations). The participants are encouraged to share their individual results with the GP.

Ethics and dissemination

The study has been approved by the SVH Human Research Ethics Committee (HREC) (File 17/080, Sydney, Australia) and the Weizmann Institutional Review Board (IRB) (File 528-3, Rehovot, Israel). Protocol modifications are communicated to the SVH HREC, the Weizmann IRB, the trial registry (ClinicalTrials.gov), the study investigators and the study participants (if relevant). De-identified participants’ data that underlie the findings reported in the research article will be available immediately following publication, ending 5 years following the article publication, to researchers who provide a methodologically sound proposal with the aim to achieve the aims reported in the approved project proposal.