The Sugar Challenge Study was designed to ascertain whether the use of the Sugar AI app along with CGM could reduce hyperglycemia in individuals with early stages of glucose dysregulation who might benefit from lifestyle changes. The Sugar AI app allows participants to log their food, water, medication, and physical activity. Glycemic index (GI), load (GL), and macronutrients for each logged food were calculated and presented information back to the participants in addition to daily summaries of nutrition and activity patterns in relation to glycemic excursions. The Sugar AI app did not provide specific dietary or food recommendations. This prospective, unblinded observational trial included volunteers from 47 US states plus the District of Columbia, spanned a wide demographic range, and included a wide range of glycemia, from healthy individuals to those with prediabetes or non-insulin-treated T2D (Table 1). Participants were provided a heart rate monitoring (HRM) device and Abbott Libre CGM and were instructed to wear them for 10 consecutive days while scanning their sensor with Abbott’s LibreLink app at least once every 8 h to maintain signal capture. They were also asked to log their physical activity, food, medication, and water consumption. Participants were subsequently able to visualize glucose curves, along with nutritional and activity summaries (Fig. 1), on the Sugar AI app on their mobile device. The Sugar AI app was designed by January AI specifically for this study. The primary endpoint of the study was a comparison of glucose TIR after 10 days of use as compared with baseline.
Recruitment for the study occurred via social media channels and online classified ads, targeting individuals from diverse geographic, socioeconomic, age, and education levels. Respondents were required to provide written informed consent and complete an online screening demographic and health questionnaire. Eligible participants could be healthy or have a diagnosis of prediabetes or non-insulin-dependent T2D (Table 1). Eligible participants were required to wear a HRM device 20–24 h/day, and flash CGM for 10 consecutive days, provide body weight at baseline, and agree to comprehensive logging of food intake, physical activity, medication, and water consumption for 10 consecutive days using the Sugar AI app. Individuals who met the online eligibility criteria were further screened by clinical coordinators via telephone who confirmed eligibility and completed enrollment. Individuals were excluded from the study if they did not meet the criteria described above or if any of the following criteria were met: use of vitamin C supplements in excess of 200% of the US recommended daily allowance at least 14 days prior to starting the trial; allergy to skin adhesives used in the trial; women who were pregnant, lactating, had given birth in the past 6 months, or were planning to get pregnant in the next 6 months; and individuals who were taking any of the following medications: insulin, oral hypoglycemic medications (sulfonylureas, meglitinides), progesterone, atypical antipsychotics, oral corticosteroids, triphasic oral contraceptives, blood thinners, deemed unfit for participation by the study coordinator, and allergy to nuts. Of the 192 people with diabetes 152 took medication; of these, 97 were on metformin and the rest were on other medications. Participants that completed the Sugar Challenge Study as outlined were remunerated with a US $50 Amazon gift card.
Because not all subjects who completed the 10-day study wore the CGM continuously as instructed, for the purposes of analysis, it was predetermined that only those who wore the CGM for at least 9 days with at least 33% daily coverage were included in the final analysis. In accordance with these criteria, 665 subjects out of 1022 enrolled were included in this study.
Enrolled participants were mailed an Abbott Freestyle Libre flash CGM (henceforth referred to as CGM) a MiBand 3 or Garmin watch to record heart rate (HRM), and instructed to download the Sugar AI app through which they could search a large food database of over 15 million foods and log all food, physical activity, water, and medications taken. Participants were provided with two nutritional bars and a dextrose solution (TRUEplus, or CVS Health Glucose Shot; 15 g) to generate glycemic responses to standardized nutrient challenges. The Sugar AI food database contains nutritional information for each logged food. GI and GL values calculated through January AI technology were also available for each food in the database. Participants were guided to scan their CGM using the Abbott LibreLink app which retrieves up to 8 h of CGM values. Aside from the standard glucose drink (day 3) and the two nutritional bars, participants determined their own dietary and activity patterns but were required to log their food, water, and physical activity for the duration of the study. No dietary recommendations were made while enrolled in the Sugar Challenge Study.
GI, GL, total calories, macronutrient content, along with heart rate, and water consumption were displayed to the participants via a tracker feature that overlays multiple data components on the CGM measurements and heart rate signals (Fig. 1b). The overlay of nutritional components of logged foods on CGM glucose curves was presented to the subjects at the time of logging. To encourage participant engagement, a point system was devised, in which participants were encouraged to attain at least 10,000 points by logging food, water, medication, and physical activity (Supplementary Table S1).
Continuous Glucose Monitoring
The Abbott Freestyle Libre was provided at no cost to participants who were instructed to apply a 10-day sensor to the back of the arm following the manufacturer’s written instructions and demonstration video. Clinical coordinators were available by phone, video conferencing, or in-person to troubleshoot CGM sensor placement for those who had difficulties. The small CGM subcutaneous sensor measures interstitial glucose values every 15 min which can be transmitted using near-field communications. Participants obtained their glucose values after scanning for up to 8 h of data (i.e., Flash CGM). No alarms were used.
Structured Food Challenges
During the 10 days of monitoring, all subjects were asked to consume three standardized nutrient challenges in order to identify patterns of glycemia that were not dependent on variability in nutrients consumed:
Oral glucose challenge: on day 3 while wearing the CGM, after fasting for at least 10 h, participants (n = 328) drank over 5 min three 15-g portions of dextrose solution (TRUEplus or CVS Health Glucose Shot), followed by 2 h of inactivity to obtain a standardized glycemic response pattern via CGM.
Mixed meal challenges: on day 5 of the challenge after fasting for at least 10 h, participants (n = 212) consumed a high-protein bar (Garden of Life; S’mores) with the following nutritional content: 200 cal per serving [total carbohydrates (25 g), fiber (14 g), protein (14 g), and total fat (8 g)] followed by 2 h of inactivity to obtain standardized glycemic response pattern via CGM. On day 7 of the challenge after fasting for at least 10 h, participants (n = 286) consumed a Kind bar (Dark Chocolate Nuts & Sea Salt) with the following nutritional content: 180 cal per serving [total carbohydrates (16 g), fiber (7 g), protein (6 g), and total fat (15 g)].
Assessment of Time in Range
As a result of potential inaccuracy of CGM during the first and last 24 h of sensor use, we defined the change in percentage TIR during days 8–9 relative to the percentage TIR at baseline which was defined as days 2–3 (excluding day 3 for those who took the glucose shot on day 3) of the 10-day study period. TIR was defined as the percentage of time glucose measurements were 54–140 mg/dL (3.0–7.8 mmol/L) for healthy and prediabetes and 54–180 mg/dL (3.0–10.0 mmol/L) for diabetes. We used 54 mg/dL as the lower limit for TIR as healthy individuals can have nonpathologic glucose values below 70 mg/dL. The American Diabetes Association has defined 54 mg/dL  as clinically significant hypoglycemia and this limit is used to diagnose hypoglycemic disorders in patients who are not taking glucose-lowering medications . Thus, we chose this lower limit so as to prevent overclassification of hypoglycemia in nondiabetic individuals. Upper limits of 140 mg/dL and 180 mg/dL were chosen for healthy/prediabetes and T2D, respectively.
Defining the Group with Poor TIR at Baseline
Because our analysis was to determine whether the use of CGM could improve TIR, we conducted a secondary analysis on individuals with suboptimal glucose patterns defined as follows: for healthy and those with prediabetes we used TIR of less than 83%, which corresponds to A1c of less than 5.7%, and for those with T2D we used TIR of less than 72%, which corresponds to A1c of less than 6.5% [25, 26].
Responders and Non-Responders
Self-reported demographic and clinical characteristics of the best and worst responders were compared to determine which participant might be the best candidate for use of CGM plus Sugar AI app. Responders were defined as the top quartile of change in TIR (from baseline to end of study), and non-responders were defined as the bottom quartile of change in TIR within the subgroups with suboptimal baseline glucose profiles as defined above.
The TIR for the last 2 days vs TIR for the first 2 days of complete data capture was analyzed via the nonparametric Wilcoxon signed-rank test [27, 28]. Mann–Whitney rank test  was used to compare the continuous variables (e.g., age, BMI) between the good responders and bad responders. Comparisons were considered statistically significant with p < 0.05. To account for the statistical analyses of multiple independent variables in this study, the Benjamini–Hochberg procedure (BH) was employed to control the false discovery rate (FDR) [30, 31] at the level of 0.05. These analyses were performed on the subset of participants who provided the required information detailed in Table 1.
Compliance with Ethics Guidelines
All named authors meet the International Committee of Medical Journal Editors (ICMJE) criteria for authorship for this article, take responsibility for the integrity of the work as a whole, and have given their approval for this version to be published. This study was performed in accordance with the Helsinki Declaration of 1964 and its later amendments. The study was approved by WCG New England IRB 120190429.