Reducing HbA1c in Type 2 Diabetes Using Digital Twin Technology-Enabled Precision Nutrition: A Retrospective Analysis

This article is published with digital features to facilitate understanding of the article. To view digital features for this article go to https://​doi.​org/​10.​6084/​m9.​figshare.​12943226.

Type 2 diabetes (T2D) is a chronic disease that has typically been thought of as incurable. Often the best outcome hoped for is improvement in the disease’s symptoms and slowing of its progression [1]. Typically, disease progression leads to increases in glucose-lowering medication use, healthcare costs, and complications [2]. The 2016 World Health Organization global report on diabetes acknowledged that diabetes reversal is possible through weight loss and reduced calorie intake [3]. Although specific criteria have not been finalized, it is commonly accepted that maintaining hemoglobin A1c (HbA1c) below 6.5% for an extended period of time without the use of insulin or oral hypoglycemic medications (with the possible exception of metformin) would be considered diabetes reversal or remission [1, 4].

Various studies have shown reductions in weight and improvements in glycemic control using bariatric surgery, low-calorie diets, or carbohydrate restriction [1, 5]. Limitations that may hinder the use of bariatric surgery include the high cost of surgery and an increased likelihood of adverse events [1]. The limitation of using a low-calorie diet (e.g., < 800 kcal/day) to manage diabetes is that the diet may not be sustainable over the long term [2]. The studies using carbohydrate restriction found improved glycemic control, but were often only short-term trials, only included small groups, focused on morbidly obese patients, and/or excluded subjects taking insulin [2, 6‐10].

Despite the importance of good glycemic control and the association of postprandial glycemic response (PPGR) with diabetes complications, predicting the impact of specific foods on PPGR has been challenging due to the high variability in different people’s response to the same food [11]. Some attempts have been made to estimate PPGR using the carbohydrate content of meals [11] or using glycemic indices [12]. Few studies have attempted to build or test predictive models of PPGR based on specific foods eaten in individual meals [11, 13, 14]. None of the studies to date have combined long-term continuous glucose monitoring (CGM), artificial intelligence methods, and precision nutrition in an attempt to reverse diabetes.

The Twin Precision Nutrition (TPN) Program uses detailed patient food intake information and CGM values as inputs to a machine learning predictive model and then provides daily precision nutrition guidance to the patient. The objective of this study was to examine changes in HbA1c, anti-diabetic medication use, insulin resistance, and other ambulatory glucose profile (AGP) metrics between baseline and after 90 days of participation in the TPN Program in patients with T2D. We hypothesized that each metric would improve significantly over the 90-day treatment period.

A total of 89 patients were enrolled in the TPN Program. Of these, 16 patients discontinued the program before 90 days of follow-up due to difficulties completing the study-related procedures (e.g., wearing the CGM for a prolonged period) or because of life-altering events, such as family emergencies. Among the remaining 73 patients, nine were eliminated from the analysis due to significant non-adherence to the nutritional instructions provided by the TNP Program. Of the 64 patients who remained in and adhered to the TPN Program, 19 (29.7%) were women. At enrollment, the average (SD) age of the 64 patients was 52.4 (10.0) years; average duration of T2D was 8.4 (6.5) years; average BMI was 29.2 (5.8) kg/m²; and 18.8% were taking insulin.

The mean (SD) HbA1c of the 64 patients decreased from 8.8% (2.2%) at baseline to 7.7% (1.6%) at 30 days and to 6.9% (1.1%) at 60 and 90 days (all p < 0.0001 vs. baseline) (Fig. 1). Six of the 64 patients (9.4%) had an HbA1c < 6.5% at baseline. The percentage of patients with measured HbA1c < 6.5% increased to 24.6% at 30 days, 38.0% at 60 days, and 39.7% at 90 days (all p < 0.003 vs. baseline) (Fig. 2). The eA1c also decreased significantly over the study period, from 6.9% (2.3%) at baseline to 5.7% (1.0%) at 30 days, 5.8% (1.0%) at 60 days, and 5.9% (1.2%) at 90 days (all p < 0.001 vs. baseline). The patients’ mean body weight decreased from 79.0 (16.2) kg at baseline to 76.1 (15.3) kg at 30 days, 75.1 (15.0) kg at 60 days, and 74.2 (14.7) kg at 90 days (all p < 0.0001 vs. baseline) (Table 1).

Mean (SD) FBG fell from 151.2 (45.0) mg/dl at baseline to 135.2 (38.1) mg/dl at 30 days (p  = 0.0043), 131.5 (31.9) mg/dl at 60 days (p = 0.0205), and 129.1 (36.7) mg/dl at 90 days (p = 0.0001) (Table 1). At baseline, 34.4% of patients had FBG < 126 mg/dl, whereas after 90 days on the TNP Program 59.7% of patients had FBG < 126 mg/dl.

Mean HOMA-IR (SD) fell from 7.4 (3.5) at baseline to 3.1 (2.5) at 30 days, 3.3 (2.4) at 60 days, and 3.2 (2.8) at 90 days (all p < 0.0001 vs. baseline). The change from baseline to 90 days was a 56.9% reduction (Fig. 3).

Mean (SD) glucose levels obtained from CGM fell from 147.4 (59.1) mg/dl at baseline to 116.7 (27.9) mg/dl at 30 days, 120.5 (29.6) mg/dl at 60 days, and 122.6 (33.3) mg/dl at 90 days (all p < 0.001 vs. baseline). Similarly, mean C-peptide values decreased from 3.2 (2.2) ng/ml at baseline to 1.7 (0.9) ng/ml at 30 days, 2.1 (0.9) ng/ml at 60 days, and 1.9 (0.9) ng/ml at 90 days (all p < 0.0001 vs. baseline) (Table 1). Some of this decrease was driven by large decreases in patients with very high baseline C-peptide values, which resulted in closer-to-normal C-peptide values for those patients.

The increase in average number of hours of sleep per night was not significant, but mean (SD) number of steps taken per day increased from 4677.4 (2804.9) steps per day at baseline to 7004.1 (3999.0) steps at 90 days (p < 0.0001). Median resting heart rate decreased from 75.0 bpm at baseline to 71.0 bpm at 90 days (p < 0.0001) (Table 1). Mean calories consumed per day decreased from 1884.0 kcal (326.5) at baseline to 1649.6 (310.1) kcal at 30 days, 1607.4 (311.4) kcal at 60 days, and 1573.0 (373.1) kcal at 90 days.

During the first week of the program, 55 patients had TIR values > than 70%. This result was maintained, and at 90 days 57 patients had TIR values > than 70%.

In only one patient was the percentage of time > 250 mg/dl (TAR level 2 of 11.9%) higher than the accepted threshold (< 5%) [17] during the first week of the program; this was subsequently reduced to 0.5% at the end of 3 months of treatment. Two patients had an initial TAR level 1 that was above the accepted threshold (< 25%) [17]; these were reduced from 41.7 and 71.3% during the first week to 2.2 and 15.8%, respectively, after 3 months.

Eighteen patients had unacceptable [17] (> 4%) levels of TBR level 1 during the first week of the program; this number had improved at the end of 3 months to 12 patients with TBR level 1 > 4%. Thirteen patients had unacceptable [17] (> 1%) levels of TBR level 2 during the first week of the program; this number improved at the end of 3 months to only four patients with TBR level 2 > 1%.

Mean TAR level 2, TAR level 1, and TIR values reached normal ranges by the end of the first week of the TNP Program and were maintained through the 90 days of the follow-up. The mean (SD) TIR in the first week was 87.1% (16.8%) and continued as such throughout the 90-day period, ending at 87.1% (19.4%). Mean TBR level 1 values were slightly above the normal range in the first week and at 90 days. Mean TBR level 2 values were slightly above the normal range in the first week and then decreased further to well within the normal range by 90 days (Table 2).

All 12 patients who were on insulin at baseline stopped insulin within 3 months after the start of the program. Of the 56 patients who were on metformin at baseline, 38 were able to stop taking metformin. One patient who was not on metformin at baseline began taking metformin during the 90-day period. Of the 28 patients on DPP-4 inhibitors at baseline (sitagliptin, vildagliptin, or teneligliptin), 26 were able to stop DPP-4 inhibitors. One patient not on sitagliptin at baseline began taking sitagliptin, and one patient switched from vildagliptin to sitagliptin. All 13 patients taking an alpha-glucosidase inhibitor medication (voglibose) at baseline were able to stop taking this medication. All 34 patients taking sulfonylureas (glibenclamide/glyburide, gliclazide, glimepiride, or glipizide) were able to stop taking them. Both patients taking pioglitazone at baseline had stopped by 90 days of follow-up. One patient began taking pioglitazone during the 90-day period. All ten patients taking sodium-glucose cotransporter-2 (SGLT2) inhibitors at baseline (dapagliflozin or empagliflozin) stopped taking their SGLT2 inhibitors prior to the end of the 90 days. The one patient taking a glucagon-like peptide 1 (GLP-1) analogue (liraglutide) stopped as well (Fig. 4). Six of the 64 patients had no medicines at baseline.

We observed no cases of diabetic ketoacidosis. No new cases of gout were diagnosed, and no episodes of symptomatic hypoglycemia were reported. There were no other serious adverse events during the study. Non-serious adverse events, such as headache, tiredness, or constipation, occurred primarily early in the study and were transient.

The TPN Program uses CGM, Internet of Things technology, food intake data, and machine learning algorithms to provide daily precision nutrition guidance to individual patients with T2D so that they can avoid foods that cause blood glucose spikes and replace them with foods that do not produce glucose spikes. The results of the study are consistent with the hypothesis that each diabetes-related outcome was significantly improved over the 90 days of the program.

The results from the present study compare well with those from prior investigations on interventions in diabetes patient populations. Studies of bariatric surgery have shown that normal HbA1c and glucose levels can be achieved in most patients following surgery, sometimes within a few days [1, 18]. Nearly 39.7% of the patients in the current study reached HbA1c < 6.5% within 90 days of initiating the TNP Program and 38.0% reached that level within 60 days. Studies have found that diets that are very low in calories quickly improve glucose and HbA1c levels and lead to discontinuation of insulin use [1, 19]. However, many low-calorie diet studies have also found that patients find these diets difficult to maintain and that the improvements in diabetic endpoints could not maintained long-term [20‐23]. All patients taking insulin in the present study discontinued the use of insulin, with most oral anti-diabetic medications also discontinued. In the very low-calorie study by Steven et al. [19], 40% of patients reached FBG < 126 mg/dl (the threshold for partial remission of diabetes [24]) at 10 weeks when isocaloric eating was resumed, and 43% of patients continued to have FBG < 126 mg/dl at 6 months. In the current study, 59.7% of patients had FBG < 126 mg/dl after 90 days. In a study of morbidly obese diabetic patients on a low-carbohydrate diet, the HbA1c of these patients decreased from 7.6 to 6.6% in 10 weeks [2]; by 10 weeks, the percentage of people with HbA1c < 6.5% had increased from 20 to 56%, and 57% had one or more diabetes medications reduced or eliminated [2]. Within 1 year, the HbA1c had dropped to 6.3%, 94% of these patients had stopped taking insulin, and 100% had stopped taking sulfonylureas [5]. In comparison, the patients in the current study started with a higher average HbA1c (8.8%) and had a larger decrease (1.9 percentage points) in HbA1c and a higher rate of elimination of diabetes medications after 90 days (13 weeks). In a study of patients with T2D on a low-carbohydrate ketogenic diet [7], HbA1c decreased from 8.8 to 7.2% and FBG fell from 178.1 to 156.4 mg/dl over a 12-week period. This decrease in HbA1c is slightly less than that observed in the present study, but the percentage decrease in FBG is the same in both studies (12.2%).

Reductions in HbA1c and HOMA-IR have been shown to be correlated with lower health risk in patients with diabetes. Stratton et al. showed that a 1.0% reduction in HbA1c is associated with a 37% reduction in diabetes microvascular complications [25]. Bonora et al. reported that lower HOMA-IR values independently predicted lower cardiovascular disease prevalence and incidence [26]. The 1.9 percentage point decrease in HbA1c and 56.9% reduction in HOMA-IR, combined with fewer medications, that was achieved in our study population have the potential to lower the number of vascular complications in this patient population and thus result in cost savings over time. Increasing the TIR (even by only 5%) and decreasing TBR have also been shown to be associated with fewer diabetes complications [17]. TIR is strongly correlated with HbA1c, and each 10% increase in TIR corresponds with a 0.4–0.8 percentage point reduction in HbA1c [27, 28]. Additionally, the hazard rates for retinopathy progression and microalbuminuria have been shown to increase by 64 and 40%, respectively, for each 10% reduction in TIR [29]. Similarly, a post hoc analysis of the Diabetes Control and Complications Trial data showed a link between TBR glucose thresholds of < 70 mg/dL and < 54 mg/dL and an increased risk for severe hypoglycemia [30]. After 90 days on the TPN Program, 57 patients (89.1%) had TIR 70–180 mg/dl ≥ 70%. The number of patients in the TPN Program with unacceptable TBR level 1 values decreased from 18 patients at week 1 to 12 patients at 90 days, and those with unacceptable TBR level 2 values decreased from 13 patients at week 1 to only four patients at 90 days of follow-up.

The strengths of this study include the significant improvements in patient outcomes and the serial HbA1c testing with continuous AGP monitoring that shed new light on changes in AGP metrics. The nutritional intervention happened in an outpatient setting employing home-cooked foods, making the intervention sustainable in the long term. App-based logging of food by patients allowed for accurate assessment of nutrition.

The limitations of this study include its retrospective nature, a relatively small study population, and no control arm. Additionally, although most of the patients discontinued diabetes medications and reached the FBG threshold for partial remission of diabetes, the length of the study was not long enough to verify remission of diabetes. This will be the goal of subsequent randomized controlled studies.

The results provide evidence that daily precision nutrition guidance based on CGM, Internet of Things technology, food intake data, and machine learning algorithms can benefit patients with T2D. Diabetes-related endpoints in the 64 participants who adhered to the TNP Program for 90 days improved significantly over the course of the study. Following 3 months of the TPN Program, patients achieved a 1.9 percentage point decrease in HbA1c, a 56.9% reduction in HOMA-IR, a significant decline in TBR, and the elimination of diabetes medication use in most of the patients. Ongoing results will shed further light on the long-term effectiveness, safety, and sustainability of the TNP Program.