CGM may be useful tool for tailoring ‘precision diets’ to improve cardiometabolic health
December 10, 2020
3 min read
McLaughlin T. Continuous glucose monitoring for precision diets. Presented at: World Congress on Insulin Resistance, Diabetes & Cardiovascular Disease; Dec. 3-6, 2020 (virtual meeting).
McLaughlin reports she serves on an advisory board for Januaryai.inc.
Artificial intelligence can be leveraged to tailor a diet to a person’s specific physiology, and continuous glucose monitors could “train” such algorithms to recognize patterns after specific foods are consumed, according to a speaker.
Dietary weight-loss benefits differ among nondiabetic people with insulin resistance vs. insulin-sensitive individuals, and studies demonstrate high interpersonal variability in postprandial glycemic responses to the same meal, Tracey McLaughlin, MD, MS, professor of medicine at Stanford University School of Medicine, said during an online presentation at the virtual World Congress on Insulin Resistance, Diabetes & Cardiovascular Disease. Using AI, data can be captured from these individuals and entered into a computer algorithm that identifies patterns. With training, predictions can be made regarding expected glucose elevations after specific foods, activities or sleep patterns, thereby creating a “precision diet,” McLaughlin said.
CGM — typically used for tracking glucose excursions for people with diabetes — can instead be harnessed to “train” such algorithms to predict the glucose responses, McLaughlin said.
“We are in the midst of all of these dietary recommendations, and the question we have is, does one glove fit all?” McLaughlin said. “It has always been my belief that it doesn’t. We have shown over many years that weight-loss benefits and dietary changes, particularly with macronutrients, have effects on insulin-resistant individuals compared with insulin-sensitive individuals, who are otherwise the same.”
Same meal, different response
There are not many studies assessing whether precision diets work, McLaughlin said; however, promising data from some trials suggest they might.
In a study published in 2015 in Cell, researchers monitored weeklong glucose levels for 800 participants using CGM, measuring responses to 46,898 meals, and found high variability in the response to identical meals, suggesting that universal dietary recommendations may have limited utility. The researchers devised a machine learning algorithm that integrated blood parameters, dietary habits, anthropometrics, physical activity and gut microbiota measured in the cohort.
A masked, randomized controlled dietary intervention based on the algorithm resulted in significantly lower postprandial responses and consistent alterations to gut microbiota configuration, McLaughlin said.
“What they showed is if you try to predict the postprandial glucose response from the meal carbohydrates or the meal calories, it wasn’t very good,” McLaughlin said. “But their model was actually quite a bit better. Once it was trained, it was better at predicting the actual postprandial glucose response.”
In the more recent PREDICT study, published in Nature Medicine in 2020, researchers analyzed data from 1,002 twins and unrelated healthy adults in the United Kingdom and assessed postprandial metabolic responses in a clinical setting and at home. Researchers observed considerable interindividual differences in postprandial metabolic responses to the same meals, challenging the logic of standardized diet recommendations, McLaughlin said. The researchers developed a machine learning model that predicted triglyceride and glycemic responses to food intake.
“These findings, in addition to the scalability of the assessment methods and the accuracy of the prediction algorithms described here, mean that, at least from a cardiometabolic health perspective, population-wide, personalized nutrition has potential as a strategy for disease prevention,” McLaughlin said.
Constructing a ‘best’ diet
McLaughlin and researchers from the Stanford Precision Health and Integrated Diagnostics Center, or PHIND, are currently collecting data from volunteers using CGM (Dexcom), sleep and activity trackers (FitBit) and a food logging app (Cronometer), in addition to performing metabolic, genomic and microbiome testing. Using AI, researchers are attempting to construct a “best” diet for each participant based on CGM responses to standardized test meals, detailed food diaries, activities, sleep patterns and omics profiles.
“By the end of the study … either an expert dietitian or a computer algorithm is recommending what foods to eat based on the ‘training’ that goes on during the food logging, and the glucose monitoring can link the various foods, timing of meals and other lifestyle factors to the glucose excursions,” McLaughlin said.
Much like the other studies, McLaughlin and colleagues have observed substantial glucose variability between participants who received the same meals, she said.
“Even when comparing [results] with their baseline diet and not a ‘bad’ diet, it looks like 60% of individuals can benefit from learning these patterns,” McLaughlin said. “They also didn’t experience as much hypoglycemia. If you’re decreasing the highs and the lows, that tells you that you cannot just look at the mean [glucose], which would be HbA1c. You really need to look at these patterns.”
Berry SE, et al. Nat Med. 2020;doi:10.1038/s41591-020-0934-0.