A new study finds that machine learning can predict differences between people in how they respond to meals If you are managing to stay lean in today’s obesogenic environment (lucky you!), you might suspect that it’s your good genes and/or disciplined adherence to a healthy diet and lifestyle. On the other hand, if you’ve struggled with body fat and dieting for most of your life, you live in hope there’s a specific type of diet (or perhaps better still, a drug) that’s perfect for you… if only you could identify which one. Enter Personalised Nutrition.
In the June issue of the prestigious journal Nature Medicine, Sarah Berry and her colleagues took a major step in that direction. They presented the findings of PREDICT (1), a large-scale study involving ~1000 people, including twins and other adults from the US and UK. Using machine learning, the goal was to use the data to derive ‘algorithms’ (mathematical formulas) that predict a person’s postprandial (after-meal) responses, that is, the rise in glucose, insulin and triglycerides (fats) in the blood after meals of varying composition.
The end-game of this kind of research is the ability to give scientifically valid ‘personalised’ dietary advice based on factors such as age, body mass index (BMI), specific genes, large bowel microbial flora (the “micobiome”) and postprandial responses.
But the findings were not what they expected. They found much more person-to-person variation than was expected, but differences in genes, the gut microbiome and insulin levels explained only a minor proportion of the differences.
By contrast, they were surprised to find a person’s response to the same foods was fairly predictable and reproducible. Food composition and macronutrient (carbohydrate, fat and protein) distribution explained some of the variation in post-meal blood glucose levels, but not in triglyceride levels. And interestingly, blood glucose responses did not predict triglyceride levels; indeed, they warned that advice based just on glucose responses (such as flash glucose monitoring) alone would be misleading.
From our point of view, the associations between the carbohydrate content of meals, post-meal blood glucose levels and other factors were among the most interesting findings. High blood glucose levels after meals are a well-established predictor of type 2 diabetes, the metabolic syndrome, fatty liver, and cardiovascular disease (2).
We have known for a long time that people vary widely in their ‘glucose tolerance’, i.e. the absolute blood glucose response to a carbohydrate challenge. In a lean, active person, the area under the curve (AUC) after a 50 g glucose challenge can be as low as 50 units, but in a sedentary person with a family history of type 2 diabetes, it can be 400 units, an 8-fold difference. Higher AUC means the beta-cells (insulin producing) in the pancreas are working hard. If you have a family history, your pancreas may not have what it takes to do this without becoming dysfunctional over time.
We know that glucose tolerance worsens (measured as higher AUC) with age, increased body weight and sedentary lifestyle. We also know that the background diet is important – low carbohydrate consumption is associated with a higher glycemic response to a glucose challenge. However, it’s reversible - just a day or so of higher carbohydrate intake will improve glucose tolerance.
Is there an optimal diet composition for your body? Is one diet better than another for you but not me? Does human evolution play a role here? Yes! Many different diets can reduce blood glucose responses on a day-to-day basis. Indeed, we have argued that this is one of key mechanisms behind the success of the Mediterranean diet, low GI diets, vegetarian diets based on legumes and lower carbohydrate diets.
Logically, reductions in blood glucose can also be achieved with carefully planned, very-low-carbohydrate diets (50-100 g/day), with parallel improvements in body weight and HbA1c (glycated haemoglobin) in people with type 2 diabetes (3). However, it would be very easy to choose a poor quality very-low-carbohydrate diet and it may be hard to sustain in the longer-term. It may not be as effective (or as easy) as changing the kind (quality) of carbohydrate.
For a given amount of carbohydrate, the glycemic index of a food predicts the degree of glycaemia relative to a standard reference food. Choosing a diet based on low GI foods such as pasta, legumes, most fruit, milk, yogurt and specific types of rice and bread can halve the AUC and reduce HbA1c in individuals with diabetes. Furthermore, meta-analyses of observational studies confirm that diets based on low GI food choices are associated with reduced risk of type 2 diabetes (4) and cardiovascular disease (5). The relative risk reduction is biologically significant, similar to increasing the amount of exercise or dietary fibre.
In our view, the potential of personalised nutritional guidance versus standard advice (national dietary guidelines) to improve weight control is far from proven. In many ways, the findings of PREDICT are important because they challenge so much of the prevailing hype.
REFERENCES:
- Berry S, and colleagues. Decoding human postprandial responses to food and their potential for precision nutrition: the PREDICT 1 study.
- The DECODE group. European Diabetes Epidemiology Group. Glucose tolerance and mortality: comparison of WHO and American Diabetes Association diagnostic criteria.
- Wycherley TP, and colleagues. Effects of energy-restricted high-protein, low-fat compared with standard-protein, low-fat diets: a meta-analysis of randomized controlled trials.
- Livesey G, and colleagues. Dietary Glycemic Index and Load and the Risk of Type 2 Diabetes: A Systematic Review and Updated Meta-Analyses of Prospective Cohort Studies.
- Livesey G, and colleagues. Coronary Heart Disease and Dietary Carbohydrate, Glycemic Index, and Glycemic Load: Dose-Response Meta-analyses of Prospective Cohort Studies.
Professor Jennie Brand-Miller holds a Personal Chair in Human Nutrition in the Charles Perkins Centre and the School of Life and Environmental Sciences, at the University of Sydney. She is recognised around the world for her work on carbohydrates and the glycemic index (or GI) of foods, with over 300 scientific publications. Her books about the glycemic index have been bestsellers and made the GI a household word.