“No one diet fits all,” the Daily Mail reports.
Israeli researchers monitored 800 adults to measure what is known as postprandial glycemic response – the amount by which blood sugar levels increase after a person eats a meal. This measure provides a good estimate of the amount of energy that a person “receives” from food.
The researchers found high variability in postprandial glycemic response across individuals who consumed the same meals.
They found these differences were related to the individual’s characteristics, and developed a model (known as a “machine learning algorithm”) to predict an individual’s response to a given meal.
When 12 individuals were put on two different tailored meal regimens predicted by this model to either give lower blood sugar levels or higher levels for a week each, the prediction was correct in most of the individuals (10 of the 12).
Results of the study should be interpreted with some caution due to limitations. The main one is that the sample in which the diets were tested was small, with a short follow-up period. The study looked at post-meal blood sugar levels and not weight, so we cannot say what the impact on weight would be.
Still, the concept that a machine learning algorithm model could be used to create a personalised diet plan is an intriguing idea. In the same way Netflix and Amazon “learn” about your TV viewing preferences, the plan could “learn” what foods were ideally suited to your metabolism.
Where did the story come from?
The study was carried out by researchers from the Weizmann Institute of Science, Tel Aviv Sourasky Medical Center and Jerusalem Center for Mental Health – all in Israel.
The study was funded by Weizmann Institute of Science, and the researchers were supported by various different institutions, such as the Israeli Ministry of Science, Technology and Space.
The study was published in the peer-reviewed scientific journal Cell.
The Daily Mail’s reporting implies the study explains why different weight loss diets perform differently in different individuals, but we cannot say this based on the research.
The study only aimed to look at blood sugar levels after a meal – not weight. It also did not compare the personalised dietary plans the researchers developed against popular weight loss diet plans such asthe 5:2 diet.
What kind of research was this?
This study aimed to measure the differences in post-meal blood glucose levels between individuals and to identify personal characteristics that can predict these differences.
The researchers then used a small randomised controlled trial (RCT) to identify whether personalising meals based on this information could help reduce post-meal blood glucose levels.
Researchers say that blood sugar levels are rapidly increasing in the population. This has led to an increase in the proportion of people with “pre-diabetes” where a person has higher blood sugar than normal, but does not meet all of the criteria required for being diagnosed with diabetes. They say that up to 70% of people with pre-diabetes eventually develop type 2 diabetes.
The researchers hoped that by understanding the factors responsible for variations in post-meal blood glucose levels they could use this information to personalise dietary intake to reduce those levels.
What did the research involve?
This study started with 800 healthy and pre-diabetic individuals (aged 18-70 years). The cohort was representative of the individuals without diabetes in Israel. Just over half (54%) of the cohort was overweight and 22% were obese.
Researchers started by collecting data on food intake, lifestyle, medical background and anthropometric measurements (such as height and weight) for all the study participants. A series of blood tests was carried out and a stool sample (used to assess gut microbial profile) was also collected.
Participants were then connected to a continuous glucose monitor (CGM) over seven days. The machine was placed on the individual’s skin to measure glucose in interstitial fluid – the fluid in and around the body’s cells – every five minutes for a week. They were also asked to accurately record their food intake, exercise and sleep using a smartphone-adjusted website developed by the researchers.
Over this period, the first meal of each day was a standardised meal given to all participants to see how their blood glucose responses differed. Other than that, they ate their normal diets.
Researchers then analysed the relationship between an individual’s characteristics and their post-meal glucose levels. They developed a model based on these characteristics that would predict what these levels would be. They then tested their model on 100 other adults.
To assess whether personally tailored dietary interventions could improve post-meal blood sugar levels, researchers carried out arandomised crossover trial.
This trial included 26 new participants who were connected to continuous glucose monitors (CGM) and had the same information collected as the 800-person cohort over a week. This allowed the researchers to identify their personal characteristics and blood glucose responses to meals.
After this, the groups were allocated to two different personalised diets. One group (the “prediction” group) was allocated to receive a meal plan based on what the researchers’ model predicted to be a “good” or a “bad” diet for them. They received these two different meal regimens for a week each, in random order:
- one regimen was based on meals that were predicted to produce “low” post-meal blood sugar levels (good diet) in the individual
- one regimen was based on meals predicted to produce “high” post-meal blood sugar levels (bad diet) in the individual
The second group (the “expert” group) took part in a similar process, but their “good” and “bad” diets were based on what a clinical dietitian and researcher selected for them based on looking at the person’s responses to different meals in the first week of the study.
Participants and researchers did not know which meal plan they were eating during the study – so both groups were blinded.
What were the basic results?
Overall, the study found high variability in post-meal blood sugar levels across the 800 individuals even when they consumed the same meal. They found that many personal characteristics were associated with their post-meal blood glucose levels, including their body mass index (BMI) and blood pressure, as well as what the meal itself contained.
One example, given in an interview to the Mail, was the case of a woman whose blood sugar levels spiked dramatically after eating tomatoes.
The researchers developed a model based on these characteristics to predict their glucose levels after a meal. This model was better at predicting post-meal glucose levels than simply looking at how much carbohydrate or calories the meal contained. The model performed similarly well when tested in a different group of 100 adults.
The researchers found that most of the individuals on the “prediction” diet (10 out of 12; 83%) had higher post-meal blood glucose levels during their “bad” diet week than their “good” diet week. This was slightly better than the “expert” diet – where eight out of 14 participants (57%) had higher post-meal blood glucose levels during their “bad” diet week.
How did the researchers interpret the results?
Researchers concluded that this research suggests: “personalised diets [including the one based on their algorithm] may successfully modify elevated postprandial blood glucose and its metabolic consequences”.
This study assessed the differences in post-meal blood sugar levels – medically known as postprandial glycemic responses (PPGR) – across 800 non-diabetic adults, and found a lot of variation between individuals.
They developed a model based on a wide range of personal characteristics, such as a person’s BMI and gut microbial profile, which could predict their response to a given meal.
In a small crossover study, it found that tailoring meals for individuals based on their model could help lower the individual’s post-meal sugar levels.
This study has some strengths and limitations. Its strengths include the relatively large sample size used to analyse the relationship between personal characteristics and post-meal blood sugar levels, and the fact the model they developed was then checked in a new group of individuals.
The main limitation of this study is that the actual testing of the personalised diets was done in a small sample of only 26 people, with only 12 of these getting the diet based on the model’s predictions.
What we can say based on these results is also limited based on its short follow-up period and the fact that only blood glucose levels were measured. We cannot say what effects these different diets have on a person’s weight or risk of diabetes in the long term.
It appears the research team is now looking into finding commercial applications for this approach. It would be feasible to combine a continuous glucose monitor with a smartphone application that creates a personalised diet plan. If successful, such an application would likely become very popular.