An increasing number of people are suffering from overweight or obesity. According to estimates, more than 50% of the global population will be affected by 2035, which could lead to serious health problems such as cardiovascular diseases, diabetes, and certain types of cancer. At the same time, there are numerous approaches to weight loss (WL), but the weight changes achieved through these methods vary greatly on an individual basis. Therefore, it is crucial to understand which factors influence the individual success of a WL intervention in order to predict weight changes as accurately as possible. Machine learning (ML) can help uncover patterns in complex datasets, providing valuable insights. Although ML has already been successfully applied in various medical disciplines, its use in predicting weight changes is still limited.
Under the scientific supervision of Prof. Dr. Köhler, head of the Assistant Professorship of Exercise, Nutrition and Health, doctoral candidate Christina Glasbrenner has published an article addressing the use of ML to predict the success of a calorie-reduction intervention. The research team developed ML algorithms to generate predictive models for individual WL following a calorie-reduced diet. The results of the study have now been published under the title “Prediction of individual weight loss using supervised learning: findings from the CALERIETM 2 study” in the American Journal of Clinical Nutrition, which has an impact factor of 6.5.
"Our publication is a great example of how we can combine machine learning with traditional research methods to improve the success of nutritional interventions. I am convinced that such approaches will not only become established in research in the coming years but also in commercial weight loss programs. We are very pleased that, with Christina Glasbrenner, we have not only the scientific expertise but also the necessary programming skills within our team," says Prof. Dr. Köhler regarding the publication and Christina Glasbrenner's expertise.
The research goal was to develop ML models that predict individual weight change following a calorie-reduced diet and to identify predictors that influence WL success. Christina Glasbrenner explains the research gap her study addresses: "Weight loss is highly individual, and existing models are usually average models, which can deviate significantly for a single individual. That’s why we tried to use machine learning to analyze large amounts of data to predict—before starting a calorie reduction—whether it will be successful or not, and how the weight will change over the course of a year of intervention."
For the algorithms, the research team used data from a well-known calorie reduction study from the USA (CALERIETM 2). Glasbrenner and her team examined the results of 130 participants after one year of calorie reduction. In the analysis, 197 variables, all of which were known before the intervention began, were considered as potential factors. The research findings show that machine learning models work well for predicting WL after twelve months of calorie-reduced dieting. According to the study, the algorithms performed best when using 20 to 40 variables.
“We identified 21 variables that have a major influence and are thus very helpful for predictions. These variables are based on psychological and demographic questionnaire data, nutritional information, biomarkers, and more complex laboratory measurements. Among them are two new predictors that may seem unusual at first: orgasm satisfaction and sexual behavior. What’s also interesting is that 16 of the variables we identified are already discussed in the literature, and three variables are even used in existing prediction models,” explains Christina Glasbrenner. The strongest influence on weight change was found in orgasm satisfaction and the respiratory quotient, with higher values associated with a lower predicted weight loss. Also relevant were higher baseline levels of the signaling molecules adiponectin and parathyroid hormone (PTH), which also hindered weight loss, while higher values for age, body fat percentage, C-reactive protein, and the glycemic index of the diet prior to the intervention were associated with a stronger predicted weight loss.
Based on these results, specialists can determine whether a patient will be successful with a calorie-reduction approach, or whether alternative WL methods should be pursued. According to Christina Glasbrenner, this has concrete implications for weight loss practice: “One can now already get an estimate of how promising a reduced caloric intake is by using basic factors that can be measured by oneself or through a questionnaire. If more complex measurements are done by a doctor, this estimate can be further refined. This would support a data-based, faster, and more objective decision-making process.”
The research article was developed in close collaboration with the Pennington Biomedical Research Center (USA), one of the world’s leading institutions in the fields of biomedicine and nutrition, as well as with the Duke University School of Medicine and the University of Glasgow's School of Medicine, Dentistry & Nursing.
Link to the Assistant Professorship of Exercise, Nutrition and Health website
Link to the study
Contact:
Prof. Dr. Karsten Köhler
Assistant Professorship of Exercise, Nutrition and Health
TUM Campus im Olympiapark
Am Olympiacampus 11
80809 München
Tel.: 089 289 24488
E-Mail: karsten.koehler(at)tum.de
Christina Glasbrenner
Assistant Professorship of Exercise, Nutrition and Health
TUM Campus im Olympiapark
Am Olympiacampus 11
80809 München
E-Mail: christina.glasbrenner(at)tum.de
Text: Jasmin Schol/Bastian Daneyko
Photos: Pixabay/Privat