Weight gain prediction models
Pro Research Analysisby 
Searched over 200M research papers
Machine Learning Models for Adult Weight Gain Prediction
Recent research has focused on using machine learning to predict weight gain in adults, leveraging large datasets and various predictors. Studies using electronic health records and behavioral survey data have developed models such as elastic net and XGBoost, achieving modest predictive performance (AUCs around 0.68–0.71). Notably, adding behavioral survey data did not significantly improve model accuracy, suggesting that other data types, like genomic information, may be needed for better predictions in the future . Similarly, models targeting young adults using demographic, health, laboratory, and neighborhood variables found that age, sex, and baseline BMI were the most important predictors, but overall accuracy remained modest (AUC up to 0.68). Researchers suggest that incorporating behavioral and genetic data could enhance these models .
Environmental and Genetic Predictors of Adiposity Gain
Environmental factors have shown strong predictive power for adiposity gain in adults, with models achieving AUCs of 0.69–0.75. In contrast, polygenic risk scores (PRS) for genetic predisposition had little predictive value for mid-to-late adulthood weight gain (AUC ~0.52), though they were somewhat more useful for predicting weight gain from early adulthood to midlife (AUC 0.60–0.62). This indicates that environmental factors are more influential for weight gain in later life, while genetics may play a larger role earlier on .
Weight Gain Prediction in Special Populations
Gestational Weight Gain Models
Several studies have developed models to predict abnormal or excessive gestational weight gain. Key predictors include pre-pregnancy BMI, season of pregnancy, parity, and behavioral factors such as increased food intake and takeaway consumption. Models incorporating these factors achieved good predictive performance (AUCs 0.75–0.88), supporting their use in personalized management for pregnant women Chen2024Heery2014McDonald2020.
Breast Cancer Survivors
For breast cancer survivors, integrated models using clinical, behavioral, and biological data (including biomarkers and proteomics) have been developed. While clinico-behavioral factors like younger age, smoking, lower income, and certain treatments were associated with higher risk, adding biological markers only slightly improved prediction. The best models reached AUCs of 0.65–0.74, indicating moderate accuracy .
Anorexia Nervosa Treatment Outcomes
In patients with anorexia nervosa, ensemble machine learning models found that the amount of weight gained during treatment was the strongest predictor of BMI six months after discharge. Other factors, such as behavioral or psychological variables, were much less predictive, highlighting the importance of treatment-related weight gain for long-term outcomes .
Childhood and Infancy Weight Gain
Models predicting excessive weight gain in children have shown that growth measures related to BMI peak and pre-peak velocity in infancy are strong predictors of later weight gain. Models using these measures achieved high accuracy (AUC up to 0.86 in derivation cohorts), though performance dropped in external validation, emphasizing the need for careful model selection and validation .
Simple Mathematical Models for Individual Weight Change
A simple mathematical model based on energy balance and body composition relationships has been validated for predicting individual weight change in response to changes in intake. This model demonstrated high accuracy, with mean absolute errors below 2.5 kg in both underfeeding and overfeeding studies, outperforming other one-dimensional models .
Conclusion
Weight gain prediction models have advanced through the use of machine learning and integration of diverse data types. Environmental and clinical factors remain the most robust predictors across populations, while genetic and behavioral data may offer incremental improvements, especially in specific groups or life stages. Continued refinement and validation of these models, including the incorporation of new data types, are needed to enhance their accuracy and clinical utility Jawara2024Murtha2023Chen2024+7 MORE.
Sources and full results
Most relevant research papers on this topic