Harnessing machine learning to predict obesity: A focus on the first 1000 days of life

The recent publication in the Scientific Reports Journal showcases a novel utilisation of machine learning (ML) to forecast obesity in adults by examining risk factors and monitoring body mass index (BMI) during the initial 1,000 days of life, spanning from two to four years old.

The rise in obesity rates in both children and adults worldwide is undeniable. Early onset obesity in children is indicative of potential adult obesity, cardiometabolic risks, and other childhood diseases.

Obesity, once entrenched, is challenging to manage and is often chronic. As a result, a preventative approach to obesity is becoming a research priority. Identifying individuals at an elevated risk of obesity in adulthood during their early years could significantly enhance these prevention efforts.

Known adjustable risk factors encompass a mother’s higher pre-pregnancy BMI, pregnancy weight gain, socioeconomic status, high neonatal weight, and local community variables such as crime rates and food availability. Despite this, the cumulative risk estimation of these factors remains underexplored.

Currently, there is a lack of initiatives aimed at estimating childhood obesity, particularly those considering prenatal and neonatal risk factors. This is despite studies highlighting that the period between two to four years of age provides a valuable window for intervention due to heightened developmental flexibility and the ability to influence health behaviours.

The study in question employs ML algorithms to pinpoint children with a higher risk of obesity, providing vital data for the creation of prevention policies and strategies. Additionally, the researchers introduced a dynamic BMI tracker for use throughout childhood to help identify obesity risks in adulthood.

The researchers utilised a machine learning technique known as least absolute shrinkage and selection operator (LASSO) regression. This allowed them to maintain features that most significantly and relevantly relate to paediatric obesity, outside of height, weight, and body mass index.

The study examined data from 149,625 visits by 19,724 individuals aged up to 48 months, with an analysis of 10,348 individuals specifically aged between 30.0 and 48.0 months. Following data correction, the supplementation of missing values, and variable normalisation, 50 variables were chosen for consideration. After application of LASSO regression and subsequent tests, a final 19 variables were scrutinised.

The proposed model comprised variables such as mean height, BMI, weight at various intervals within the first two years, time differences between visits, and percentile ranks for weight and height at two years.

The predictive ability of the model was tested with a validation dataset comprising 20% of the patients. It showed an impressive accuracy in estimating childhood BMI, with a mean error of 1.0 across all three age ranges (30.0 to 36.0 months, 36.0 to 42.0 months, and 42.0 to 48.0 months).

Most variables in the model showed a significant association with paediatric BMI across all estimated ranges. These findings suggest that this predictive model could bolster both clinical and population-wide obesity prevention efforts in the earliest days of life.

Risk factors associated with higher childhood BMI identified in the study included maternal risks during pregnancy, C-section delivery, higher infant birth weight, and sleep disturbances in infants requiring assistance to sleep.

Interestingly, living in a food desert and having Hispanic ethnicity were factors that appeared protective against high BMI.

In summary, this study highlights that machine learning can help track paediatric BMI trajectories and identify modifiable risk factors during early childhood. This supports efforts to intervene before the onset of unhealthy weight gain, aiming to alleviate the health burden of obesity.

Factors such as maternal health, a child’s sleep quality, and socioeconomic influences can shape the weight trajectories of children into later