By applying machine learning, we identified key environmental stressors affecting children, analyzed their interactions, and quantified the risk of children developing health conditions due to these environmental exposures and clinical factors.

Environmental risks to child health

Using data from over 1,600 European mothers and children from six countries, our study looked at how living in cities, being exposed to chemicals, metabolic profiles, and other prenatal and childhood experiences work together to influence children’s mental, cardiovascular, and respiratory health.

In this study, we found that maternal stress, exposure to noise from neighbors and other kids, and various lifestyle factors, such as a child's diet or level of physical activity, play the biggest role in children's mental health (figure 1). We also identified biological factors, including child BMI and the presence of specific proteins, that can predict cardiometabolic and respiratory diseases.

Being the first to develop such early-life environmental and clinical risk scores encompassing a wide range of factors, this study has significant implications for preventative care and treatment.

Using machine learning to study environmental impacts

The approach used to identify these risk factors and assess their relationships with health was a key part of this research. The study used supervised machine learning, a field of artificial intelligence in which an algorithm is trained using large, labeled datasets to make predictions on new datasets. Given that environmental factors rarely act in isolation, our study focuses on the complete set of environmental exposures encountered throughout life, commonly referred to as the human exposome.

Studying the human exposome means being open to complex and unexpected interactions. While traditional statistical methods often make assumptions about how things are related based on pre-defined patterns or formulas, our machine learning methods do not. They are more flexible and therefore helped us uncover and more accurately understand the complex relationships between our environment and health.

Addressing the challenges of machine learning

New approaches also bring new challenges. One of the difficulties of using machine learning is that it can be hard to interpret the data, due to its complexity and the number of parameters involved. This is something that our approach already considers.

The second challenge is the amount of data needed to train the algorithms that are used in machine learning properly. Gathering data sets on thousands of people as well as hundreds of different environmental and clinical factors is costly and time-consuming. Although steps are being taken to address this through federated learning, a technique that allows multiple servers to collaboratively train a model without sharing their data, additional funding and effort are needed to advance this area of research.

To ensure that our model produces the most trustworthy and generalizable environmental risk scores possible, we are now looking to incorporate well-known environmental health effects from existing literature into the machine learning process.

Despite these challenges, our research marked a great stride forward in better predicting how the environment a child grows up in will impact their health in the future with the help of machine learning models. By further understanding these environmental risk factors, healthcare leaders and practitioners can put in place preventative care measures to better protect children across the EU.

Read the full study

Learn more about the ATHLETE research project