Precision Medicine at Novo Nordisk Uses AI To Combat Obesity

AI can be used to figure out who is at a high risk of developing obesity and its consequences.

Importantly, we're not just looking solely at obesity, but also the risks that come with it. It's well known that if you develop obesity, you are also at risk of getting obesity-related complications. So, we are looking at both.

At the TPU, we use AI to predict who is at the highest risk by analyzing both real-world data – measurable factors from everyday life – and biological data, such as omics and genomics, which can provide deeper insights about an individual. 

At the TPU, our primary goal is to increase the number of obesity-free life years people can enjoy. We aim to prevent individuals from developing obesity and delay the onset of obesity-related conditions as much as possible.

By leveraging AI to identify those at highest risk early on, we hope to enable targeted interventions that promote healthier lives and reduce the long-term health burden associated with obesity. 

We want to predict the risk of becoming or having obesity and its complications, but we also want to then be able to provide interventions for you, and they need to be tailored to you – and that's based on the AI.

For example, if the AI model identifies that increasing daily physical activity would have the greatest impact on your health, your recommendation might be to take more steps each day. For someone else, the focus might be on optimizing nutrition, improving sleep quality, managing stress, or addressing other lifestyle factors. Because our models continuously integrate new data – such as changes in your health measurements or lifestyle habits – your personalized risk profile and recommendations can evolve. 

In our work, we use a wide variety of data types to build comprehensive risk profiles for obesity. This includes real-world data – things you can easily measure or track yourself. We also incorporate diverse biological data sets, often referred to as “omics,” which can include genomics, proteomics, metabolomics and more, depending on what’s available for the individual.

Importantly, our approach is flexible and not limited to any single type of data. We recognize that not everyone is comfortable or able to provide blood samples, saliva tests or biopsies. Therefore, we also use simpler, non-invasive inputs – like answers to targeted questionnaires about lifestyle, health history and habits – to assess risk.

By combining these multiple layers of information and applying advanced AI algorithms, we can analyze complex patterns and interactions that contribute to obesity risk.

One of the biggest challenges is ensuring that the technologies we develop are truly scalable – that they can be applied widely across diverse populations and settings without losing accuracy or effectiveness. Scalability requires robust data infrastructure and models that can handle large, varied data sets efficiently.

Another key challenge lies in data integration and consistency. There are many valuable data sets collected from different studies and cohorts, but often these contain different types of data – for example, one cohort might have detailed omics information, while another only has lifestyle or clinical data. Bridging these gaps and creating translatable models that can leverage longitudinal data across diverse cohorts is essential for building reliable predictions and interventions.

Currently, we often use simple measures like weight and height to calculate BMI, which is a useful but limited indicator. However, our goal is to move beyond static snapshots and predict an individual’s obesity risk over the course of years or even their entire lifespan. This requires coherent, high-quality data that captures dynamic changes in health and behavior over time.

We also face a global data gap. Much of the data we rely on comes from specific regions or populations, and there’s a need to include diverse data from different parts of the world to ensure our solutions are applicable to all.