Science and technology

Research to strengthen prevention.

To prevent disease, we need to understand how it develops. We combine earlier health data with new samples and computational models to look for patterns that individual tests cannot reveal.

Our approach

  1. 01

    Connect health information

    In Sweden, personal identity numbers allow researchers to link records from different registers and studies to the same person. Combined with new measurements, these records make it possible to study health over time.

  2. 02

    Investigate relationships

    Analysing different kinds of information together helps researchers study how biology, lifestyle and living conditions relate to health.

  3. 03

    Plan new measurements

    When existing data cannot answer a question, researchers need to decide what to measure, when and in whom.

All use of data takes place within approved research protocols and applicable data protection requirements.

The learning loop

Every prediction is tested in reality.

AI models study the health data we already have and predict what may happen next. Researchers then test those predictions with new samples and examinations. The results show where the models are right, and where more research is needed.

01Health history02AI analysis03Knowledge gaps04Study design05New measurements06Validation

One of the first things we want to find out is whether AI can point to the measurements that teach us the most. Researchers always decide what is measured.

How models are tested

Compare methods on equal terms.

We compare different AI models with one another and with statistical methods. They are tested on the same participants and data, with the same prediction horizon. The comparison includes open-weight models, specialised health models and models developed in-house.

  • Earlier data and later outcomes

    Models are trained on information from an earlier period and tested against what happened afterwards.

  • What does each data type add?

    We leave out one type of information at a time to see how it affects the results.

  • Synthetic data with known relationships

    We also test methods on constructed data where the underlying relationships are known.

  • Predictions and new measurements

    We compare predictions with findings from new examinations.

Different measurements

Several views of health.

  • Clinical examinations

    Blood analyses, ECG, cardiorespiratory fitness, lung function, blood pressure, grip strength, vision and hearing.

  • Molecular measurements

    Genes and gene activity, epigenetics, proteins and metabolites, the small molecules formed by the body’s metabolism.

  • Health between visits

    Wearables, home blood and saliva sampling, and repeated questionnaires provide information between examinations.

When more evidence is needed

Three ways to investigate further.

  • Repeat a measurement

    Measure the same person again to see how a value changes over time.

  • Add a different measurement

    A new type of examination can provide information that existing data cannot.

  • Include more people

    A larger group may be needed to answer the question.

Colleagues in discussion at ABC Labs

Scientific network

Research across disciplines.

Our researchers, laboratory specialists and engineers work with scientists in genetics, epidemiology, proteomics and machine learning.

  • Sara Hägg
  • Rickard Sandberg
  • Jochen M. Schwenk
  • Björn Reinius
  • Gunnel Tybring
  • Wei Ouyang
  • Paul Franks
  • Sebastian Bujwid