I have a keen interest in methodological issues in epidemiological research and how the Swedish register infrastructure can be used to improve validity. How can we trust results from observational studies, which often get substantial publicity in media? How can we improve risk predictions and our understanding of causes and effects by smarter study designs and novel analytic approaches? 

One such novel analytic approach is machine learning, i.e. methods where computer programs learn to complete a task, perform classifications or to foresee outcomes based on extensive input data. In an on-going interdisciplinary project called AIR Lund (see www.lupop.lu.se/airlund), partly financed by VINNOVA, we investigate how machine learning methods applied on the Swedish register infrastructure can contribute to improved prevention, diagnosis and prognosis of cardiometabolic diseases. We will critically assess the added value of these methods compared to more simplistic risk scoring schemes and statistical approaches. In this project we also addressing well-founded ethical and legal concerns related to the use of such complex and data-demanding algorithms in clinical practice.

Another on-going research project is on selection bias where we use register data to improve internal validity and generalizability of findings from population studies. We expect that this project, which is funded by FORTE, will have an impact on how data from the population-based cohort studies are analysed and results presented.

Besides methodological issues and clinical prediction modelling, I also have a strong research interest in health effects of outdoor environments - traffic noise, green and blue environments and air pollution. I have been involved in several studies on effects of these exposures on both well-being and chronic diseases.