Comparative effectiveness research with an emphasis on causal inference methods

Our work will include comparative effectiveness research, with an emphasis on causal inference methods for complex longitudinal data from both observational studies and randomized trials, e.g. to assess effects of interventions on cardiovascular disease, cancer, and others. The research will mainly be based on data from Swedish population and health data registers.

Overall program

Causal inference methods have succeeded in extracting meaningful insights from otherwise unmanageably large biomedical databases and have added to the understanding of how these data can help clinicians make better informed decisions in health care. However, their use within Swedish register data is still in its infancy. We aim to develop and tailor causal inference methods for use in Swedish register data. Our initial focus is to use the target trial framework to design observational studies that ask casual questions. Ideally, questions about comparative effectiveness or safety would be answered using an appropriately designed and conducted randomized experiment. When we cannot conduct a randomized experiment, we analyze observational data. The target trial framework recommends that observational studies should be designed using the same approach as when designing randomized trials. If the aim is to ask a causal question, one can always think of protocol for the ideal randomized experiment: the target trial. By explicitly defining the protocol of the target trial and emulating it in observational data, we are able to prevent paradoxes and biases that are common in epidemiological studies of observational data. Additionally, we aim to expand the use of the parametric g-formula to tackle time-varying confounding and handle complex causal questions. We will apply this framework and methods to two areas, as follows.

Estimating the effect of dietary interventions on cardiovascular risk

Nutritional epidemiology uses observational data extensively, as randomized controlled experiments are expensive and often difficult to conduct. However, traditional methodological approaches in the field are often limited to the comparison of cardiovascular risk at different level of a nutrients or at different dietary patterns, but cannot address well-defined, explicitly formulated causal questions that would have important public health relevance. Many of the “real-life” nutritional interventions involve substituting one nutritional factor with another while keeping overall energy intake unchanged. Furthermore, dietary interventions must sustain for a longer period to be effective, therefore factors that may influence adherence to these interventions should be considered during the analysis. The target trail framework provides a conceptual framework and causal analytical methods that allows us to address such research questions and emulate a hypothetical experiment by using observational data.

Replicating and extending the results from randomized trials for cardiovascular disease

A fundamental problem of causal inference is the impossibility to determine when the observational data are sufficient to approximately emulate the target trial. Though several attempts have been made at comparing the results of randomized trials and observational studies, the results are hard to interpret because 1) the observational studies did not explicitly emulate the trials that were actually conducted, and 2) randomized and observational studies were conducted in different populations. We will use observational data from the SWEDEHEART register to emulate several register-based

randomized trials that were embedded within SWEDEHEART. This is a unique opportunity to conduct a systematic comparison of randomized and observational studies in the same population with an explicit emulation of the trials that were actually conducted. If the observational emulation is successful, we will then attempt to extend the results of the randomized trials using the observational data. This would enable prompt estimation of the long-term effects of interventions and allow adaptive clinical or regulatory changes to be put quickly into practice.


Swedish Research Council (Vetenskapsrådet)

Swedish Research Council for Health, Working Life and Welfare (FORTE)


Anthony Matthews

Postdoctoral researcher

Katalin Gémes

Assistant professor

Miguel Hernan

Visiting professor