Responsible Health Data Analysis: New Research Group at the BIH

Dr. Rohmann, what is your field of research, and what are your main research focuses?

I am an epidemiologist interested in applying and adapting methods to improve the quality, reliability, and interpretability of health research. I am particularly interested in how observational health data can be used to answer descriptive, predictive, and causal questions, with a strong focus on causal inference. My goal is to produce better evidence that can support decisions in health care and public health. Working on causal inference involves asking "what if?" questions. In simple terms, this means using research methods to understand whether and by how much a treatment (or other exposure) causes a particular health outcome.

So far, much of my research has focused on questions related to stroke and other neurovascular diseases. More recently, I have also worked on other clinical areas, including intensive care medicine and infectious diseases. In each application area, I need to collaborate very closely with discipline experts. This is because knowledge of the specific clinical area is essential for asking the right questions and for choosing suitable causal inference methods.

What projects and research topics have you worked on in the past that are particularly important to you?

I have worked extensively with registry data and other secondary health data, particularly in neurovascular research. I am still very passionate about using real-world health data to answer important questions in stroke and other neurovascular diseases. I have also worked as a methods consultant on research projects spanning many different clinical research areas. I particularly enjoy this work because it allows me to see how unique methodological challenges arise in different field-specific settings.

Another important part of my work has been research on research, or "meta-research". I study how research methods are used and reported in published scientific articles, including quantifying how common certain suboptimal practices are. I combine this work with practical suggestions and tutorials to help researchers recognize these problems and make better use of available suitable methods.

More recently, I have become interested in how generative AI is changing research and how we should adapt epidemiology teaching to this new environment. Together with colleagues from Germany and across Europe, I have worked on recommendations for the responsible use of generative AI in epidemiology teaching.

What are your plans for your research group in terms of the direction of your research?

Looking ahead, I want to develop my research group around responsible health data analysis, research quality and causal inference.

One important part of this direction is 3P-CAUSAL ("Promises, Pitfalls, and Pathways for Causal Inference with Synthetic and Anonymized Health Data"), a newly funded BMFTR Zukunft eHealth Junior Consortium project, for which I will serve as consortium lead. Our group aims to understand how privacy-preserving changes to health data, such as removing, aggregating, or changing parts of the data, can affect the “causal utility” of these data and impact the resulting causal effect estimates.

Additionally, together with a multidisciplinary group of researchers from several Berlin research institutions, my research group will work towards realizing a vision for an interdisciplinary Center for Causal Inference in Berlin. This initiative builds directly on the Einstein Circle on Causal Inference from Observational Health Data, which launched in 2025 and for which I was speaker. The idea is to bring researchers from different fields together to develop new methods, translate them into real-world applications, and raise awareness of causal inference within the wider Berlin research community.

How do you see your research aligning with the BIH’s translational mission, and how might patients benefit from your work in the future?

For me, an important part of translation is asking causal questions. We are often interested in “what if” questions: What if we introduce a new treatment? What if we change how care is delivered? What if we introduce a new technology? Getting the right answers to these questions is key to informing decisions about what to do.

The way we ask and approach research questions also affects how we interpret the results. If we only use prediction methods, we may find patterns without knowing what action would actually improve outcomes. We could mistake a correlation for a causal relationship and make the wrong decision. Causal inference helps us determine when associations are actually causal, so that we can use the evidence to inform decisions, clinical guidelines, and recommendations.

There is also another way in which I see this as part of translation. By improving how researchers are trained to analyse, review, and communicate research, we can improve the quality of science itself. This can help good research move into practice and, ultimately, benefit patients.