Petra-AI: AI to support women in perimenopause

Women have long been underrepresented in medical research – with noticeable consequences for diagnostics, therapy, and data. Symptoms are often not correctly classified, evidence-based information is difficult to access.

With the research project "Petra‑AI", an interdisciplinary consortium led by Dr. Theresa Ahrens from Fraunhofer IESE aims to develop an AI-supported app that scientifically soundly, understandably, and safely supports women in perimenopause.

In the interview, Ahrens explains the role an AI chatbot can play and why structured, interoperable health data are crucial for closing care gaps.

What is Petra-AI?

The project "Petra-AI: AI-supported, educational therapy support for perimenopause" is a publicly funded research project in which we are developing an AI-supported app to support women in perimenopause. The goal is to provide evidence-based information and help women better understand and alleviate their symptoms. The project is funded by the Federal Ministry of Research, Technology and Space (BMFTR) and runs for three years.

Who is working on the project?

We are an interdisciplinary consortium. In addition to Fraunhofer IESE as the consortium leader, the digital health start-up Femna Care is involved, as well as the Junior Professorship for Health and E-Health at Ruhr University Bochum, the Social Research Unit of the Faculty of Social Sciences at the Technical University of Dortmund, and the Berlin Institute of Health at Charité.

Femna contributes expertise in women's health. The social science partners support the user-centered development and co-creation processes. The Berlin Institute of Health supports us with interoperable data standards such as FHIR and SNOMED, so that research data can be collected in a structured and reusable manner.

Is commercialization of PETRA-AI also planned?

No, it is purely a research project for now. However, we are developing the technology with the aim that it can be transferred to healthcare in the long term. We don't want to do research "for the drawer."

What role does the AI chatbot play specifically?

The chatbot is intended to convey evidence-based information in an understandable and accessible way. A central challenge of the project lies in adapting the AI-supported content to the respective knowledge level and linguistic abilities of the users as needed.

At the same time, safety is extremely important to us. Language models can hallucinate – that is, generate false information. This is particularly problematic in the healthcare context. Therefore, we are working intensively on technical safeguards and evaluating various existing language models. Training our own complete language model would not be economically sensible, but we are technically extending and securing the chosen models.

Can AI in healthcare actually be error-free?

The expectation that AI systems will work one hundred percent error-free is unrealistic. It is crucial to be transparent about what AI can do – and what it cannot.

For example, there is the so-called "automation bias": people can tend to trust AI systems too much. Therefore, we must understand AI as a support, not as a substitute for medical expertise. AI systems are socio-technical systems – the interaction between humans and machines must always be considered and researched.

How do you position yourself regarding the classification of menopause? There is also the fear that it is treated as a disease.

Menopause and the transitional phase called perimenopause are not diseases in the classic sense. Nevertheless, quality of life can be significantly impaired. Some women experience symptoms for many years – sometimes up to 15 years.

Many symptoms are not recognized as hormonally related in everyday healthcare. This can lead to long "odysseys." Therefore, education is important – both for those affected and for doctors.

Does lifestyle play a role in the perception of menopause?

Certainly, there are individual differences, but I currently see primarily a societal destigmatization. For a long time, there was little open discussion about women's health. Social media has certainly played a positive role here, as communication has become easier – even if the quality of information there varies greatly.

Will the app provide concrete therapy recommendations, for example for hormone replacement therapy?

The goal is not for the chatbot to independently suggest or even prescribe therapy. However, it can explain the pros and cons and present current study findings in an understandable way. Especially with hormone replacement therapy, there are still many fears based on older studies. Here, evidence-based information can help make informed decisions together with doctors.

How do you handle data protection and research data?

In the pilot phase, participants will be fully informed. Data will be collected pseudonymously and made accessible for research according to FAIR criteria (Ed. note: Findable, Accessible, Interoperable, and Reusable).

When will there be initial results?

The project will initially run for three years. We are starting with interviews and workshops to involve women's needs from the outset. At the end, a structured pilot phase is planned in which we will evaluate acceptance, usability, and effectiveness.

Despite the challenges, you seem optimistic.

Yes, otherwise I wouldn't have become a scientist. These are complex problems – medical, technical, and social. But we have a strong, interdisciplinary team. And let's not forget: women make up 50 percent of the world's population. Nevertheless, women's health has long been underrepresented in research and development. That is changing now – and that is a good thing.

(mack)