Voices for Women (VfW) is a foundation dedicated to gender-specific care. Traditionally, medical research has focused on men, even though the female body responds to illness differently than the male body. Consequently, a staggering 80% of patients with unexplained health complaints are women. By examining just four specific diseases, timely recognition, accurate diagnosis and appropriate treatment could save €7.6 billion annually for women in the Netherlands alone [1]! Furthermore, the projected ‘unnecessary’ costs are estimated to reach €750 billion across the European Union [2]. VfW’s mission is crystal clear: give women with unexplained health complaints a voice and improve healthcare for all women. 

Since experience expert Mirjam Kaijer founded the organization, thousands of women with unexplained health complaints have contacted VfW. The overwhelming volume of responses posed a challenge: How to extract valuable insights from the vast sea of unstructured text? That’s why IG&H stepped in. Our team, consisting of healthcare and AI experts, provided support to VfW during a 3-week project in which a GenAI solution was built to extract those insights. With confidentiality in mind, we are unable to disclose the findings of our research. However, we would like to share our approach and lessons learned. 

The GenAI Solution  

As a first step, we created focus on the most critical topics for VfW, while ensuring that we could deliver high-quality results on these topics. For example, exploring whether valuable insights could be derived on these topics by solely asking questions that yield numerical or “Yes/No” responses. While understanding the most common unexplained health issues for women with unexplained health complaints is both interesting and valuable, it requires relatively complex GenAI modeling that may not yield high-quality results. In contrast, extracting the length of a woman’s patient journey can be done relatively easily. 

The next step was setting up the data processing process. GenAI performs very well on bite-sized chunks of information. Rather than overwhelming it with all the data at once, we processed each story individually and always asked a separate question for extracting a single data point. Once we set the scope and decided how we were going to process the data, we used a three-step approach to create powerful insights: 

1) Unstructured to Structured 

An LLM (GenAI) transformed the unstructured stories into a structured dataset consisting of columns and rows, instead of numerous separate text files. 

2) Data Analysis Techniques

Once we had a structured dataset, we applied simple yet effective analysis techniques. The data only consisted of simple categories and numbers, allowing very useful insights to be gathered without complex data processing techniques.  

3) Compelling Storytelling  

Insights alone are powerful, but when put into a compelling story supported by strong visuals, it extracts even more value from the data. In only a few visualizations we were able to tell the stories of over a thousand women.