Synergizing human insight and machine learning: A dual-lens approach to uncovering healthcare research and innovation outcomes

Abstract

Many healthcare organisations have extensive documentation detailing the processes behind their various research and innovation projects. Analysing this data can provide valuable insights into why some projects succeed without major issues, others encounter and overcome problems, and some ultimately fail. This study introduces an approach that combines narrative interviews and Natural Language Processing (NLP) to identify patterns associated with innovation project outcomes. We analysed 618 documents from 67 projects provided by ZonMw, a major Dutch healthcare research funder, and conducted 32 narrative interviews across seven cases of healthcare innovation projects. By using narrative interviews to inform and pre-train a text embedding model, we demonstrate the potential to create a proxy for human judgement, allowing for a more natural identification of contextual patterns in project documentation. The findings indicate that successful projects are more likely to adopt a proactive approach to role changes and uncertainty (due to ambiguous laws and regulations) and to allow flexibility, which enhances stakeholder engagement, compared to failed projects. However, while we were able to conduct descriptive analysis to gain these insights, significant interpretation is still required to fully understand the findings. Our study makes two primary contributions: first, it offers a new approach for future research on the factors that determine project success or failure, closely aligning with Structuration Theory. Additionally, it suggests potential efficiency improvements in theory development by enabling multiple pattern configurations within Grounded Theory. Second, it offers practical strategies for organisations to more effectively capture and use contextual information in their project documentation for future success.