Data-driven implementation science synthesis: a topic modeling approach to understand implementation science in the biomedical literature

Implementation science (IS) is a far-reaching field covering innumerable concepts to define overarching topics, methods, and perspectives. As the field has grown, there are increasing concerns about effectively synthesizing large swaths of information to interpret key factors important to dissemination and implementations science studies. To understand the scope and trends within Implementation Science (IS) we reviewed the IS literature using Latent Dirichlet Allocation (LDA). We applied an explanatory mixed methods approach (quantitative then qualitative) to identify and characterize groups of themes in the peer reviewed literature. We included PubMed articles from 1975-2023 using a broad set of IS terms in the title and/or abstract. We employed Latent Dirichlet Allocation (LDA) to identify the number of latent clusters in the literature. We confirmed the number of clusters using majority vote from perplexity, CV coherence scores, and UMass coherence scores. Finally, we iteratively reviewed the titles from each cluster giving each cluster a topical label through manual consensus. We analyzed 9,098 LDIS articles from PubMed using natural language processing techniques, namely topic modeling, and identified 15 distinct themes. These themes represent two categories of IS: Metascience (e.g., Theories, Models, and Frameworks) and IS Topics in IS (e.g., Healthcare and Chronic Disease). The number of IS-related articles has increased exponentially with similar articles in both Metascience and IS Topics in IS. Our findings demonstrate an expanding scope of IS over time, providing valuable insights into areas within the field that may require further exploration and discussion. With the incredible growth of IS, these results suggest that informatics approaches may be well-suited to address the magnitude and heterogeneity of the field. This article provides a novel approach in informatics applied to the IS literature.