Analysis of the orthopaedic trauma literature utilizing machine learning and latent dirichlet allocation

Abstract

Objectives: To demonstrate a new method to review literature utilizing machine learning and latent Dirichlet allocation and summarize the past 20 years of orthopaedic trauma research. Methods: All original research articles published in the Journal of Bone and Joint Surgery American volume, Journal of Orthopaedic Trauma, Journal of Bone and Joint Surgery British volume, Trauma, Injury, Clinical Orthopaedics and Related Research, and the Journal of the American Academy of Orthopaedic Surgeons from 2000-2020 were analyzed using latent Dirichlet allocation (LDA), which is a form of machine learning. 100 topics were created by the algorithm and only topics that were relevant to trauma were included, leaving 30 topics ranked by popularity and associated with a best-fitting article and trend over the past 20 years. Results: Research article abstracts totaling 21,968 from 2000-2020 in the orthopaedic trauma literature were analyzed to create 30 topics. The topics were ranked by popularity, trended over the specified time period, and associated with a best fitting article. The 3 “hottest” and “coldest” topics were visualized in graphical form. Conclusions: This is the first study of its kind to utilize machine learning as a method of reviewing the orthopaedic trauma literature. Machine learning possesses the ability to rapidly synthesize a large body of literature to assess the current state of research and trends of research topics. Machine learning can aid clinicians and researchers in time-intensive tasks to quickly provide clues that will promote avenues further research.