Machine learning models can predict subsequent publication of North American Spine Society Annual General Meeting abstracts

• Published in: Canadian Journal of Surgery 2022-12, Vol.65, p.S160-S160
• Main Authors: Olotu, Olumide, Abbas, Aazad, Bhatia, Akshdeep, Selimovic, Denis, Larouche, Jeremie, Yee, Albert, Lewis, Stephen, Finkelstein, Joel, Toor, Jay
• Format: Article
• Language: English
• Subjects: Accuracy; Data collection; Machine learning; Meetings
• ISSN: 0008-428X; 1488-2310

Background: Academic meetings serve as an opportunity to present and discuss novel ideas, with manuscript publication in a peer-reviewed journal being the eventual goal of presented research. Previous studies have identified factors predictive of publication without generating predictive models. However, machine learning (ML) presents a novel tool capable of generating these models. As such, the objective of this study was to use ML models to predict subsequent publication of abstracts presented at a major spine surgery meeting. Methods: All abstracts from the North American Spine Society (NASS) annual general meetings (AGM) from 2013-2015 were reviewed. The following information was extracted: number of authors, institution, location, conference category, subject category, study type, data collection methodology, human subject research and US Food and Drug Administration (FDA) approval. Abstracts were searched on the PubMed, Google Scholar, and Scopus databases for publication. ML models were trained to predict whether the abstract would be published or not. The quality of models was determined by using the area under the receiver operating characteristic curve (AUC). The top 10 most important factors were extracted from the most successful model. Results: A total of 1119 abstracts were presented, with 553 (49%) abstracts published. During training, the model with the highest AUC and accuracy metrics was the partial least squares (AUC of 0.77 [standard deviation (SD) 0.05], accuracy of 75.5% [SD 4.7%]). During testing, the model with the highest AUC and accuracy was the random forest (AUC of 0.69, accuracy of 67%). The top 10 features for the random forest model were (descending order): number of authors, year, conference category, subject category, human subjects research, continent and data collection methodology. Conclusion: This was the first study attempting to use ML to predict the publication of complete articles after abstract presentation at an academic conference. Our study used ML to identify key predictive features, combining them to create a potent predictive model. This technique can be used to improve the quality of scientific meetings while demonstrating the potential for broader applications of ML in academia and health care.