Dynamite plots in surgical research over 10 years: a meta-study using machine-learning analysis

Abstract

Abstract Purpose Bar charts of numerical data, often known as dynamite plots, are unnecessary and misleading. Their tendency to alter the perception of mean’s position through the within-the-bar bias and their lack of information on the distribution of the data are two of numerous reasons. The machine learning tool, Barzooka, can be used to rapidly screen for different graph types in journal articles. We aim to determine the proportion of original research articles using dynamite plots to visualize data, and whether there has been a change in their use over time. Methods Original research articles in nine surgical fields of research were sampled based on MeSH terms and then harvested using the Python-based biblio-glutton-harvester tool. After harvesting, they were analysed using Barzooka. Over 40 000 original research articles were included in the final analysis. The results were adjusted based on previous validation data with 95% confidence bounds. Kendall τ coefficient with the Mann–Kendall test for significance was used to determine the trend of dynamite plot use over time. Results Eight surgical fields of research showed a statistically significant decrease in use of dynamite plots over 10 years. Oral and maxillofacial surgery showed no significant trend in either direction. In 2022, use of dynamite plots, dependent on field and 95% confidence bounds, ranges from ~30% to 70%. Conclusion Our results show that the use of dynamite plots in surgical research has decreased over time; however, use remains high. More must be done to understand this phenomenon and educate surgical researchers on data visualization practices.