Graph Quality in Top Medical Journals

Well-designed graphs can portray complex data and relationships in ways that are easier to interpret and understand than text and tables. Previous investigations of reports of clinical research showed that graphs are underused and, when used, often depict summary statistics instead of the data distribution. This descriptive study aims to evaluate the quantity and quality of graphs in the current medical literature across a broad range of better journals.

We performed a cross-sectional survey of 10 randomly selected original research articles per journal from the 2012 issues of 20 highly cited journals. We identified which figures were data graphs and limited analysis to a maximum of 5 randomly selected data graphs per article. We then described the graph type, data density, completeness, visual clarity, special features, and dimensionality of each graph in the sample.

We analyzed 342 data graphs published in 20 journals. Our sample had a geometric mean data density index across all graphs of 1.18 data elements/cm². More than half (54%) of the data graphs were simple univariate displays such as line or bar graphs. When analyzed by journal, excellence in one domain (completeness, visual clarity, or special features) was not strongly predictive of excellence in the other domains.

Despite that graphs can efficiently and effectively convey complex study findings, we found their infrequent use and low data density to be the norm. The majority of graphs were univariate ones that failed to display the overall distribution of data.