Article type
Year
Abstract
Background: A wide variety of different methods are currently used to present the results of primary diagnostic accuracy studies and systematic reviews of such studies.
Objective: To conduct a methodological review of the graphical methods used to present diagnostic accuracy data in primary studies and systematic reviews.
Methods: We identified primary diagnostic accuracy studies published in 2004 by hand searching 12 journals. We also used the DARE database to identify diagnostic systematic reviews published in 2003. We extracted the different graphical methods used to present diagnostic information, defined as any summary measure of diagnostic accuracy including sensitivity, specificity, likelihood ratios (positive and negative), and post-test probability of disease.
We investigated associations of the use of graphs and pictures with: measures of diagnostic accuracy reported in the publication; type of diagnostic test; and type of study (primary study or systematic review).
Results: A total of 49 primary studies and 65 systematic reviews were included in the review. None of the systematic reviews and only one of the primary studies used pictures to present diagnostic information. Forty-seven percent of primary studies and 51% of systematic reviews incorporated graphical results. The table below summarises the number of publications using each type of graph. Receiver operating characteristic (ROC) plots were the most common types of graph used in both primary studies and systematic reviews. Several systematic reviews also used forest plots to summarise results, whereas primary studies frequently used various combinations of dot plots and box (and whisker) plots. In studies that used graphical methods to present results, the number of graphs ranged from 1 to 51 (median 2, IQR 1 to 2.5 for primary studies and median 3, IQR 1.5 to 6 for systematic reviews). The study that included the most figures was a systematic review published as a report rather than in a journal.
Conclusions: Graphs and pictures to present diagnostic information are currently underused. As the new Cochrane software for diagnostic systematic reviews automatically produces graphs, the use of graphical representation of results should become standard in all Cochrane reviews.
Objective: To conduct a methodological review of the graphical methods used to present diagnostic accuracy data in primary studies and systematic reviews.
Methods: We identified primary diagnostic accuracy studies published in 2004 by hand searching 12 journals. We also used the DARE database to identify diagnostic systematic reviews published in 2003. We extracted the different graphical methods used to present diagnostic information, defined as any summary measure of diagnostic accuracy including sensitivity, specificity, likelihood ratios (positive and negative), and post-test probability of disease.
We investigated associations of the use of graphs and pictures with: measures of diagnostic accuracy reported in the publication; type of diagnostic test; and type of study (primary study or systematic review).
Results: A total of 49 primary studies and 65 systematic reviews were included in the review. None of the systematic reviews and only one of the primary studies used pictures to present diagnostic information. Forty-seven percent of primary studies and 51% of systematic reviews incorporated graphical results. The table below summarises the number of publications using each type of graph. Receiver operating characteristic (ROC) plots were the most common types of graph used in both primary studies and systematic reviews. Several systematic reviews also used forest plots to summarise results, whereas primary studies frequently used various combinations of dot plots and box (and whisker) plots. In studies that used graphical methods to present results, the number of graphs ranged from 1 to 51 (median 2, IQR 1 to 2.5 for primary studies and median 3, IQR 1.5 to 6 for systematic reviews). The study that included the most figures was a systematic review published as a report rather than in a journal.
Conclusions: Graphs and pictures to present diagnostic information are currently underused. As the new Cochrane software for diagnostic systematic reviews automatically produces graphs, the use of graphical representation of results should become standard in all Cochrane reviews.