BACKGROUND: Systematic reviewers often encounter incomplete or missing data, and the information desired may be difficult to obtain from a study author. Thus, systematic reviewers may have to resort to estimating data from figures with little or no raw data in a study's corresponding text or tables. METHODS: We discuss a case study in which participants used a publically available Web-based program, called webplotdigitizer, to estimate data from 2 figures. We evaluated and used the intraclass coefficient and the accuracy of the estimates to the true data to inform considerations when using estimated data from figures in systematic reviews. RESULTS: The estimates for both figures were consistent, although the distribution of estimates in the figure of a continuous outcome was slightly higher. For the continuous outcome, the percent difference ranged from 0.23% to 30.35% while the percent difference of the event rate ranged from 0.22% to 8.92%. For both figures, the intraclass coefficient was excellent (>0.95). CONCLUSIONS: Systematic reviewers should consider and be transparent when estimating data from figures when the information cannot be obtained from study authors and perform sensitivity analyses of pooled results to reduce bias.
BACKGROUND: Systematic reviewers often encounter incomplete or missing data, and the information desired may be difficult to obtain from a study author. Thus, systematic reviewers may have to resort to estimating data from figures with little or no raw data in a study's corresponding text or tables. METHODS: We discuss a case study in which participants used a publically available Web-based program, called webplotdigitizer, to estimate data from 2 figures. We evaluated and used the intraclass coefficient and the accuracy of the estimates to the true data to inform considerations when using estimated data from figures in systematic reviews. RESULTS: The estimates for both figures were consistent, although the distribution of estimates in the figure of a continuous outcome was slightly higher. For the continuous outcome, the percent difference ranged from 0.23% to 30.35% while the percent difference of the event rate ranged from 0.22% to 8.92%. For both figures, the intraclass coefficient was excellent (>0.95). CONCLUSIONS: Systematic reviewers should consider and be transparent when estimating data from figures when the information cannot be obtained from study authors and perform sensitivity analyses of pooled results to reduce bias.
Authors: Takashi Hirase; Jeremiah F Ling; Varan Haghshenas; Jeyvikram Thirumavalavan; David Dong; Darrell S Hanson; Rex A W Marco Journal: Spine Deform Date: 2021-11-01
Authors: Allison Gates; Michelle Gates; Ben Vandermeer; Cydney Johnson; Lisa Hartling; David W Johnson; Terry P Klassen Journal: Cochrane Database Syst Rev Date: 2018-08-22