Evidence for differences in patterns of temporal trends in meta-analyses of diagnostic accuracy studies in the Cochrane Database of Systematic Reviews.
Murphy J., Fanshawe TR.
OBJECTIVES: Temporal trends in comparative meta-analyses of interventions are well-recognised in the medical literature. For studies of diagnostic test accuracy (DTA), evidence of temporal trends is growing and the importance of assessing and reporting them has been highlighted in recent guidelines on post-market surveillance in several jurisdictions. In this study we evaluate the prevalence and patterns of time trends using a larger and more up-to-date set of DTA systematic reviews than has previously been examined, from the Cochrane Database of Systematic Reviews. STUDY DESIGN AND SETTING: Cumulative meta-analysis was conducted on bivariate random effects meta-analysis estimates of sensitivity and specificity, after ranking studies by publication date. Trends for all studies were assessed graphically using plots of summary estimates by study rank, and using ROC plots of sensitivity vs. specificity. Linear trends were also described using weighted linear regression with autocorrelated errors of summary estimates against study rank. Various patterns of non-linear trends were characterised descriptively. RESULTS: The analysis included 46 reviews (92 meta-analyses) conducted between 2017 and 2022. The total number of studies within all reviews was 1,486, with a median [IQR] 7,134 [2,782, 16,406] participants per review. Reviews had a median [IQR] time span of 19 [15,25] publication years. Time trends in at least one DTA measure were observed in 40 (87%) reviews, and statistically significant linear trends in 32 (71%) reviews. Non-linear time trends were observed in 16 (35%) reviews. There was no evidence for a trend in either DTA measure in 27 (59%) reviews. CONCLUSION: The study contributes evidence on the variety in patterns of linear and non-linear temporal DTA trends which has not previously been described. We recommended researchers check statistical assumptions of trend analysis methods, for example using graphical methods. Further research into potential reasons for time trends could contribute to the robustness of future meta-analyses.