A graphical display of reporting factors associated with systematic reviews: an example from diabetic retinopathy

Tags: Oral
Hui X1, Clearfield E1, Lindsley K1, Virgili G2, Scherer R1
1Cochrane Eyes and Vision Group, Johns Hopkins University, USA, 2Cochrane Eyes and Vision Group, University of Florence, Italy

Background: Systematic reviews (SRs) are considered to be the highest level of evidence, but may lead to biased conclusions if not properly conducted. We aim to evaluate SRs of interventions for diabetic retinopathy (DR), the leading cause of blindness in the working-age population in developed countries.

Objectives: To assess the descriptive, methodological, quantitative, qualitative, and other characteristics of SRs of interventions for DR.

Methods: We searched PubMed/MEDLINE, EMBASE and the Cochrane Central Register for Controlled Trials (CENTRAL) up to May 2014 for all SRs related to eyes and vision with a focused question and explicit scientific approach. We identified all SRs that examined an intervention for DR. Two authors independently evaluated 15 characteristics adapted from the Critical Appraisal Skills Programme (CASP) and A Measurement Tool to Assess Systemic Reviews (AMSTAR; Table 1). We noted whether each item was reported or not reported/can’t tell and calculated the proportion of SRs reporting that characteristic. We plotted the calculated proportions using a radar plot in MS Excel (Figure 1).

Results: We identified 25 SRs published from 2002 to 2014; four were Cochrane SRs. The SRs included a median of six studies (range: 0-11). Common interventions included anti-vascular endothelial growth factor (VEGF) treatments (14 trials) and corticosteroids (10 trials). The majority of SRs presented eligibility criteria (96%), however only 36% reported that the criteria were pre-specified. Only 56% of searches were comprehensive. A risk of bias assessment was performed in 52% of SRs, and 61% discussed limitations. Seventy-nine per cent used correct methods when results were combined quantitatively. Non-industry or no funding was reported by 56%.

Conclusions: The radar graph visually displays substantial gaps in the reporting of descriptive factors and appropriate SR methodology for SRs in DR. Reporting of quantitative factors was slightly better compared with qualitative methods, such as discussing limitations.