gms | German Medical Science

Background/research question: Incomplete reporting of clinical trial results is frequent for harms outcomes [1]. Outcome reporting bias in systematic reviews (SRs) may be mitigated by using clinical study reports (CSRs). We have recently conducted an SR comparing icodextrin vs. glucose in peritoneal dialysis [2]. In contrast to preceding SRs on the same topic, our dataset included CSR data made available by trial sponsors. The aim of this case study was to investigate, at outcome level, the effect of adding CSR data on data completeness and meta-analytical results of the icodextrin SR for harms outcomes. This adds to available evidence from other case studies.

Methods: Of the 19 clinical trials (1688 patients) included in the meta-analysis, data were available from publications for 18 trials (1644 patients), and from CSRs for 10 trials (1187 patients). One trial (44 patients) was unpublished but available as a CSR. The five main harms outcomes, i.e. those from the SR Summary of Findings table, were reanalysed using publication data only, and publication data with added CSR data. The number of included trials, point estimates and 95% CIs were compared at outcome level. A χ² test was used to test for significant differences in meta-analytical results.

Results: Mortality was reported in 16 of 18 trial publications (89%), conversion to haemodialysis in 11 (61%), serious adverse events (SAE) in 2 (11%), peritonitis in 7 (39%), and ultrafiltration failure in 3 publications (17%). Mortality was reported in 10 of 10 CSRs (100%), conversion to haemodialysis in 9 (90%), serious adverse events (SAE) in 10 (100%), peritonitis in 9 (90%), and ultrafiltration failure in 4 CSRs (40%). Point estimates were not significantly different for publication data, compared to publication data with added CSR data, for any outcome. Only for peritonitis, the direction of effect was changed (0.90 [0.74, 1.10] for publications vs. 1.09 [0.88, 1.34] when adding CSRs). The 95% CIs overlapped for every outcome. For SAE, the width of the 95% CI was narrower when adding CSR data (0.89 [0.51, 1.54] for publications vs. 0.92 [0.76, 1.11] when adding CSRs).

Conclusion: The fraction of publications reporting results varied substantially by outcome. SAEs were most underreported in publications. In contrast to other case studies, the conclusions of the meta-analysis remained unchanged when adding unpublished data. Future work should systematically analyse to what extent unpublished harms data influence SR outcomes.

Competing interests: The authors received funding from Baxter International for the conduct of a systematic review comparing icodextrin vs. glucose in peritoneal dialysis. MRM is an employee of Baxter Healthcare (Asia).