Increased risks for random errors are common in outcomes graded as high certainty of evidence.

Autor: Gartlehner G; RTI International, Research Triangle Park, NC, USA; Cochrane Austria, Danube University Krems, Krems, Austria. Electronic address: ggartlehner@rti.org., Nussbaumer-Streit B; Cochrane Austria, Danube University Krems, Krems, Austria., Wagner G; Cochrane Austria, Danube University Krems, Krems, Austria., Patel S; RTI International, Research Triangle Park, NC, USA., Swinson-Evans T; RTI International, Research Triangle Park, NC, USA., Dobrescu A; Genetics Department, Victor Babes University of Medicine and Pharmacy, Timisoara, Romania., Gluud C; Copenhagen Trial Unit; Centre for Clinical Intervention Research, Rigshospitalet, Copenhagen University Hospital, Copenhagen, Denmark.
Jazyk: angličtina
Zdroj: Journal of clinical epidemiology [J Clin Epidemiol] 2019 Feb; Vol. 106, pp. 50-59. Date of Electronic Publication: 2018 Oct 19.
DOI: 10.1016/j.jclinepi.2018.10.009
Abstrakt: Objectives: The aim of article was to assess the risk for random errors in outcomes graded as high certainty of evidence (CoE).
Study Design and Setting: We randomly selected 100 Cochrane reviews with dichotomous outcomes rated as high CoE using Grading of Recommendations Assessment, Development, and Evaluation. To detect increased risks for random errors, two investigators independently conducted trial sequential analysis using conventional thresholds for type I (α = 0.05) and type II (β = 0.10) errors. We dually regraded all outcomes with increased risks for random errors and conducted multivariate logistic regression analyses to determine predictors of increased risks for random errors.
Results: Overall, 38% (95% confidence interval: 28-47%) of high CoE outcomes had increased risks for random errors. Outcomes assessing harms were more frequently affected than outcomes assessing benefits (47% vs. 12%). Regrading of outcomes with increased random errors showed that 74% should have been downgraded based on current guidance. Regression analyses rendered small absolute risk differences (P = 0.009) and low number of events (P = 0.001) as significant predictors of increased risks for random errors.
Conclusion: Decisionmakers need to be aware that outcomes rated as high CoE often have increased risks for false-positive or false-negative findings.
(Copyright © 2018 Elsevier Inc. All rights reserved.)
Databáze: MEDLINE
Externí odkaz: