A META-EPIDEMIOLOGICAL SURVEY OF THE REPORTING OF EFFECT MODIFICATION IN NETWORK META-ANALYSES.
Author(s)
Kovic B1, Zoratti M2, Michalopoulos S2, Silvestre C3, Thorlund K1, Thabane L1
1McMaster University, Hamilton, ON, Canada, 2Redwood Outcomes, Vancouver, BC, Canada, 3University of Waterloo, Milton, ON, Canada
OBJECTIVES: To evaluate the current state of reporting and handling of effect modification in network meta-analyses (NMAs), as well as perform exploratory analyses to identify factors that are potentially associated with incomplete reporting of effect modifiers in NMAs. METHODS: We conducted a meta-epidemiological survey utilizing a systematic review of NMAs published in 2013 and identified through MEDLINE and Embase databases. We extracted information and reported descriptive statistics on effect modifiers analyzed, adjustment methods used, study outcomes, and potential independent factors for incomplete reporting. We evaluated three study outcomes: the reported statistical analysis plan, the reported analyses results, and complete reporting (i.e. reporting of both plan and analyses results). We performed univariate logistic regression analyses exploring potential associations between factors and outcomes. RESULTS: The review identified 77 NMAs. The most common type of effect modifiers identified and explored were patient characteristics (50.7% or 39/77), and most common adjustment method used was sensitivity analysis (51.7% or 30/58). Over 45% (35/77) of studies did not describe a plan, nearly 40% (30/77) did not report the results of analyses, and approximately 47% (36/77) of studies had incomplete reporting. Exploratory univariate regression analyses yielded a statistically significant association for the factors of journal impact factor, ratio of randomized controlled trials to number of comparisons, and total number of randomized controlled trials. CONCLUSIONS: Current reporting practices are largely deficient, given that almost half of identified published NMAs do not explore or report effect modification. Journal impact factor and amount of available information in a network were associated with completeness of reporting. This study highlights the need for readers to be aware of whether statistical analysis plans include steps that report and address potential effect modifiers due to the impact they may have on analyses.
Conference/Value in Health Info
2017-05, ISPOR 2017, Boston, MA, USA
Value in Health, Vol. 20, No. 5 (May 2017)
Code
PRM106
Topic
Methodological & Statistical Research, Study Approaches
Topic Subcategory
Modeling and simulation
Disease
Multiple Diseases