APPLICATION OF SIMPLE IMPUTATION TECHNIQUES FOR MISSING PAIRWISE CONTRASTS FROM MULTI-ARM TRIALS WHEN USING FREQUENTIST NETWORK META ANALYSIS
Author(s)
Petto H1, Brnabic A2, Kadziola Z1, Belger M3
1Eli Lilly Regional Operations GmbH, Vienna, Austria, 2Eli Lilly, Sydney, Australia, 3Eli Lilly and Company Ltd, Windlesham, UK
OBJECTIVES: When conducting frequentist (fixed effects or random effects) network meta-analysis (NMA), input data is usually required in contrast form. In practice, multiple-arm trials are quite common and results for only the contrast relative to one treatment group are presented. However, some frequentist NMA require all possible pairwise treatment effects and standard errors combinations. While the missing effect sizes can still be directly derived, additional assumptions about co-variances are needed to calculate standard errors. METHODS: Simple imputation techniques are used for substituting the standard errors of the missing comparisons and this has been applied to both simulated data as well as a real world data example. After imputation data is analyzed using standard frequentist NMA, incorporating multi arm studies by the method described in Rücker (2015). RESULTS: We derive simple imputations techniques by (1) assuming independence between contrasts, (2) estimating missing co-variances from the available contrasts in the multi arm trials and (3) from the other two arm studies in the network. Comparable results to networks including all pairwise contrasts can be obtained, especially if only few contrasts are missing in multi arm studies and if variances of the comparisons are not too different. In the first case, even (1) can give acceptable results. If variances differ, but are similar to that from two arm studies then (3) might be preferable over (2). CONCLUSIONS: Our results suggests that from a practical point of view, simple imputation techniques might be useful tools for incorporating multi arm trials with incomplete pairwise contrasts into frequentist NMA, although limitations need to be carefully considered. Rücker G: Network meta-analysis, electrical networks and graph theory. Research Synthesis Methods, 2012, 3, 312–324.
Conference/Value in Health Info
2015-11, ISPOR Europe 2015, Milan, Italy
Value in Health, Vol. 18, No. 7 (November 2015)
Code
PRM218
Topic
Methodological & Statistical Research
Topic Subcategory
Confounding, Selection Bias Correction, Causal Inference
Disease
Multiple Diseases