MATCHING-ADJUSTED INDIRECT COMPARISON (MAIC)- SENSITIVITY ANALYSES AND GRAPHICAL DIAGNOSTICS
Author(s)
Signorovitch J, Gao W, Rybkin V, Yao Z, Hellstern M
Analysis Group, Inc., Boston, MA, USA
Presentation Documents
OBJECTIVES: MAICs have been used to compare treatment outcomes across separate clinical trials and to inform health technology assessments, especially when anchor-based or network meta-analyses are not feasible or may be biased by cross-trial differences. As in any adjusted analysis of non-randomized treatment groups, an important decision in the design of an MAIC is the set of baseline characteristics used for adjustment. We present analyses and graphical summaries that can be used to help evaluate this choice. METHODS: Given a set of baseline characteristics, univariate MAICs matching one baseline characteristic at a time, and multivariable MAICs matching all selected baseline characteristics but removing one at a time, are performed. Impacts on the estimated treatment effect, and on its estimated level of uncertainty (the standard error), are tabulated from each analysis and summarized graphically in conjunction with the baseline differences in each characteristic. RESULTS: Example applications based on simulated and real data illustrate how these analyses can be used to identify sensitivity (or lack of sensitivity) to the choice of adjustment variables in an MAIC. The analyses also help assess the face validity of the analysis (e.g., does the estimated treatment effect move in the expected direction after adjustment) and help identify baseline characteristics for which adjustment increases uncertainty without substantially impacting estimated treatment effects. CONCLUSIONS: Analytical and graphical approaches to evaluating the contribution of each baseline characteristics to an MAIC can be helpful for assessing the sensitivity and face validity of estimated treatment effects.
Conference/Value in Health Info
2017-11, ISPOR Europe 2017, Glasgow, Scotland
Value in Health, Vol. 20, No. 9 (October 2017)
Code
PRM224
Topic
Methodological & Statistical Research
Topic Subcategory
Confounding, Selection Bias Correction, Causal Inference
Disease
Multiple Diseases