ACCOUNTING FOR HETEROGENEITY IN COST-EFFECTIVENESS ANALYSIS AND VALUE OF INFORMATION ANALYSIS: IVIG FOR SEVERE SEPSIS

Tuesday, October 22, 2013
Key Ballroom Foyer (Hilton Baltimore)
Poster Board # P3-7
Quantitative Methods and Theoretical Developments (MET)

Nicky J. Welton, PhD, Bristol University, Bristol, United Kingdom
Purpose: To describe a strategy to account for heterogeneity in meta-analysis of treatment effects and explore the implications for cost-effectiveness and value of  information analyses.

Methods: Heterogeneity is ubiquitous in meta-analysis and random effect models are typically used to obtain pooled effect estimates when there is variability in treatment effects across studies. However, for the results to be used as inputs to cost-effectiveness models, it is necessary to understand what is causing the heterogeneity. In the absence of observed treatment-effect modifiers, various summaries from the random effects distribution (random effects mean, predictive distribution, single study estimate, shrunken estimate, or random effects distribution) can be used depending on the relationship between the included study populations and the target population for the decision. If covariates have been measured that could potentially explain the heterogeneity, then these can be included in a meta-regression model. For example, common study-level treatment effect modifiers are indicators of risk of bias, and variable treatment definitions. We outline a model selection procedure to help choose between competing meta-regression models, including clinical input.

Results: We illustrate the approach with a recent health technology assessment of intravenous immunoglobulin (IVIG) for the management of adult patients with severe sepsis and septic shock in an intensive care setting. We show that all of the heterogeneity can be explained by including one covariate that represents risk of bias and one covariate that describes the delivery of the intervention. The results of the cost-effectiveness analysis and value of information analyses are sensitive to the approach taken to account for heterogeneity in the meta-analysis model. 

Conclusions: It is important to attempt to understand the cause of heterogeneity in meta-analysis, and how this relates to the target population for the decision when meta-analysis results are used as inputs to cost-effectiveness models. We highlight the value of clinical input to assist model selection in order to restrict attention to clinically plausible models. Sensitivity analysis to model choice is should be considered essential.