Online Program

Multivariate techniques for combining information

*Christopher Schmid, Brown University 

Keywords: network meta-analysis, longitudinal data, meta-regression, multivariate analysis, missing data, ecological bias

Patient-centered comparative effectiveness studies are inherently multidimensional, dealing with different treatments, different outcomes, different data collection schedules and different patient and treatment environments. The ideal of complete information at the patient and provider level is almost never available and frequently only data summaries at a study level are provided. This talk discusses issues involved with appropriately modeling incomplete multivariate study-level data where inferences are desired about multiple treatments with multiple outcomes, multiple measurement times and multiple covariates. The talk with emphasize practical applications and issues in collecting, extracting, analyzing and summarizing multivariate data. These include heterogeneity, missing and unbalanced data, direct and indirect comparisons, correlation, heterogeneity and network inconsistency. The talk will summarize the promise and the pitfalls of analyzing multivariate data in the context of existing data availability.