NEW METHODS FOR ADDRESSING CONFOUNDING IN COMPARATIVE EFFECTIVENESS AND COST-EFFECTIVENESS STUDIES

Overview: This is an intermediate-level course about quantitative methods for addressing confounding by indication in comparative effectiveness and cost-effectiveness studies.

Background: The increasing availability of large observational datasets provides major opportunities for providing evidence on the effectiveness and cost-effectiveness of new interventions to inform decisions. However, large data are themselves insufficient. For such data to provide an appropriate basis for decision-making requires expertise in the design, analysis and interpretation of observational studies. In observational studies the main methodological challenge is selection bias from confounding by indication. The aim of this course is to outline new approaches to the design, analysis and interpretation of observational studies for decision-making.

Format Requirements: The format of the course is a mixture of lectures, introducing the methods, and practical exercise, where the participants can gain hands on experience in applying the methods using the R software, through a series of real world data examples. The course requires basic familiarity with regression methods. The course includes a brief introduction to the R software. Participants who wish to have practical exposure to the methods are requested to bring their own laptop computers, with the latest version of R installed.

Description and Objectives:

This is an intermediate-level course introducing new quantitative methods for addressing confounding by indication in comparative effectiveness and cost-effectiveness studies.

We provide an overview of requisite steps in study design that can help minimise the risk of confounding, including the definition of the intervention and the potential outcomes. We provide new alternatives to standard regression or propensity score methods which are frequently used when estimating treatment effectiveness, but can be highly sensitive to the choice of model specification.  The course will introduce novel methods for addressing confounding, including data adaptive matching and regression methods that exploit machine learning. We will also outline approaches to sensitivity analyses to the ‘no unobserved confounding assumption’.

Throughout we will illustrate the conceptual ideas with examples that have been published in prestigious clinical and medical decision-making journals. The course will end with a discussion of approaches for handling new emerging problems, such as new approaches for estimating the comparative effectiveness of longitudinal, individualised treatment strategies.

By the end of the course participants will:

-              have a strategy for the design, analysis and interpretation of an observational study for estimating treatment effectiveness from observational data

-              have a clear conceptual understanding of confounding due to indication, and the implications for bias in the estimation of treatment effectiveness

-              understand the  assumptions underpinning alternative approaches for tackling  confounding

-              understand the relative advantages of new flexible regression (Super Learner) and matching approaches (Genetic Matching)  compared with conventional alternatives

-              have practical tools (including code) required to implement these methods in practice.

-              gain insights into new debates and opportunities for the use of advanced analytical methods in providing estimates of comparative effectiveness suitable for decision-making.