STATISTICAL METHODS FOR PATIENT-CENTERED COMPARATIVE EFFECTIVENESS RESEARCH

Format Requirements: This will be a half-day intermediate level course which requires only a basic understanding of regression analysis. There are no limitations in terms of number of participants. The course will be primarily lecture-based. Example analyses will also be conducted using standard statistical software (e.g. Stata).

Outline and Objectives:

The primary goals of this workshop are to 1) describe what CER and PCOR are, and why they are important, 2) introduce the fundamental methods that adjust for confounding, and 3) describe and contrast other advanced approaches.

 I. From CER to PCOR, and a case for why methods matter

a) What is CER and why is it important?

b) Priorities and past funding for CER

c) Evolution from CER to PCOR

e) Why methods matter to PCOR

II. The basics of assessing effectiveness for observational data

a) Appeal of randomized clinical trials (RCTs) and need for observational CER

b) Bias from treatment self-section

c) A standard approach: covariate adjustment with regression

d) The concept of propensity scores to capture measured confounders

e) The concept of instrumental variables to capture unmeasured confounding

III. Application of propensity scores and instrumental variables

a) Matching,  inverse probability weighting and doubly robust propensity score methods

b) Selecting an instrumental variable: principles and examples

c) Other advanced methods

d) Assumptions and diagnostics

IV. Conclusions and recommendations