Displaying One Session

Teaching Course Fri, Sep 11, 2020
Session Type
Teaching Course
Date
Fri, Sep 11, 2020
Invited Presentations Invited Abstracts

TC14.01 - Presentation 01 (ID 629)

Speakers
Authors
Presentation Number
TC14.01
Presentation Topic
Invited Presentations

Abstract

Abstract

I will present how to analyse improvement in MS studies and the maning of univariate and multivariate analysis in MS studies.

Collapse
Invited Presentations Invited Abstracts

TC14.02 - Presentation 02 (ID 630)

Speakers
Authors
Presentation Number
TC14.02
Presentation Topic
Invited Presentations
Invited Presentations Invited Abstracts

TC14.03 - Propensity Score: Neither a Black Box nor a Magic Bullet (ID 631)

Presentation Number
TC14.03
Presentation Topic
Invited Presentations

Abstract

Abstract

There is a trend in the medical literature towards the use of real-world data to inform aspects related with efficacy and safety in people living with MS. Indeed, there are some key aspects, mostly related with safety (e.g. rare and life-threatening adverse effects), for which the use of real-world data is the most convenient approach (if not the only feasible). However, we should carefully deal with confounding as a key issue for achieving causal inference, and thereof, getting real-world (unbiased) evidence through the use of real-world data from non-randomized study designs.

(1) To describe the use of propensity-score (PS)-based methods as an attempt for controlling for confounding.

(2) To learn the advantages, limitations and assumptions of these approaches to interpret the results from studies using these methods based on our own judgement

After a brief introduction on different available methods for controlling for confounding, we will narrow the lecture on propensity-score (PS)-based methods, particularly on PS-matching, a frequent approach in studies using observational data in MS.

In this course, we will explain the key concepts of PS-methods including the methodology of use, advantages, limitations and assumptions of these approaches.

A PS is a continuous variable that defines the probability of exposition (e.g. “use of a given DMD”) conditional (“as a function”) on the confounders. In this session, we will explain how to build a valid logistic regression model to estimate a PS. Then, we will describe different uses of PS with especial emphasis on PS-matching for which the estimation will be described step-by-step. We will explain the concept of caliper (the most commonly-used criterion for matching) and the assumptions derived from the bias-variance trade off. We will explain how standardized variable differences and Love Plots can help us to judge whether the matching on PS has been successfully performed in a study.

After the completion, the attendees will be able to better interpret the results coming from studies using PS-based methods. They shall use their own judgement to assess whether the reported analyses may or may not provide be reliable, unbiased findings.

Collapse