In this webinar, Dr. Dormuth outlines the process whereby a research request from a stakeholder such as Health Canada results in a finished multicentre observational study from CNODES. A recent study of statins and diabetes is presented as an example. Emphasis is on the dissemination phase of the process, which is sometimes plagued to a degree by popular misunderstandings of confounding.
Analyst Session: Menglan and Valerie walk us through coding applications of Marginal Structural Model designs.
This workshop was originally presented by and for CNODES analysts to provide guidance in practical applications of marginal structural models. Menglan and Valerie cover: model specifications for marginal structural models; how to implement a marginal structural model design in a real study; diagnostic tools for the distribution of weights; methods to deal with extreme weights; and sensitivity/exploration analyses for different modeling approaches. Stay tuned – the corresponding methods lecture by Dr. Robert Platt is coming soon.
Dr. Suissa draws from his extensive teaching experience to lead us through an introduction to nested case-control design and analysis.
Observational studies of drug effects using healthcare databases usually involve large cohorts of subjects with long follow-up. At times, the cohorts are huge and involve time-varying exposure and covariates, presenting formidable technical challenges in data analysis. To analyze such cohorts, the nested case-control approach has been the standard design based on sampling within the cohort. In this talk, Dr. Suissa describes various sampling schemes and the corresponding measures of association, and illustrates this approach using a study of pneumonia risk associated with inhaled corticosteroid use in a large cohort of patients with chronic obstructive pulmonary disease (COPD).
This lecture by Dr. Platt covers an exposure-based method of minimizing confounding known as inverse probability of treatment weighting.
For observational studies of drugs, researchers seeking to control for confounding often have a better understanding of the predictors of drug exposure than they do of the complex etiology of the disease under study. Since conventional methods aim to eliminate confounding by controlling for predictors of disease, they may sometimes be less useful than methods of control that are based on predicting exposure. This methods lecture will cover an exposure-based method known as IPTW, including: counterfactuals and average causal effects; inverse probability weighting to create a pseudo-population; stabilization; and implementation/an example.
The conclusion of Jeremy Rassen’s engaging presentation on the implications of analyses using high-dimensional propensity scores.
In Part II of this presentation, Dr. Rassen covers optimal approaches to 2-group PS matching, an approach to 3-group PS matching, and a group discussion of “how would you design a study?”
Join one of the co-developers of the high dimensional propensity score for this engaging presentation on the implications of these analyses.
Dr. Rassen begins with an intuitive introduction to propensity scores, and then goes on to discuss variable selection in propensity score models and use of confounder time patterns to improve bias adjustment with hdPS. Continue on to Part 2 for optimal approaches to 2-group PS matching, an approach to 3-group PS matching, and a group discussion of “how would you design a study?”
Dr. Maclure walks us through his creative and very visual explanation of propensity scores.
Is understanding propensity scores a headache for you? Join the crowd! You’ve heard of crowd-sourcing, this is crowd-sorting. Imagine you are in a crowd of 216 people who have packed a lecture theatre for a day-long course on managing migraines. Dr. Maclure guides us through a pictorial explanation of how the people in the theatre can be sorted into groups based on confounding variables such as drinking coffee and smoking. He then shows how to balance the different groups by matching. This is a very concrete and visual example of how propensity scores can be used in observational studies.
In this talk, Dr. Dormuth offers a great introduction to why and how to use propensity scores.
“As epidemiologists and data analysts doing observational studies, only you can control confounding.” When randomization is not an option, we need to develop other methods to get at the right result when we have potential confounding variables that cannot be removed with a coin toss. The propensity score is a numerical technique for addressing confounding in the association between the exposure and the outcome. This talk is a great introduction to why and how to use propensity scores.