The British & Irish Region of the International Biometric Society is delighted to announce our first in-person meeting in nearly three years!
We invite all BIR members and friends to reunite in central London for a gala scientific meeting, which includes the 2022 Presidential Address & AGM, and features our special guest Prof. Odd Aalen. Odd is the creator of the stochastic process architecture that underpins and enables modern event-history analysis.
But there's more: we also want to hear from you! We invite oral contributions from all attendees, presenting one slide (1s) and describing one thing (1t) you have learned to do differently in the last three years. Maybe it's a new software tool. Maybe a statistical model. Maybe a way of managing email or diaries. Maybe it's a piece of scientific notation, or a productivity tip. Maybe it's a mathematical trick that keeps being useful. We're interested in methods in the widest possible sense, so anything that is about how we do our work is fair game.
Book your place on our Eventbrite page!
10:30—11:00: Annual General Meeting of the IBS-BIR
11:00—11:30: This 1s 1t! (1 slide, 1 thing) Part One
12:00—12:30: This 1s 1t! (1 slide, 1 thing) Part Two
13:30—14:30: Presidential Address (Daniel Farewell)
15:00—16:00: Time-dependent mediators in survival analysis (Odd Aalen)
2022 Presidential Address
It's About Time! (Daniel Farewell, Cardiff University)
Stochastic processes (collections of random variables indexed by time) have an abstract, technical aura that belies their intimate connection to real-world reasoning. In this presentation, I will attempt to convince the open-minded skeptic that the forbidding formalism of measure theory is a mountain well worth climbing. There are clear skies and beautiful vistas to be found above the fog that sometimes surrounds missing data in clinical trials, causal inference from observational studies, and even ordinary regression.
Time-dependent mediators in survival analysis
Odd Aalen, University of Oslo
We discuss causal mediation analyses for survival data. The emphasis is on a dynamic point of view, that is, understanding how the direct and indirect effects develop over time. Mediation analysis in a survival setting is a challenge. One issue is that a patient may survive in one counterfactual setting, and not in the other, or they may be censored in one and not in the other. An additional aspect is the need for understanding the development over time. The relevant mediators will typically be stochastic processes.
An approach to mediation analysis is based on nested counterfactuals. However, this is implies a cross-world assumption which has been disputed by several authors. A simpler and more intuitive procedure was recently developed by Vanessa Didelez, based on earlier work by Robins and Richardson. The idea is to consider different components of the treatment; one component describes how the treatment affects the outcome through the mediator, while another describes the other effects of treatment.
To illustrate and evaluate assumptions we use causal graphs for local independence. This is a concept for describing how the evolution of a stochastic process depends on other processes. In comparison with conditional independence, local independence has the advantage that there is a direction in the dependence relationship. In this talk, we shall focus on a time-discrete version of local independence. We shall show how local independencies can be read off from the graph using δ-separation (which is an extension to stochastic processes of d-separation).
We illustrate the approach with an example.