Schedule:
14:00-14:05 Welcome and opening remarks
 
14:05-14:45 Talk: Multivariable and multi-response Mendelian randomization: Using genetic evidence to select causes of multimorbidity
 
14:45-15:25 Talk: A scalable Bayesian double machine learning framework, with application to racial disproportionality assessment
 
15:25-15:40 Coffee Break
 
15:40-16:20 Talk: On inferential equivalence classes of causal models
 
16:20-17:00 Talk: Target Trial Emulation: From Observational Data to Causal Inference
 
17:00-17:05 Closing remarks
 
Please note that the order of the talks is provisional and may change.


Verena Zuber, Imperial College London
 
Multivariable and multi-response Mendelian randomization: Using genetic evidence to select causes of multimorbidity
 
Mendelian randomization (MR) uses genetic variation as instrumental variables to test and estimate the effect magnitude of an exposure on an outcome. Traditional MR methods are designed for one exposure and one outcome and do not reflect complex relationships among multiple exposures and outcomes as is typical for real-life applications. I am going to discuss three extensions of the standard one-exposure versus one-outcome MR model. First, multivariable MR considers multiple exposures in a joint model and estimates the direct effect of an exposure conditional on other exposures included. MR-BMA implements exposure selection in a Bayesian model averaging framework to prioritize the most likely causal exposures from a larger set of candidate exposures. Second, multi-response MR (MR2) uses the seemingly unrelated regression model to include multiple outcomes and connects their residual errors in a sparse Bayesian Gaussian copula regression framework. Third, MR-DAG is a Bayesian causal graphical model to detect dependency relations within the exposures, within the outcomes, and between exposures and outcomes to improve causal effect estimation. All methods are illustrated in real-world examples that explore how human phenotypes relate to disease.
 
 
Yu Luo, King's College London
 
A scalable Bayesian double machine learning framework, with application to racial disproportionality assessment
 
Racial disproportionality in stop and search practices elicits substantial concerns about its societal and behavioral impacts. In London, Black individuals are about four times more likely to be stopped and searched than White individuals. Using data on stop and search events in London from January 2019 to December 2023, this paper aims to investigate disproportionality in the volume of stops for expressive crimes involving Black individuals compared to other ethnicities. We employ a semi-parametric partially linear structural regression method and introduce a Bayesian empirical likelihood procedure combined with double machine learning techniques to control for high-dimensional confounding and to accommodate strong prior assumptions. In addition, we show that the proposed procedure yields a valid posterior in terms of coverage. Applying this approach to the stop and search dataset, we find that racial disproportionality aimed at the Black community may be influenced by the borough racial composition when focusing on expressive crimes.
 
 
Heather Battey, Imperial College London
 
On inferential equivalence classes of causal models
 
Causal models in the same Markov equivalence class are, in the absence of strong assumptions, statistically indistinguishable at any sample size. For modest sample sizes there is also the possibility that several such classes are compatible with the data, pointing to a confidence set of causal models as the appropriate presentation of evidence.  In a Gaussian setting, non-identifiability of models from the same Markov equivalence class corresponds to a plurality of inverse-covariance representations in the unconstrained parametrisation of Cox and Wermuth (1993). This parametrisation facilitates construction of, and a theoretical analysis for, a confidence set of causal models based on standard likelihood theory. The talk will clarify connections between the proposed inferential equivalence classes of causal models and the geometric equivalence classes of Evans (2020). I will also introduce and discuss mixtures of causal models from the same perspective.
 
 
Venexia Walker, University of Bristol
 
Target Trial Emulation: From Observational Data to Causal Inference
 
Target trial emulation provides a framework for using observational data to answer causal questions when randomised trials are not available, yield inconclusive results, or do not fully address clinically relevant subgroups or follow-up durations. This talk will introduce the target trial framework and show how explicitly specifying the protocol of a hypothetical randomised trial can improve the design, analysis, and interpretation of observational studies. Through practical examples, key concepts such as treatment strategies and time zero will be explored. The talk will also highlight common pitfalls, including immortal time bias, and illustrate how thinking explicitly about the trial we would ideally like to conduct can help us make better use of observational data.

 
 

Dean Bodenham & Marina Evangelou on the behalf of the RSS Emerging Applications Section. 

Member - £15.00

Non-member - £20.00