February 29: Amsterdam Causality Meeting with Mirthe van Diepen and Philip Boeken
37 views
Skip to first unread message
Stéphanie van der Pas
Feb 1, 2024, 7:43:09 AM2/1/24
to n.litvak--- via stoch-ned-l, amscau...@googlegroups.com, mirthe.v...@ru.nl, p.a.b...@uva.nl
Dear all,
This meeting is financially supported by the ELLIS unit Amsterdam and the Big Statistics group at Amsterdam UMC.
Best wishes,
Sara Magliacane, Joris Mooij and Stéphanie van der Pas
The next Amsterdam Causality Meeting will take place on Thursday February 29. The aim of the seminar is to bring together researchers in causal inference from the VU, UvA, Amsterdam UMC and CWI, but it is open to everyone.
Date & time: February 29, 14.30-18.00
Location: UvA Science Park, room L3.36 in the Lab42 building
Schedule:
14.30-15.30: Mirthe van Diepen (Radboud University), Detecting New Risk Factors via Causal Discovery in Aortic Surgery
15.30-16.30: Philip Boeken (UvA), A Markov property for sample paths of stochastic processes
16.30-18.00: drinksPlease find the abstracts below this message. We would also like to announce our second spring meeting on Monday April 22, 14.30-17.30 at the VU campus.
If you're interested in this event or in the seminar series, please check our website. For announcements regarding upcoming meetings, you can also register to our Google group.
This meeting is financially supported by the ELLIS unit Amsterdam and the Big Statistics group at Amsterdam UMC.
Best wishes,
Sara Magliacane, Joris Mooij and Stéphanie van der Pas
Abstracts:
Mirthe van Diepen: Detecting New Risk Factors via Causal Discovery in Aortic Surgery
Understanding the causal relationships between demographic information and biomarkers can be extremely useful to get a better understanding of causal risk factors in healthcare. It can motivate future studies to search for an intervention that lowers the risk or the search for possible treatment alternatives that can improve quality of life expectations. Using random controlled trials (RCTs) we can try to infer specific causal relationships However, it is not always possible to directly intervene on (proxy) variables due to ethical reasons or it is just impossible in practice. Causal discovery algorithms try to address this problem, by searching for the causal structure between variables in an observational data set instead of using interventions on the variables. However, currently, in medical journals, the methods to analyze data are usually not based on causal discovery methods due to the assumptions made which are difficult to test for, and the non-intuitive definitions that are required for this field. Here we show how to handle these using a specific case study that exhibits many of these challenges. This study is motivated by a data set containing subjects who had aortic surgery at the St. Antonius Hospital in Nieuwegein. We use this data set to demonstrate what important steps are needed for the analysis. Challenges of this aortic surgery data set are (1) small sample size, (2) consisting of a complex combination of very different variables, both discrete and continuous, (3) unknown causal structure (there might be unknown confounders or cycles in the causal structure), (4) context variables and time-dependent variables (variables from the different phases in the perioperative period), and (5) missing values. We will show what to consider when choosing a causal discovery method and the impact of different choices for the hyperparameters for it. Moreover, we suggest how one can combine the outputs of a causal discovery method with bootstrapping to make it more robust for small data sets, how to deal with context variables, and how to deal with mixed data.
Philip Boeken: A Markov property for sample paths of stochastic processes
We prove a graphical Markov property for sample paths of various discrete- and continuous-time stochastic processes. When a dynamical system is modelled by a set of (ordinary, partial, or stochastic) differential equations, the existence and uniqueness of solutions can yield the existence of a (deterministic) solution function, that takes initial conditions and sample paths of exogenous noise processes as input, and maps them to the sample path of the solution of the differential equation. Similarly, sample paths of variables of a dynamic Bayesian Network can be expressed as measurable (deterministic) functions of sample paths of other variables, and an exogenous noise process. This approach allows to model the sample paths of various stochastic processes with SCMs, where the variables are entire sample paths of the process, and the structural equations are the measurable solution functions as described above. For the graph of this SCM, existing work by Forré, Mooij and Bongers yields a Markov property. This provides a ‘global’ alternative to the local independence graph as developed by Didelez, Mogensen and Hansen. Our Markov property implies a do-calculus for interventions on the level of entire sample paths. Combined with recent developments in conditional independence testing for functional data, this might be a promising approach for ‘global’ causal discovery and causal reasoning for time series data.
Understanding the causal relationships between demographic information and biomarkers can be extremely useful to get a better understanding of causal risk factors in healthcare. It can motivate future studies to search for an intervention that lowers the risk or the search for possible treatment alternatives that can improve quality of life expectations. Using random controlled trials (RCTs) we can try to infer specific causal relationships However, it is not always possible to directly intervene on (proxy) variables due to ethical reasons or it is just impossible in practice. Causal discovery algorithms try to address this problem, by searching for the causal structure between variables in an observational data set instead of using interventions on the variables. However, currently, in medical journals, the methods to analyze data are usually not based on causal discovery methods due to the assumptions made which are difficult to test for, and the non-intuitive definitions that are required for this field. Here we show how to handle these using a specific case study that exhibits many of these challenges. This study is motivated by a data set containing subjects who had aortic surgery at the St. Antonius Hospital in Nieuwegein. We use this data set to demonstrate what important steps are needed for the analysis. Challenges of this aortic surgery data set are (1) small sample size, (2) consisting of a complex combination of very different variables, both discrete and continuous, (3) unknown causal structure (there might be unknown confounders or cycles in the causal structure), (4) context variables and time-dependent variables (variables from the different phases in the perioperative period), and (5) missing values. We will show what to consider when choosing a causal discovery method and the impact of different choices for the hyperparameters for it. Moreover, we suggest how one can combine the outputs of a causal discovery method with bootstrapping to make it more robust for small data sets, how to deal with context variables, and how to deal with mixed data.
Philip Boeken: A Markov property for sample paths of stochastic processes
We prove a graphical Markov property for sample paths of various discrete- and continuous-time stochastic processes. When a dynamical system is modelled by a set of (ordinary, partial, or stochastic) differential equations, the existence and uniqueness of solutions can yield the existence of a (deterministic) solution function, that takes initial conditions and sample paths of exogenous noise processes as input, and maps them to the sample path of the solution of the differential equation. Similarly, sample paths of variables of a dynamic Bayesian Network can be expressed as measurable (deterministic) functions of sample paths of other variables, and an exogenous noise process. This approach allows to model the sample paths of various stochastic processes with SCMs, where the variables are entire sample paths of the process, and the structural equations are the measurable solution functions as described above. For the graph of this SCM, existing work by Forré, Mooij and Bongers yields a Markov property. This provides a ‘global’ alternative to the local independence graph as developed by Didelez, Mogensen and Hansen. Our Markov property implies a do-calculus for interventions on the level of entire sample paths. Combined with recent developments in conditional independence testing for functional data, this might be a promising approach for ‘global’ causal discovery and causal reasoning for time series data.
Reply all
Reply to author
Forward
0 new messages