Interactive causal discovery and inference for comorbidity onset and progression in SCI population


Loading...

Date

2023-04-19

Publication Type

Other Conference Item

ETH Bibliography

yes

Citations

Altmetric

Data

Abstract

Spinal Cord Injury (SCI) is a lifelong chronic condition affecting many functions and systems in the body. Despite extensive research on risk factors towards SCI comorbidity onset, the underlying causal relations and relevant biomarkers for early detection remain unclear, which would help medical workers make timely diagnoses, prevention, and interventions. Therefore, within the framework of continuous health monitoring in long-term chronic patients, we aspire to build a general pipeline for the intelligent assistive causal learning system based on clinical and sensory-systems data combining expert knowledge for continual causal discovery and inference in disease onset and progression. The human-machine interactive causal discovery with the consequent theoretical guarantee is this project’s focus, with a novel application on multi-modal data including both static descriptive features extracted retrospectively from patient medical records and dynamic sensor signals collected from subjects under controlled interventions. We will present an iterative approach to learning the directed maximal ancestral graph, where an expert’s belief is embedded into the constrained-based causal discovery algorithm by adjusting the corresponding rejection threshold in conditional independence tests. In this way, we aim to improve causal discovery with expert knowledge integration that simultaneously accounts for the uncertainty of the experts’ beliefs. For application-driven development and validation, we evaluate our approach in precision and robustness against a benchmark on synthetic datasets generated from the expert-knowledge simulation. Finally, we will demonstrate the learning results and findings with different encoding and levels of granularity on clinical datasets from the Swiss Paraplegic Center of secondary conditions in SCI individuals.

Publication status

published

External links

Editor

Book title

Journal / series

Volume

Pages / Article No.

Publisher

ETH Zurich

Event

European Causal Inference Meeting (EuroCIM 2023)

Edition / version

Methods

Software

Geographic location

Date collected

Date created

Subject

Graphical modeling; causal inference; medical data; Pressure injury

Organisational unit

03654 - Riener, Robert / Riener, Robert check_circle

Notes

Abstract for Poster Presentation 33 / Funded by ETH-SPS Digital Transformation in Personalized Health Care for SCI

Funding

Related publications and datasets