# A novel method for Causal Structure Discovery from EHR data, a demonstration on type-2 diabetes mellitus

**Authors:** Xinpeng Shen, Sisi Ma, Prashanthi Vemuri, M. Regina Castro, Pedro J. Caraballo, Gyorgy J. Simon  
**arXiv:** [2011.05489](https://arxiv.org/abs/2011.05489)  
**Date:** 2020-11-11  
**Summary page:** https://www.summarizepaper.com/en/paper/2011.05489v1/

## Key points

- Understanding causal mechanisms underlying diseases is crucial for improved diagnosis, prognosis, and treatment selection.
- Traditional clinical trials have limitations in determining causality, prompting the use of Electronic Health Records (EHR) as a valuable source of real-world data.
- The study introduces a new data transformation method and a novel Causal Structure Discovery (CSD) algorithm that outperforms existing methods in correctness, stability, and completeness.
- Focused specifically on type-2 diabetes mellitus, the study uses large EHR datasets from Mayo Clinic and Fairview Health Services for internal evaluation and external validation.
- The proposed method successfully incorporates study design considerations and remains robust even with unreliable EHR timestamps.
- It accurately infers causal effect directions and improves the clinical correctness of discovered graphs.
- This research significantly advances our understanding of disease mechanisms through innovative methods utilizing real-world healthcare data effectively.

## Summary

In their study titled "A novel method for Causal Structure Discovery from EHR data, a demonstration on type-2 diabetes mellitus," Xinpeng Shen, Sisi Ma, Prashanthi Vemuri, M. Regina Castro, Pedro J. Caraballo, and Gyorgy J. Simon address the importance of understanding causal mechanisms underlying diseases for improved diagnosis, prognosis, and treatment selection. The authors highlight the limitations of traditional clinical trials in determining causality and propose the use of Electronic Health Records (EHR) as a valuable source of real-world data for uncovering disease mechanisms. To effectively leverage this data, they introduce a new data transformation method and a novel CSD algorithm that outperforms existing methods in terms of correctness, stability, and completeness. The study focuses specifically on type-2 diabetes mellitus and uses large EHR datasets from Mayo Clinic and Fairview Health Services for internal evaluation and external validation respectively. The results show that the proposed method successfully incorporates study design considerations and remains robust even with unreliable EHR timestamps. It also accurately infers causal effect directions and improves the clinical correctness of discovered graphs. Overall, this research contributes significantly to advancing our understanding of disease mechanisms through innovative methods that utilize real-world healthcare data effectively. By improving the accuracy and reliability of causal structure discovery from EHR data, it has implications for enhancing personalized medicine approaches and ultimately improving patient outcomes in various medical conditions like type-2 diabetes mellitus.