An algorithm for direct causal learning of influences on patient outcomes
作者：
Highlights：
• We propose a new algorithm (DCL) to learn the direct causal influences of a target such as a disease outcome.
• DCL uses Bayesian network scoring and a novel deletion algorithm.
• Results show DCL clearly outperforms PC with respect to accuracy and runtime.
• Found SNPs directly causal of LOAD on NISCH & PRKG1 and validated by prior studies.
• Further Validated ER cat. & HER2 status causal of 5 & 10-year breast cancer survival/death, resp.
摘要
•We propose a new algorithm (DCL) to learn the direct causal influences of a target such as a disease outcome.•DCL uses Bayesian network scoring and a novel deletion algorithm.•Results show DCL clearly outperforms PC with respect to accuracy and runtime.•Found SNPs directly causal of LOAD on NISCH & PRKG1 and validated by prior studies.•Further Validated ER cat. & HER2 status causal of 5 & 10-year breast cancer survival/death, resp.
论文关键词：Bayesian-score based learning,Constraint-based learning,Causal discovery,Simulated data,Predictive medicine,Clinical decision support
论文评审过程：Received 22 April 2016, Accepted 25 October 2016, Available online 5 November 2016, Version of Record 12 November 2016.
论文官网地址：https://doi.org/10.1016/j.artmed.2016.10.003
