Leveraging implicit expert knowledge for non-circular machine learning in sepsis prediction
作者：
Highlights：
• Accurate prediction of sepsis requires grounding in real-world clinical practice.
• Current AI models risk circularity if training and test data are not independent.
• A ground truth is collected from independent expert judgements via a questionnaire.
• Implicit expert knowledge can be collected with high inter-rater reliability.
• AI models trained on this ground-truth data yield state-of-the-art sepsis prediction.
摘要
•Accurate prediction of sepsis requires grounding in real-world clinical practice.•Current AI models risk circularity if training and test data are not independent.•A ground truth is collected from independent expert judgements via a questionnaire.•Implicit expert knowledge can be collected with high inter-rater reliability.•AI models trained on this ground-truth data yield state-of-the-art sepsis prediction.
论文关键词：Machine learning in health care,Sepsis prediction
论文评审过程：Received 17 September 2018, Revised 24 June 2019, Accepted 10 September 2019, Available online 24 September 2019, Version of Record 24 September 2019.
论文官网地址：https://doi.org/10.1016/j.artmed.2019.101725
