Postoperative Delirium Phenotype Analysis via SHAP and Clustering

2026-03-23 is the last update date for this 5.5 KB XLS file by Xiaochen Zheng, which contains results from a study on postoperative delirium (POD) phenotypes. The dataset presents model performance metrics and SHAP feature scores from a case study using real-world electronic health record data covering preoperative, intraoperative, and postoperative stages. It is licensed under CC BY 4.0.

Use Cases

• Analyze SHAP feature scores to identify key drivers of personalized POD risk prediction across preoperative, intraoperative, and postoperative stages.
• Apply unsupervised clustering on SHAP feature scores to uncover distinct patient subgroups with differing clinical characteristics.
• Compare model performance metrics for machine learning models trained on heterogeneous EHR data to predict POD risk in elderly patients.

Strengths

• Data is recent, with a last update of 2026-03-23.
• The study uses a multi-stage approach, covering preoperative, intraoperative, and postoperative patient data.
• The dataset is openly shared under a CC BY 4.0 license.

Limitations

• The dataset is very small at 5.5 KB, indicating limited scope and likely containing summary results rather than raw patient records.
• Specific row count, column names, and sample data are unavailable, limiting detailed analysis.
• The data is derived from a specific case study on elderly POD patients, which may limit generalizability to other populations.

Provenance

Source

Xiaochen Zheng via figshare.

Collection Method

Results from a study combining supervised machine learning for POD risk prediction with unsupervised clustering on SHAP scores from real-world EHR data.

Freshness

Last updated 2026-03-23.