Chronic Obstructive Pulmonary Disease Patient Data for Predicting Acute Exacerbations

A retrospective study of 2,102 COPD patients admitted between 1 January 2019 and 31 December 2024. The dataset was used to develop and compare machine learning models for predicting acute exacerbation risk, with the XGBoost model achieving an AUC of 0.824 on a test set. The data was contributed by Dapeng Kuang and is shared under a CC-BY-4.0 license.

Use Cases

• Training machine learning models to predict COPD exacerbation severity based on clinical and laboratory features.
• Identifying key risk factors for acute COPD exacerbations using SHAP analysis.
• Benchmarking predictive performance of algorithms like XGBoost against other statistical methods in a clinical context.
• Developing interpretable clinical decision support tools for early intervention in COPD management.

Strengths

• Contains data from 2,102 patient records, providing a substantial sample size.
• Model performance is quantified with specific metrics, including an AUC of 0.960 on the training set and 0.824 on the test set.
• The study design includes a temporal split for training and testing, with the first 70% of chronological data used for training.
• Key predictive features like creatinine, neutrophil percentage, and D-dimer were identified through SHAP analysis.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The geographic origin of the patient data is not specified, which may limit generalizability.

Provenance

Source

figshare

Collection Method

Retrospective analysis of patient admission records.

Time Range

2019-01-01 to 2024-12-31

Freshness

Last updated 2026-05-14 05:43:29; freshness should be verified.