Clinical Data for ICU-Acquired Weakness Prediction with Sarcopenia Assessment

A retrospective analysis of 1,324 critically ill patients from a prospectively maintained database, used to develop a machine learning model for predicting ICU-acquired weakness. The dataset includes predictors such as age, APACHE II score, sarcopenia, sepsis, mechanical ventilation, and lactic acid. The data was published by Peng Zheng on figshare in May 2026.

Use Cases

• Developing a machine learning model for ICU-acquired weakness risk prediction based on clinical and sarcopenia indicators.
• Analyzing the relative importance of predictors like age, APACHE II score, and sarcopenia for a critical care outcome.
• Validating the performance of an XGBoost model for clinical prediction tasks using a dataset with 927 training and 397 validation records.

Strengths

• Includes 1,324 patient records with a defined 7:3 split for training and validation.
• Model performance is documented with an AUC of 0.838 in training and 0.843 in validation.
• Key predictors were identified using LASSO regression and the Boruta algorithm, providing a feature selection rationale.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the underlying raw data is unknown, which may limit suitability assessment.
• The geographic and institutional source of the patient data is not specified, which may indicate bias.

Provenance

Source

Peng Zheng via figshare.

Collection Method

Retrospective analysis of a prospectively maintained database.

Freshness

Last updated 2026-05-14 04:24:46; freshness should be verified.