Elderly Hip Fracture Patient Mortality Predictions with CT Imaging

A 2026 study by Jiasen Hu presents a dataset of 334 elderly hip fracture patients from two institutions for predicting 1-year postoperative mortality. It combines clinical baseline features with skeletal muscle tissue images from chest CT scans at the 12th thoracic vertebra. The dataset was used to train and validate a deep learning model and a LightGBM fusion model.

Use Cases

• Predict 1-year postoperative mortality from clinical baseline features and CT-derived skeletal muscle images using a LightGBM fusion model.
• Compare the performance of a pure deep learning model (Densenet161) against a feature fusion model using AUC and F1 score metrics.
• Analyze the relationship between thoracic vertebral skeletal muscle tissue features from CT scans and patient mortality risk.
• Validate predictive models externally using patient data from a second institution to assess generalizability.

Strengths

• Multicenter design with 221 patients for internal training and 113 for external validation.
• Model performance metrics are provided, with the fusion model achieving an AUC of 0.815 and F1 score of 0.819 on the validation set.
• Data combines specific clinical baseline features with radiological image inputs from chest CT scans.

Limitations

• Small sample size of 334 total patients, which may limit statistical power and model generalizability.
• Dataset scope is limited to a specific clinical prediction task (1-year mortality after hip fracture) and a specific patient demographic (elderly).
• The primary data file is a 442 KB DOCX document, indicating the dataset is likely a summary or supplementary file rather than a raw data repository.

Provenance

Source

figshare, authored by Jiasen Hu.

Collection Method

Retrospective multicenter study using patient data from two medical institutions.

Time Range

Study period not specified; publication date is 2026.

Freshness

Last updated in April 2026.

Geography

Institution locations not specified; presumed to be hospital data.