stratify-hip

stratify-hip predicts risk levels for patients who have undergone surgical treatment for hip fractures by estimating probabilities of in-hospital death, 30-day post-discharge mortality, and changes in residence to support postoperative risk stratification.

Key Features:

• Predicted outcomes: Estimates risk for in-hospital death, 30-day post-discharge mortality, and change in residence after hip fracture surgery.
• Risk categorization: Produces low, medium, and high-risk group assignments based on multivariable prediction models.
• Predictor variables: Includes age, sex, pre-fracture mobility status, presence of dementia, and pre-fracture residence type, with pre-fracture residence excluded when predicting residence change.
• Statistical modeling: Uses multivariable Fine-Gray regression to account for competing risks for time-to-event outcomes and logistic regression for the binary residence change outcome.
• Development cohort: Developed on audit data linked with hospital records for older adults in England and Wales (2011–2014) comprising 170,411 cases.
• External validation cohort: Validated on a separate cohort of 90,102 patients from 2015–2016.
• Calibration metrics: Assessed using observed:expected ratios of 0.90 for in-hospital death, 0.99 for 30-day death, and 0.94 for residence change.
• Discrimination and accuracy: Reported area under the curve (AUC) values of 73.1 for in-hospital death, 71.1 for 30-day mortality, and 71.5 for residence change, with Brier scores of 5.7, 5.3, and 5.6 respectively.
• Assumptions and missing data: Models were evaluated for assumptions, performance metrics, and sensitivity to missing data, including complete-case analyses.
• Cohort outcomes and risk distribution: In complete-case development cohort analyses, 4.8% experienced in-hospital death, 5.8% died within 30 days post-discharge, 3.7% had a change in residence, and approximately 31%/28%/41% were classified as low/medium/high risk respectively.
• External performance consistency: External validation showed similar performance metrics across the validation cohort.

Scientific Applications:

• Postoperative risk stratification: Supports stratifying patients after hip fracture surgery by individualized risk for death and residence change.
• Clinical care planning: Informs tailoring of postoperative care plans based on predicted risk profiles.
• Generalizability assessment: Enables evaluation of model performance across temporally distinct cohorts from England and Wales.

Methodology:

Models were developed using multivariable Fine-Gray regression for competing risks (in-hospital and 30-day mortality) and logistic regression for residence change, with evaluation by observed:expected ratios, AUC, Brier scores, sensitivity analyses for missing data, and complete-case analyses.