Stratify-Hip is a tool which predicts the risk of inhospital death, 30-day death, and of a change in residence (to a higher level of dependency).
The tool’s predicted risk is based on analysis of observational data from hundreds of thousands of older adults who underwent hip fracture surgery in England and Wales. The tool provides estimates from the average risk of each individual outcome based on past outcomes observed for older adults of a similar age, gender, prefracture mobility, prefracture residence, and the presence/absence of a diagnosis of dementia. The risks of individual outcomes were then considered together in a clinically informed algorithm to determine ‘overall’ risk classification (Figure 1). This algorithm assigns patients to overall low (low risk across outcomes) overall medium (medium or high risk of change in residence), and overall high (high risk of inhospital death, high or medium risk of 30-day death), risk across outcomes.Figure 1: : Stratify-hip algorithm to enable patient assignment to three overall risk groups based on predicted risk of inhospital death, 30-day death and change in residence.
The tool provides an estimate of risk and not a definitive declaration of whether an individual will die, or transition from living at home to living in a care home.
The target users are healthcare professionals working with older adults during the index admission to hospital for hip fracture surgery.
Stratify-Hip applies to adults over the age of 60-years who are scheduled to/have already undergone surgery following admission to hospital with a hip fracture. A hip fracture is defined by the International Classification of Disease 10th Edition codes: S72.00, S72.01, S72.09, S72.10, S72.19, S72.20.
Data submitted to the National Hip Fracture Database for 170,411 patients surgically treated for hip fracture between January 1st 2011 and December 31st 2014 were selected for development and internal validation. Individual patient National Hip Fracture Database data were linked to electronic hospital records for England and Wales, and the Office of National Statistics for additional data on dementia diagnosis and death. Further details on data cleaning and person-level linkage across databases may be found in Supplementary File 1 of the open access publication here.
Fine-Gray regression was used to build risk prediction models that estimate the direction of the association between the five predictors (age, sex, prefracture function, prefracture residence, and dementia) and the cumulative incidence of inhospital death and of change in residence. A five-predictor logistic regression was used to predict the risk of 30-day death. Internal validation was completed using 100 bootstrap samples with replacement from the development dataset. Further details on model development and performance may be found in the completed publication here.
|In-hospital death model||∼1−0.9909exp(LP)||age, sex, prefracture ambulation, prefracture residence, dementia|
|30 days mortality model||∼1÷(1+exp(−LP))||age, sex, prefracture ambulation, prefracture residence, dementia|
|Change in residence model (30 days)||∼1−0.9800114exp(LP)||age, sex, prefracture ambulation, dementia|
Cut offs for the risk groups for each model:
|Model||Low risk estimate||Medium risk estimate||High risk estimate|
|In-hospital death model||<5.66%||[5.66%-12.69%[||≥12.69%|
|30 days mortality model||<5.43%||[5.43%-11.96%[||≥11.96%|
|Change in residence model||<5.19%||[5.19%-10.00%[||≥10.00%|
Data submitted to the National Hip Fracture Database for 90,102 patients treated between January 1st 2015 and December 31st 2016 were selected for external (temporal) validation. Model performance was similar for development and validation datasets. Further details on external validation performance may be found in the completed publication here.