Identifying Age-Specific Risk Factors for Poor Outcomes After Trauma With Machine Learning
Heather R Kregel1, Gabrielle E Hatton1, John A Harvin2
1Division of Acute Care Surgery, Department of Surgery, McGovern Medical School at UTHealth, Houston, Texas; Center for Surgical Trials and Evidence-Based Practice, McGovern Medical School at UTHealth, Houston, Texas; Center for Translational Injury, McGovern Medical School at UTHealth, Houston, Texas.
The Journal of Surgical Research
|February 6, 2024
Summary
Risk factors for poor outcomes after trauma differ by age. Age-specific models can improve care for older trauma patients by identifying unique risk factors and improving hospital benchmarking.
Area of Science:
- Trauma Surgery
- Geriatric Medicine
- Data Science in Healthcare
Background:
- Current trauma risk stratification models are not age-specific.
- This can lead to suboptimal care and inaccurate quality benchmarking for older adults.
- Age-dependent risk factors for trauma outcomes are hypothesized.
Purpose of the Study:
- To assess if risk factors for poor outcomes after trauma are age-dependent.
- To determine if the relative importance of risk factors varies by age.
Main Methods:
- A cohort study of severely injured adult trauma patients (2014-2018) using trauma registry data.
- Random forest algorithms were used to predict poor outcomes (death or complication) in all, younger, and older (≥55 years) patient cohorts.
- Mean Decrease in Accuracy (MDA) was calculated to assess variable importance and significant differences between age cohorts.
Main Results:
- Older patients (25%) had higher poor outcome rates (33%) compared to younger patients (12%).
- Head injury was the top predictor across all cohorts; age was key in the full cohort after head injury.
- Risk factor importance varied by age: surgery for younger, Glasgow Coma Scale for older patients.
Conclusions:
- Supervised machine learning revealed age-specific differences in trauma risk factors and their associations with poor outcomes.
- Age-specific models can enhance hospital benchmarking and identify quality improvement targets for older trauma patients.
