Field Measures Are All You Need: Predicting Need for Surgery in Elderly Ground-Level Fall Patients via Machine...

Area of Science:

• Geriatric Medicine
• Emergency Medical Services
• Data Science in Healthcare

Background:

• Ground-level falls (GLFs) are a major cause of mortality in elderly patients.
• Effective field triage is crucial for improving patient outcomes following GLFs.
• Traditional statistical methods may not fully capture complex patterns in medical data.

Purpose of the Study:

• To investigate the utility of machine learning algorithms in analyzing medical data for elderly GLF patients.
• To compare the effectiveness of machine learning (XGBoost) against traditional t-tests in identifying factors predicting surgical need.
• To develop data-driven clinical guidelines for paramedics using interpretable AI models.

Main Methods:

• Retrospective analysis of 715 elderly patients (over 75 years) who experienced ground-level falls.
• Calculation of P-values to assess the statistical significance of factors related to surgical necessity (P < .05).
• Application of the XGBoost machine learning algorithm to rank factor importance, interpreted using SHapley Additive exPlanations (SHAP) values and decision trees.

Main Results:

• The most significant factors predicting surgery need were Glasgow Coma Scale (GCS) (P < .001), absence of comorbidities (P < .001), and transfer-in status (P = .019).
• XGBoost identified Glasgow Coma Scale (GCS) and systolic blood pressure as the strongest contributing factors.
• The XGBoost model achieved a prediction accuracy of 90.3% on the test/train split.

Conclusions:

• XGBoost provides more comprehensive and robust insights into factors predicting surgical need compared to traditional P-values.
• Machine learning models, like XGBoost, demonstrate significant clinical applicability for enhancing medical decision-making.
• Decision trees derived from SHAP values can empower paramedics with real-time clinical guidance, with potential for broader hospital application.