Objective:

To develop a dynamic AI model that predicts 30-day mortality in trauma patients using data from prehospital and in-hospital electronic health records, enhancing decision-making in trauma care.

Approach:

• Model Development: The model integrates tabular and time-series data through sequential data fusion, utilizing advanced machine learning techniques to enhance the representation of dynamic patterns and relationships.
• Dynamic Prediction: It recalculates mortality risk at every timestep across treatment phases, allowing for real-time updates as new data is collected, thereby improving the accuracy of predictions.

Key Findings:

• Machine learning models outperform traditional scoring systems like TRISS in predicting trauma outcomes, demonstrating improved accuracy.
• Dynamic risk assessment models can enhance the precision and timeliness of mortality-risk prediction, offering a more responsive approach to patient care.

Interpretation:

The integration of continuous patient-data streams and advanced machine learning techniques holds potential for enhancing trauma care outcomes.

Limitations:

• The study does not detail the specific performance metrics of the developed model, which limits the assessment of its effectiveness.
• Potential biases in data extraction from electronic health records were not addressed, raising concerns about the reliability of the data used.

Conclusion:

The model aims to provide a more accurate and timely assessment of mortality risk in trauma patients.

Sources:

• Salim A et al.
• Lapp L et al.
• Preti LM et al.
• Maurer LR et al.
• Bonde A et al.