Accepted for/Published in: Journal of Medical Internet Research
Date Submitted: Oct 20, 2025
Open Peer Review Period: Oct 20, 2025 - Dec 15, 2025
Date Accepted: May 8, 2026
(closed for review but you can still tweet)
Warning: This is an author submission that is not peer-reviewed or edited. Preprints - unless they show as "accepted" - should not be relied on to guide clinical practice or health-related behavior and should not be reported in news media as established information.
Dynamic prediction of mortality in trauma patients: Development of a hybrid neural network model for risk assessment
ABSTRACT
Background:
The trajectory of a trauma patient is often complex and non-linear. Real-time estimation of the mortality risk from prehospital care to discharge is critical for point-of-care decision-making and for benchmarking the quality of care. Using data automatically extracted from electronic health records, this study aims to create an artificial intelligence risk prediction model for 90-day mortality in trauma patients capable of performing predictions at any time point through treatment phases from prehospital to discharge.
Objective:
To develop risk prediction models based on transformer architectures for dynamic mortality prediction
Methods:
Comprehensive data on pre- and in-hospital care of trauma patients treated in Denmark, Capital Region between 2017 and 2024 was used to train, validate, and evaluate the AI model. Demographic, comorbidity, and injury specific variables were structured as tabular data. Temporal data on vital signs, laboratory test results, and medications were structured as sequences with temporally determined dynamic bin sizes. The model architecture combines both tabular and sequential input data. A unified input is processed by multiple layers of transformer encoders before being passed into a neural network for binary classification. Model performance was assessed using area under the receiver operating characteristic curve (AUROC) and area under precision-recall curve (AP) on a holdout dataset comprising 20% of the total data.
Results:
A total of 8,372 patients were included and the 90-day mortality rate was 4.55%. The model achieved an AUROC of 0.963 (95% CI: 0.934-0.993) and an AUPRC of 0.800 (95% CI: 0712-0.870) predicting 90-day mortality in the holdout evaluation set using full-length trajectories, comprising 1675 patients.
Conclusions:
We designed a dynamic, automated, and highly accurate model that allows 90-day mortality risk prediction at any point in time during the complex and non-linear trajectory of the trauma patient. The model could be used as decision support for triage, patient deterioration alerts, bedside decision-making and family counselling. Clinical Trial: NA
Citation
Request queued. Please wait while the file is being generated. It may take some time.
Copyright
© The authors. All rights reserved. This is a privileged document currently under peer-review/community review (or an accepted/rejected manuscript). Authors have provided JMIR Publications with an exclusive license to publish this preprint on it's website for review and ahead-of-print citation purposes only. While the final peer-reviewed paper may be licensed under a cc-by license on publication, at this stage authors and publisher expressively prohibit redistribution of this draft paper other than for review purposes.