A New Model Improves Prediction of Death Risk in Severely Injured Patients

A new model improves the prediction of death risk in severely injured patients. A death risk prediction tool for severely injured patients has been revised to adapt to the evolving population and missing data. This model, used since 2013, needed updating due to the aging of trauma patients and the need for more precise management of missing information.

The study relied on data from 53,738 severely injured patients documented between 2022 and 2023. Of these, 75% were used to develop the new model, while the remaining 25% were used to validate it. The patients had an average age of 55, with a majority being male (69%), and an average injury severity score of 18 points. Missing data, which could reach up to 17.5% for certain parameters such as excess base rate, were handled differently. Now, a missing value is considered normal unless other clinical elements suggest altered physiology. For example, pre-injury health status was estimated based on age, while pupil size or light reaction were inferred from the severity of brain injuries.

The new model, called RISC III, has demonstrated an exceptional ability to distinguish between survivors and non-survivors. Its accuracy rate, measured by the area under the ROC curve, reaches 0.946, a result confirmed by validation data at 0.949. Observed mortality was 13.1% in the development sample, while the model-predicted mortality was 13.0%, showing an almost perfect match. In the validation sample, the figures were 13.2% and 13.0%, respectively.

The most notable improvement concerns the handling of missing data. In the previous model, a missing value was neutral, which could distort predictions. Now, a missing value is presumed normal unless clinical evidence suggests otherwise. This approach has improved the model’s overall accuracy, particularly for elderly patients, whose risk of death was previously underestimated.

Adjustments made to the model include increased weighting for age, particularly for patients over 85, as well as better consideration of physiological parameters such as blood pressure, hemoglobin levels, or coagulation status. These changes reflect a clinical reality where elderly patients, often frail, require a more nuanced assessment of their prognosis.

The RISC III model retains the same fifteen predictors as its predecessor but with revised categories and weightings. For example, age has been divided into more precise brackets, and the normal values of certain physiological parameters have been redefined to better reflect their impact on survival. Blood pressure, previously divided into multiple categories, has been simplified to better align with clinical observations.

Model validation showed that predicted mortality almost perfectly matched observed mortality, with a gap of less than 0.2%. This represents a significant improvement over the previous model, which underestimated the risk of death, particularly in the elderly. Additionally, the model remains robust even when some data are missing, provided that fewer than five predictors are absent.

This new model will now be used to assess the quality of care in hospitals and for scientific analyses. It allows for the comparison of hospital performance by adjusting observed mortality rates against predictions, thus providing a fairer and more accurate view of the management of severely injured patients.

A new model improves the prediction of death risk in severely injured patients. A death risk prediction tool for severely injured patients has been revised to adapt to the evolving population and missing data. This model, used since 2013, needed updating due to the aging of trauma patients and the need for more precise management of missing information.

The study relied on data from 53,738 severely injured patients documented between 2022 and 2023. Of these, 75% were used to develop the new model, while the remaining 25% were used to validate it. The patients had an average age of 55, with a majority being male and an average injury severity score of 18 points. Missing data, which could reach up to 17.5% for certain parameters such as excess base rate, were handled differently. Now, a missing value is considered normal unless other clinical elements suggest altered physiology. For example, pre-injury health status was estimated based on age, while pupil size or light reaction were inferred from the severity of brain injuries.

The new model, called RISC III, has demonstrated an exceptional ability to distinguish between survivors and non-survivors. Its accuracy rate, measured by the area under the ROC curve, reaches 0.946, a result confirmed by validation data at 0.949. Observed mortality was 13.1% in the development sample, while the model-predicted mortality was 13.0%, showing an almost perfect match. In the validation sample, the figures were 13.2% and 13.0%, respectively.

The most notable improvement concerns the handling of missing data. In the previous model, a missing value was neutral, which could distort predictions. Now, a missing value is presumed normal unless clinical evidence suggests otherwise. This approach has improved the model’s overall accuracy, particularly for elderly patients, whose risk of death was previously underestimated.

Adjustments made to the model include increased weighting for age, particularly for patients over 85, as well as better consideration of physiological parameters such as blood pressure, hemoglobin levels, or coagulation status. These changes reflect a clinical reality where elderly patients, often frail, require a more nuanced assessment of their prognosis.

The RISC III model retains the same predictors as its predecessor but with revised categories and weightings. For example, age has been divided into more precise brackets, and the normal values of certain physiological parameters have been redefined to better reflect their impact on survival.

Model validation showed that predicted mortality almost perfectly matched observed mortality, with a gap of less than 0.2%. This represents a significant improvement over the previous model, which underestimated the risk of death, particularly in the elderly. Additionally, the model remains robust even when some data are missing, provided that fewer than five predictors are absent.

This new model will now be used to assess the quality of care in hospitals and for scientific analyses. It allows for the comparison of hospital performance by adjusting observed mortality rates against predictions, thus providing a fairer and more accurate view of the management of severely injured patients.

Sources Used

Report Source

DOI: https://doi.org/10.1007/s00068-026-03224-2

Title: Prediction of risk of death in severely injured patients: the revised injury severity classification score, version 3 (RISC III)

Journal: European Journal of Trauma and Emergency Surgery

Publisher: Springer Science and Business Media LLC

Authors: Rolf Lefering; Sebastian Imach; Dan Bieler