Parkland Trauma Index of Mortality
![](https://pccinnovation.org/wp-content/uploads/2023/04/src-hbs.png)
PREDICTING MORTALITY IN TRAUMA PATIENTS
The Parkland Trauma Index of Mortality (PTIM) is a real-time predictive model of in-hospital mortality assessment of trauma patients for clinical decision support. Most deaths due to trauma will occur in the immediate aftermath of the injury – in the critical first hours and days of an admission. This model was developed because provider teams needed a dynamic tool to estimate patient mortality risk for prioritization of interventions as the teams must make many consequential decisions about trauma care in rapid succession; i.e., when to stabilize versus intervene or how to sequence interventions. In many cases, these decisions can feel like equal parts instinct, art, and science. Prior industry models provided only a static, one-time score at the time of admission. The PTIM machine learning algorithm is the only known model that uses Electronic Medical Record (EMR) data to predict―every hour―48-hour mortality during the first 72 hours of hospitalization, thus evolving with the patient’s physiologic response to trauma. The model has been running at Parkland since August 2019. In the first year of go-live, the model made over 18.000 hourly predictions on approximately 432 trauma level 1 patients at Parkland, with consistent performance.
This year, to enhance the solution’s usefulness to end-user clinicians, PCCI designed, leveraging Islet, a new, Epic-integrated interface that provides three distinct areas of patient curated information: (1) current PTIM risk score and its top contributors; (2) over-time trend of PTIM; and (3) historic table of PTIM scores and all the underlying variables that were used by the model to predict that timestamp’s risk level. This interface enables Trauma care teams to collaborate more efficiently and standardize PTIM within patients’ care-management journeys in the Trauma center.
![](https://pccinnovation.org/wp-content/uploads/2023/04/Trauma-h.jpg)