The development of these tools was inspired by the realization that traditional methods of caring for wounded service members in conflicts in Iraq and Afghanistan were not working as well as they could—despite advances in technology such as body armor, tourniquets, and rapid clotting. After examining the processes (joint casualty management) and the rates of wound failure in service members, we were motivated to develop an enhanced decision-making tool that would help the critically wounded.
SC2i is currently working to develop a sepsis prediction tool, referred to as AIDEx (Artificial Intelligence Decompensation Expert). This tool is the data delivery pipeline / infrastructure which will be utilized in conjunction with our sepsis prediction algorithm, AISE (Artificial Intelligence Sepsis Expert), to improve sepsis management amongst adult ICU patients through rapid identification (4-6 hours prior to onset) and treatment of sepsis. The interface will allow clinicians to monitor large patient populations or individual patients in near-real-time, while also sending alerts to the clinician when specific patients reach specific sepsis risk thresholds. This tool and algorithm combination lay the groundwork for cutting edge artificial intelligence and data-based Clinical Decision Support Tool deployments in the Military Health System.
Description Coming Soon
The open abdomen project evaluates trauma patients who underwent at least one open laparotomy during their hospital course. Sub studies include predicting severe sepsis, organ space infection, pneumonia, acute kidney injury, and hollow viscus injuries. Clinical, biomarker, and flow cytometry data are considered in the analyses and models.
Description Coming Soon
Severe traumatic brain injury (sTBI) often leads to the complications of vasospasm or death, but they are not well predicted at this time. We have worked with researchers from Emory to identify promising statistical models for predicting these outcomes by combining clinical data with cytokine panels. Machine learning models including LASSO, CART, and random forest have identified patients at risk with strong discrimination (AUC >= 0.9).
Heterotopic ossification (HO) is defined by bone formation in an abnormal place, frequently injured muscles, and remains a common treatment complication associated with the treatment of combat-related extremity trauma. It is also known to be associated with a substantial systemic inflammatory response, increased bacterial colonization in extremity wounds, and extremity amputations. Surgical treatment with removal of the ectopic bone tissue is indicated for some symptomatic cases, which are frequent and may occur in more than 60% of the patients with traumatic and combat-related extremity amputations. Once developed, HO represents a substantial challenge to patient recovery and functional mobility. Estimating the risk of development of HO early following trauma, immediately after hospital admission, may help identify patients who could benefit most from early prophylactic treatments such as the use of nonsteroidal anti-inflammatory drugs (NSAIDs). Our group is using precision medicine tools such as machine learning techniques to investigate the clinical data of patients with combat related extremity wounds, their local, and systemic inflammatory response to develop predictive models and identify, as early as possible, patients with a higher risk of developing HO.
Acute respiratory distress syndrome (ARDS) is a serious respiratory dysfunction and remains a severe and frequent complication of critical illness, and can occur in diverse settings, including following trauma. It is estimated to affect up to 16% of trauma and 10% of all ICU patients. The development of ARDS is associated with worse outcomes and high mortality rates despite of the advancements in supportive treatments including new mechanical ventilatory protocols. Underdiagnose remains a substantial concern, therefore anticipating and recognizing the individual risk each patient may have to develop ARDS may allow for an improvement in patient treatment and ultimately outcomes. Our group is studying standard-of-care data, the inflammatory response, and gene expression of critical care patients to identify key biomarkers associated with ARDS to enable the development of statistical models using various machine learning techniques. This will allow us to predict and estimate the individual risk of patients, facilitate resource management in ICUs, and ultimately develop clinical decision support tools (CDST) to aid in treatment of patients with acute respiratory dysfunction and ARDS.