A22 Casualty infection risk prediction and diagnosis field care decision support

Objectives Advancements in combat casualty care and personnel protective equipment led to increased survivability of grievous battlefield injuries during U.S. Operations Iraqi Freedom and Enduring Freedom (OIF/OEF). Accompanying these important advances were high rates of post-trauma infections (25–30%), including high-consequence infections, such as invasive fungal wound infections (IFIs), sepsis, and infections attributed to multidrug-resistant Gram-negative bacilli (MDRGN), which are associated with long-term morbidity (e.g., limb loss) and mortality. 1–3 The Trauma Infectious Disease Outcomes Study (TIDOS) collected data from wounded warriors from 2009–2014 and findings have been utilized to further the understanding of combat trauma-related infections, as well as identifying infection risk factors and examining clinical outcomes. 3 While TIDOS has focused on infections following trauma sustained during OIF/OEF, there is a high likelihood that similar high-consequence infections will occur in future conflicts. A further concern is that near-peer scenarios in future conflicts would result in a lack of air superiority for medical evacuation, resulting in prolonged field care. Between 2001 and 2010, combat casualties injured in Iraq or Afghanistan arrived at Landstuhl Regional Medical Center (LRMC; Germany) a median of 38 hours (~1.5 days) following injury. As sepsis and IFIs typically develop a median of 3 days post-injury (median of 7 days post-injury for MDRGN infections), 1 3 4 any delays in casualty evacuation from the operational theater to higher role facilities will result in prolonged field care and presumably greater morbidity from these high-consequence infections. Thus, healthcare providers across a range of expertise along the evacuation chain of care, coupled with limited forward diagnostic and therapeutic capabilities need maximal decision support tailored to the specific patient and injury setting to mitigate these infections. As information collected from combat casualties on injury characteristics, clinical signs, symptoms, and laboratory findings have the potential to support both risk stratification and clinical decision-making along the pathway of care from the prehospital setting through medical evacuation to more advanced capabilities, 5 we are utilizing TIDOS data to develop a machine learning-based clinical decision support tool (CDST) for longitudinal use from the prehospital setting through hospital admission. Methods Design: Retrospective cohort study utilizing machine learning. Setting: Injury characteristics, clinical signs, symptoms and laboratory findings collected in the austere (prehospital) setting and following admission to a higher role of care military hospital. Participants: U.S. active-duty military personnel wounded during deployment. Main Outcome(s) and Measure(s): In the prehospital setting, particularly with prolonged field care, use of the CDST will improve triage and early decision support through use of immediately available data related to injury characteristics (severity, wounding pattern, mechanism, and environmental setting) and vital signs (within 1st 72 hours), identifying those at highest risk for developing high-consequence infections (figure 1). The tool may also be used on a longitudinal basis as the patient transitions through the echelons of care, acquiring greater accuracy for risk prediction as clinical and laboratory data become available in the hospital setting. Results and Conclusions Machine learning and deep learning will provide novel insights into how combinations of independent risk factors influence infection aetiology and clinical outcome coupled with information on pretest probability (prevalence) of high-priority microbial threats. The use of a functional tool that meets the needs of front-line care givers and focuses on early indicators available in the prehospital setting will directly support frontline casualty care, as well as prolonged and en-route care. Abstract A22 Figure 1 Predictive modelling workflow and application. The initial risk stratification model will be created using the earliest data obtained before admission to hospitals, while longitudinal support model will be generated using data obtained after admission to hospitals. The lower arrows of left and right columns indicate the future use of the models Disclaimer The view(s) expressed herein are those of the author(s) and do not reflect the official policy or position of Uniformed Services University of the Health Sciences, Henry M. Jackson Foundation for the Advancement of Military Medicine, Inc., National Institutes of Health or the Department of Health and Human Services, the Defense Health Agency, the Departments of the Air Force, Navy, Army, or the Department of Defense, or the U.S. Government. Funding Support for this work was provided by the Infectious Disease Clinical Research Program (IDCRP), a Department of Defense program executed through the Uniformed Services University of the Health Sciences, Department of Preventive Medicine and Biostatistics through a cooperative agreement with The Henry M. Jackson Foundation for the Advancement of Military Medicine, Inc. (HJF). This project has been funded by the National Institute of Allergy and Infectious Diseases (NIAID), National Institutes of Health (NIH), under Inter-Agency Agreement Y1-AI-5072, the Defense Health Program, U.S. DoD, under award HU0001190002, and through Defense Health Program award HU00012320031. The funders had no role in study design, data collection, data analysis, data interpretation, or writing the manuscript. References Tribble DR, Murray CK, Lloyd BA, et al . After the battlefield: infectious complications among wounded warriors in the trauma infectious disease outcomes study. Mil Med. 2019; 184 (Suppl 2):18–25. Campbell WR, Li P, Whitman TJ, et al . Multi-drug-resistant gram-negative infections in deployment-related trauma patients. Surg Infect. 2017; 18 :357–67. Lewandowski LR, Weintrob AC, Tribble DR, et al . Early complications and outcomes in combat injury related invasive fungal wound infections: a case-control analysis. J Orthop Trauma. 2016; 30 :e93–9. Lloyd B, Weintrob A, Rodriguez C, et al . Effect of early screening for invasive fungal infections in U.S. service members with explosive blast injuries. Surg Infect . 2014; 15 :619–26. Potter BK, Forsberg JA, Silvius E, et al . Combat-related invasive fungal infections: development of a clinically applicable clinical decision support system for early risk stratification. Mil Med. 2019; 184 :e235–42. Competing Interests All authors have completed the ICMJE uniform disclosure form at http://www.icmje.org/disclosure-of-interest/ and declare: DT, NE, ISP, LS, DC, and KM received funding through their institution for this study as reported in the funding statement. DT, NE, ISP, and LS also received funding from the Defense Health Program and USU through their institution for other research protocols during the past 36 months. DT, NE, and LS received support from grants paid to their institutions for travel to meetings/conferences. DT serves on Data Safety Monitoring Boards (DSMB) for the National Institute of Allergy and Infectious Diseases Division of Microbiology and Infectious Diseases and Department of Defense, and serves on the University of Vermont COBRE Advisory Board. DC serves on a DSMB for a clinical trial being run by IDCRP hosted at the Uniformed Services University of the Health Sciences. DC is also an editor of the journal, Machine Learning and Artificial Intelligence in Diagnostics . NE received payment from the George Washington University Milken Institute School of Public Health for lectures per being part-time faculty.