AI-enabled Transfusion Algorithm for Personalized Trauma Care

Funding opportunity

AI-enabled Transfusion Algorithm for Personalized Trauma Care

Anticipated timeline and budget

Background

Hemorrhage, the main cause of death on the battlefield, is difficult to treat and early identification of hemorrhagic shock is challenging. Massive hemorrhage protocols (MHP) with blood products, such as red blood cells, fresh frozen plasma, platelets, and whole blood, are recommended after severe bleeding injuries. However, different transfusion protocols and blood products are controversial, including presumptive formula-driven transfusion and algorithm-guided transfusion. Canadian Forces Health Services (CFHS) requires advanced technologies and methods to improve damage control resuscitation and survivability on operations. With the availability of large volume retrospective data and AI R&D efforts, it is possible to build predictive models and transfusion algorithms to assist in timely decision-making for optimal blood transfusions. This funding opportunity seeks proposals to develop predictive models for trauma-induced blood transfusion, serving as a foundation for a large-scale, multi-year AI/ML assisted blood transfusion translational research program.

Objective

The funding opportunity seeks submissions to develop prototype predictive AI/ML models to assist trauma-induced blood transfusions. The models will use advanced machine learning algorithms and large-scale data to predict the need for massive hemorrhage protocol, the need for massive transfusion, and the transfusion of blood products (RBCs, plasma, and platelets) over time. The research can be conducted on a broader population, not just the Canadian Armed Forces (CAF) and Canadian Veteran populations.

Phases of progress/milestones

The project is expected to be completed in two years, with each year having achieved milestones. Currently, $140,000 is provided for year one, with potential for $250,000 additional funding in year two, pending on funding.

Year one: modelling development

Milestone 1 - The goal is to develop AI/ML models for personalized transfusion using retrospective data and advanced principles. The models aim to accurately predict MHP activation requirements for massive transfusions.

Milestone 2 -Use ML strategies to identify data features most important to the modelling tasks described above. These features would be interpreted to improve clinical practices for trauma induced transfusion decision-making.

Year two (pending funding availability): model validation and deployment solution development

Milestone 1 - Develop validation models as a baseline. Develop a ML validation model with type and severity of bleeding as outcomes: This would tell us if the data in the two models from the primary objectives are consistent (if type and severity of bleeding predict MHP). This validation model can also be used as a standalone bleed prediction assistant.

Milestone 2 - The model deployment solution should include software, hardware, and networking elements, considering feasibility and accessibility.

Milestone 3 - The final goal is to validate and improve the ML algorithms using retrospective and prospective data for standard MHPs, massive transfusion status, and blood transfusion outcomes within 24-hour post-injury among severely injured patients.

Desired outputs

Year one

In the first year, AI/ML models will be prototyped to target MHP activation, massive transfusion needs, and best transfusion strategies in hemorrhagic trauma. Key findings will be disseminated in research papers and conference presentations.

Year two

Additional validation and refined AI/ML models were refined, and recommendations for further transfusion algorithm development to be presented. Research papers will report main findings of the project and provide recommendations for further development of transfusion algorithms. Communications will be presented through conferences and presentations in national and international meetings. Finally, model deployment solution for model validation.

Applicant qualifications and requirements for selection

The research team must have relevant advanced degrees in all areas of research within the approach proposed.

Proposals should be led by a senior investigator with an MD in trauma care, involving data science experts and a multidisciplinary team with expertise in clinical practice and trauma resuscitation.

The research group should have computational power for local AI/ML model training and development, be located within the Greater Toronto Area, have access to a large trauma center population in a Canadian Hospital, and include follow-up plans for model validation, refinement, and deployment solution development, ideally with prospective data.

Collaborative agreement

The successful research group will collaborate closely with a team at DRDC Toronto Research Centre, involving various elements such as:

• Intellectual input and collaboration on project design, data collection, data sharing, data analysis and interpretation, and modelling.
• Co-authorship in publications, briefings and conference presentations.

Intellectual property (IP) ownership

The award recipient will own any foreground IP created by virtue of the funding agreement.

The sponsor may request permission to use such intellectual property for their own purposes. Funding agreement with award recipient will stipulate that award recipients will not unreasonably withhold such permission.

Application deadline

Please download and submit the Research Funding Application form.

Enquiries

Questions about this funding opportunity can be emailed to the VAC Research office.