Prospective Validation of the SHOCKMATRIX Hemorrhage Predictive Model (SHOCKMATRIX)

External Validation of a Real-time Machine Learning-based Predictive Model for Early Severe Hemorrhage and Hemorrhage Resource Needs in Trauma Patients

Management of post-traumatic severe hemorrhage remains a challenge to any trauma care system. Studying integrated and innovative tools designed to predict the risk of early severe hemorrhage (ESH) and resource needs could offer a promising option to improve clinical decisions and then shorten the time of intervention in the context of pre-hospital severe trauma. As evidence seems to be lacking to address this issue, this ambispective validation study proposes to assess on an independent cohort the predictive performance of a newly developed machine learning-based model, as well as the feasibility of its clinical deployment under real-time healthcare conditions.

Study Overview

• Status: Completed
• Conditions: Wounds and Injuries; Hemorrhagic Shock; Traumatic Shock
• Intervention / Treatment: Other: Ambispective validation of machine learning-based predictive model

Detailed Description

Background: Hemorrhagic shock remains the leading cause of early preventable death in severely injured patients. When a severe hemorrhage occurs shortly after serious trauma, thus defining an early severe hemorrhage (ESH), its management becomes highly challenging. In this context, improving clinical decisions and shortening the time of intervention, known as a critical endpoint, may require designing innovative tools for early detection as well as studying their integration within the routine healthcare process.

Objective: Part of the TRAUMATRIX project led by the Traumabase Group in partnership with Capgemini Invent and several research centers (Ecole polytechnique, CNRS, EHESS), this study aims to externally validate a recently developed machine learning-based predictive model for ESH in trauma patients. This model, previously trained on a high-quality trauma database named Traumabase, offers a specific ability to handle missing values.

Materials and Methods: At least 1500 adult trauma patients from 8 French trauma centers will be included for a six-24 month period with a retrospective and prospective sample. ESH will stand as our primary outcome, defined as any of the following events occurring within the first hours of trauma management: any packed red blood cell (RBC) transfusion in the resuscitation room, or transfusion exceeding 4 RBCs within the first 6 hours, or emergency hemostatic intervention (surgery or interventional radiology), or death in an unambiguous setting of uncontrolled, objectified hemorrhage. Data of interest will be collected in two phases: (1) from the prehospital phase of the trauma management, where the variables needed to calculate the algorithmic prediction of ESH (10 inputs) as well as the clinical prediction from the attending trauma leader receiving in the resuscitation room a pre-alert call from the dispatch center, will be recorded in real-time using a dedicated user-friendly smartphone interface developed by the Capgemini Invent teams; (2) from a delayed phase where a classic inclusion in the Traumabase® will be performed to retrieve the component variables of the ESH composite endpoint, and a feedback survey will be sent to the trauma teams involved in the study to collect additional informative data. The prospective data collected, we will compare to a retrospective cohort predictive performance of two systems, namely the clinical trauma expert versus our machine learning-based predictive model.

• Study Type: Observational
• Enrollment (Actual): 1584

Contacts and Locations

Study Locations

• France
  • Clichy, France, 92110
    • Beaujon Hospital AP-HP, Anesthesia-Intensive Care Department
  • La Tronche, France, 38700
    • Grenoble Alpes University Hospital
  • Le Kremlin-Bicêtre, France, 94270
    • Bicêtre Hospital AP-HP, Anesthesia-Intensive Care Department
  • Lille, France, 59037
    • Lille University Hospital, Anaesthesia and Intensive Care Unit
  • Paris, France, 75013
    • Pitié-Salpêtrière Hospital AP-HP, Anesthesia-Intensive Care Department
  • Paris, France, 75015
    • Georges-Pompidou European Hospital AP-HP, Anesthesia-Intensive Care Department
  • Strasbourg, France, 67200
    • University Hospitals Strasbourg, Anaesthesia, Intensive Care and Peri-Operative Medicine Department
  • Toulouse, France, 31059
    • University Hospital of Toulouse, Polyvalent Intensive Care

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: No

• Sampling Method: Non-Probability Sample

Study Population

Every severe trauma adult patient to be admitted to a participating center excluding those already diagnosed with active hemorrhage from computed tomography findings, and those with prior traumatic cardiac arrest

Description

Inclusion Criteria:

• every severe trauma adult patient to be admitted to a participating center

Exclusion Criteria:

• patients already diagnosed with active hemorrhage from computed tomography findings;
• patients with prior traumatic cardiac arrest
• patient under 18 years of age
• opposition of patient or relative

Study Plan

How is the study designed?

Number of groups / cohorts

1

Cohorts and Interventions

Group / Cohort	Intervention / Treatment
Prehospital severe trauma patients; Every severe trauma patient 18 years of age or older to be admitted to a participating center excluding those already diagnosed with active hemorrhage from computed tomography findings and those with prior traumatic cardiac arrest	Other: Ambispective validation of machine learning-based predictive model; Retrospective and prospective validation of a machine learning model to predict major haemorrhage in trauma patients compared to clinician prediction

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Fβ-score, with β = 4	A configurable single-score metric for evaluating a binary classification model. The parameter β allows placing more emphasis on false-negative prediction error. The formula for Fβ-score is given below (TP true positives, FN false negatives, FP false positives): Fβ= ((1+β^2 ).TP)/((1+β^2 ).TP+ β^2.FN+FP)	18 months

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Common binary classification metrics	Sensitivity Se, Specificity Sp, Accuracy Acc, Positive Predictive Value PPV, Negative Predictive Value NPV	18 months

Collaborators and Investigators

• Sponsor: Assistance Publique - Hôpitaux de Paris
• Collaborators: Traumabase Group; Capgemini Invent; Ecole polytechnique; EHESS (Ecole des hautes études en sciences sociales); CNRS (Centre national de la recherche scientifique)

Study record dates

Study Major Dates

• Study Start (Actual): July 1, 2022
• Primary Completion (Actual): June 24, 2024
• Study Completion (Actual): June 24, 2024

Study Registration Dates

• First Submitted: February 14, 2024
• First Submitted That Met QC Criteria: February 14, 2024
• First Posted (Actual): February 21, 2024

Study Record Updates

• Last Update Posted (Actual): March 25, 2025
• Last Update Submitted That Met QC Criteria: February 14, 2025
• Last Verified: February 1, 2025

More Information

Terms related to this study

• Keywords: Predictive model; Machine learning; Ambispective validation; Missing values
• Additional Relevant MeSH Terms: Pathologic Processes; Wounds and Injuries; Shock; Hemorrhage; Shock, Hemorrhagic; Shock, Traumatic
• Other Study ID Numbers: CRCBDD1712

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No
• product manufactured in and exported from the U.S.: No