Clinical Evaluation of an AI Risk Prediction System (AI-TRiPS)

NCT07634185 | Not Yet Recruiting Early Phase 1

• 2 other identifiers: 354225179821
• Study Type: interventional
• Target: 1,200
• Locations: 0 countries
• Sites: N/A

Brief Summary

The goal of this clinical study is to evaluate a software device and its impact on clinician behaviour during the initial management of trauma patients in a real-world clinical setting. Known as the AI-TRiPS Device this software uses real-time prehospital data and machine learning-based risk predictions which are displayed digitally for hospital trauma teams prior patient arrival. The investigators will use a Stepped Wedge Cluster Randomised Controlled study design with an integrated process evaluation. The Device will be deployed across the London Major Trauma System where the Major Trauma Centres will be the clusters. Each cluster will transition from control (standard care) to intervention at a pre-specified time (time of transition is randomised). Primary Outcome: Clinician behaviour, assessed via the accuracy of risk prediction and clinician confidence. Secondary Outcome: Clinician acceptability, care process metrics, patient outcomes, and safety endpoints. Primary study population: Hospital trauma clinicians, following initial resuscitation of each eligible trauma patient, who will complete electronic questionnaires. Secondary study population: Adult trauma patients, data will be collected for the duration of their index admission to hospital, to assess outcomes and enable comparison with clinician risk predictions.

Conditions

Trauma; Injury; Decision Support Systems, Clinical

Keywords

Device trial, prediction tool, trauma; clinical decision support

Outcome Measures

Primary Outcomes (5)

• Clinician Risk Prediction - Mortality, Trauma Induced Coagulopathy, and Acute Kidney Injury

  Clinician participants will make probability estimates (0-100%) on index admission in each of the 3 domains.

  Baseline

• Clinician Risk Prediction - Estimation of Blood Transfusion Volume

  Clinicians will estimate the number(n) of packed red blood cell (pRBC) units required for transfusion in the first 24 hours. The estimation will take place immediately after initial resuscitation.

  Baseline

• Clinician Confidence

  Clinician Participants will self-report their confidence in their predictions using the Post-Task Confidence Scale (PTCS), a Likert scale from 1-7, where the higher the score the higher the level confidence.

  Baseline to 24 Hours - Immediately following initial clinician predictions

• Clinician Cognitive Effort

  Clinician participants self-report the mental effort required to make each prediction using the Paas Mental Effort Scale ( Likert Scale 1-9) where a lower score corresponds to low mental effort.

  Baseline to 24 Hours - immediately following risk predictions

• Risk Prediction Accuracy

  For each of the 4 domains in which predictions have been made, accuracy of these predictions will be determined with a comparison to patient outcomes. This will be done using the Brier score, however other metrics of predictive performance may also be used to perform comparisons, including measures of discrimination, calibration, and accuracy (Brier skill Score, Mean Absolute Error)

  From Discharge through to study completion, an average of 1 year.

Secondary Outcomes (13)

• Clinician Decision-Making Behaviour - Decision Making

  From discharge through to study completion, an average of 1 year.

• Clinician Decision-Making Behaviour - Appropriateness of Decision Making

  From Discharge through to study completion, an average of 1 year.

• Clinician Decision-Making Behaviour - Clinician Confidence

  Baseline to 24 hours - Immediately following initial clinician decision making

• Clinician Decision-Making Behaviour - Clinician Cognitive Effort

  Baseline to 24 Hours - immediately following risk predictions

• Clinician Decision-Making Behaviour - Time Pressure

  Baseline to 24 Hours - immediately following decision making

Study Arms (2)

AI TRIPS device intervention

EXPERIMENTAL

Patients who fit the eligibility criteria are triaged and treated at the participating trauma centre by trauma clinicians who have been exposed to the individualised risk predictions for that patient.

Device: AI-TRiPS Device

Usual Standard Care

NO INTERVENTION

Patients who fit the eligibility criteria are triaged and treated at the participating trauma centre by trauma clinicians under standard conditions.

Interventions

AI-TRiPS Device DEVICE

This is Software as a Medical Device designed to function as an aid to inform clinical situational awareness by presenting predictions of patient trajectory (probability of death, probability of trauma induced coagulopathy, probability of red cell transfusion, probability of acute kidney injury).

AI TRIPS device intervention

Eligibility Criteria

• Age: 16 Years+
• Sex: all
• Healthy Volunteers: No
• Age Groups: Child (0-17), Adult (18-64), Older Adult (65+)

You may qualify if:

• Clinician Participants
• Senior clinical decision-maker involved in the initial trauma resuscitation (e.g. consultant or senior trainee in emergency medicine, anaesthesia, intensive care medicine, or surgery).
• Based at one of the four participating Major Trauma Centres.
• Able and willing to provide informed consent.
• Completed the required study-specific training.
• Trauma Patients
• Aged 16 years and above.
• Treated and transported to a participating Major Trauma Centre by London's Air Ambulance.
• Managed by one or more participating trauma clinicians during the resuscitation.

You may not qualify if:

• Clinician Participants
• ● Decline or withdraw informed consent at any stage.
• Trauma Patients
• Aged under 16
• Not treated by London's Air Ambulance.
• Transported to a non-participating hospital.
• Not managed by any participating clinicians.
• Presenting with injuries resulting from burns, hangings, drownings, or isolated psychiatric emergencies.
• Have registered a national NHS data opt-out or otherwise requested that their routine clinical data not be used for research.

Sponsors & Collaborators

• Queen Mary University of London: lead
• Imperial College Healthcare NHS Trust: collaborator
• Congressionally Directed Medical Research Programs: collaborator
• University of Aberdeen: collaborator
• Barts & The London NHS Trust: collaborator
• St George's University Hospitals NHS Foundation Trust: collaborator
• King's College Hospital NHS Trust: collaborator
• London Ambulance Service NHS Trust: collaborator
• London's Air Ambulance Charity: collaborator

Related Publications (22)

• Kyrimi E, Neves MR, McLachlan S, Neil M, Marsh W, Fenton N. Medical idioms for clinical Bayesian network development. J Biomed Inform. 2020 Aug;108:103495. doi: 10.1016/j.jbi.2020.103495. Epub 2020 Jun 30.

  PMID: 32619692 BACKGROUND

• McLachlan S, Kyrimi E, Wohlgemut J, Perkins Z, Lagnado D, Marsh W. Explainable AI: Definition and characteristics of a good explanation for health AI. AI and Ethics. 2025:1.

  BACKGROUND

• Wohlgemut JM, Pisirir E, Stoner RS, Perkins ZB, Marsh W, Tai NRM, Kyrimi E. A scoping review, novel taxonomy and catalogue of implementation frameworks for clinical decision support systems. BMC Med Inform Decis Mak. 2024 Nov 1;24(1):323. doi: 10.1186/s12911-024-02739-1.

  PMID: 39487462 BACKGROUND

• Kyrimi E, McLachlan S, Wohlgemut JM, Perkins ZB, Lagnado DA, Marsh W. Explainable AI: definition and attributes of a good explanation for health AI. AI and Ethics. 2025:1-14.

  BACKGROUND

• Pisirir E, Wohlgemut JM, Kyrimi E, et al. A process for evaluating explanations for transparent and trustworthy ai prediction models. IEEE; 2023:388-397.

  BACKGROUND

• Kyrimi E, Stoner RS, Perkins ZB, Pisirir E, Wohlgemut JM, Marsh W, Tai NRM. Updating and recalibrating causal probabilistic models on a new target population. J Biomed Inform. 2024 Jan;149:104572. doi: 10.1016/j.jbi.2023.104572. Epub 2023 Dec 9.

  PMID: 38081566 BACKGROUND

• Wohlgemut JM, Pisirir E, Kyrimi E, Stoner RS, Marsh W, Perkins ZB, Tai NRM. Methods used to evaluate usability of mobile clinical decision support systems for healthcare emergencies: a systematic review and qualitative synthesis. JAMIA Open. 2023 Jul 12;6(3):ooad051. doi: 10.1093/jamiaopen/ooad051. eCollection 2023 Oct.

  PMID: 37449057 BACKGROUND

• Marsden MER, Perkins ZB, Pisirir E, Marsh W, Kyrimi E, Rossetto A, Lyon RL, Weaver A, Davenport R, Tai NR. Early clinical evaluation of a machine-learning system for risk prediction of trauma-induced coagulopathy in the prehospital setting. Emerg Med J. 2025 Sep 24;42(10):654-661. doi: 10.1136/emermed-2024-214396.

  PMID: 40234019 BACKGROUND

• Marsden M, Perkins Z, Marsh W, et al. Evaluation of an Artificial Intelligence (AI) system to augment clinical risk prediction of Trauma Induced Coagulopathy in the pre-hospital setting: a prospective observational study: 3. BMJ Military Health. 2022;168(5):e12.

  BACKGROUND

• Alptekin C, Wohlgemut JM, Perkins ZB, Marsh W, Tai NRM, Yet B. Presenting predictions and performance of probabilistic models for clinical decision support in trauma care. Int J Med Inform. 2025 Feb;194:105702. doi: 10.1016/j.ijmedinf.2024.105702. Epub 2024 Nov 14.

  PMID: 39579585 BACKGROUND

• Wohlgemut JM, Kyrimi E, Stoner RS, Pisirir E, Marsh W, Perkins ZB, Tai NRM. The outcome of a prediction algorithm should be a true patient state rather than an available surrogate. J Vasc Surg. 2022 Apr;75(4):1495-1496. doi: 10.1016/j.jvs.2021.10.059. Epub 2021 Dec 16. No abstract available.

  PMID: 34921966 BACKGROUND

• Tandle S, Wohlgemut JM, Marsden MER, Pisirir E, Kyrimi E, Stoner RS, Marsh W, Perkins ZB, Tai NRM. Enhancing the clinical relevance of haemorrhage prediction models in trauma. Mil Med Res. 2023 Sep 20;10(1):43. doi: 10.1186/s40779-023-00476-6. No abstract available.

  PMID: 37726859 BACKGROUND

• Perkins ZB, Yet B, Sharrock A, Rickard R, Marsh W, Rasmussen TE, Tai NRM. Predicting the Outcome of Limb Revascularization in Patients With Lower-extremity Arterial Trauma: Development and External Validation of a Supervised Machine-learning Algorithm to Support Surgical Decisions. Ann Surg. 2020 Oct;272(4):564-572. doi: 10.1097/SLA.0000000000004132.

  PMID: 32657917 BACKGROUND

• Kyrimi E, Mossadegh S, Tai N, Marsh W. An incremental explanation of inference in Bayesian networks for increasing model trustworthiness and supporting clinical decision making. Artif Intell Med. 2020 Mar;103:101812. doi: 10.1016/j.artmed.2020.101812. Epub 2020 Jan 31.

  PMID: 32143808 BACKGROUND

• Yet B, Perkins ZB, Tai NR, Marsh DWR. Clinical evidence framework for Bayesian networks. Knowledge and Information Systems. 2017;50(1):117-143.

  BACKGROUND

• Yet B, Perkins ZB, Rasmussen TE, Tai NR, Marsh DW. Combining data and meta-analysis to build Bayesian networks for clinical decision support. J Biomed Inform. 2014 Dec;52:373-85. doi: 10.1016/j.jbi.2014.07.018. Epub 2014 Aug 9.

  PMID: 25111037 BACKGROUND

• Yet B, Perkins Z, Fenton N, Tai N, Marsh W. Not just data: a method for improving prediction with knowledge. J Biomed Inform. 2014 Apr;48:28-37. doi: 10.1016/j.jbi.2013.10.012. Epub 2013 Nov 2.

  PMID: 24189161 BACKGROUND

• Perkins ZB, Yet B, Marsden M, Glasgow S, Marsh W, Davenport R, Brohi K, Tai NRM. Early Identification of Trauma-induced Coagulopathy: Development and Validation of a Multivariable Risk Prediction Model. Ann Surg. 2021 Dec 1;274(6):e1119-e1128. doi: 10.1097/SLA.0000000000003771.

  PMID: 31972649 BACKGROUND

• Durrands TH, Murphy M, Wohlgemut JM, De'Ath HD, Perkins ZB. Diagnostic accuracy of clinical examination for identification of life-threatening torsos injuries: a meta-analysis. Br J Surg. 2023 Nov 9;110(12):1885-1886. doi: 10.1093/bjs/znad285. No abstract available.

  PMID: 37847819 BACKGROUND

• Wohlgemut JM, Pisirir E, Stoner RS, Kyrimi E, Christian M, Hurst T, Marsh W, Perkins ZB, Tai NRM. Identification of major hemorrhage in trauma patients in the prehospital setting: diagnostic accuracy and impact on outcome. Trauma Surg Acute Care Open. 2024 Jan 12;9(1):e001214. doi: 10.1136/tsaco-2023-001214. eCollection 2024.

  PMID: 38274019 BACKGROUND

• Marsden MER, Kellett S, Bagga R, Wohlgemut JM, Lyon RL, Perkins ZB, Gillies K, Tai NR. Understanding pre-hospital blood transfusion decision-making for injured patients: an interview study. Emerg Med J. 2023 Nov;40(11):777-784. doi: 10.1136/emermed-2023-213086. Epub 2023 Sep 13.

  PMID: 37704359 BACKGROUND

• Wohlgemut JM, Marsden MER, Stoner RS, Pisirir E, Kyrimi E, Grier G, Christian M, Hurst T, Marsh W, Tai NRM, Perkins ZB. Diagnostic accuracy of clinical examination to identify life- and limb-threatening injuries in trauma patients. Scand J Trauma Resusc Emerg Med. 2023 Apr 7;31(1):18. doi: 10.1186/s13049-023-01083-z.

  PMID: 37029436 BACKGROUND

MeSH Terms

Conditions

Wounds and Injuries

Study Officials

• Prof. N Tai

  Queen Mary University London

  PRINCIPAL INVESTIGATOR

Central Study Contacts

Dr Mays Jawad

CONTACT

+4402078827275; research.governance@qmul.ac.uk

Prof. N Tai

CONTACT

+4402073777044; bartsheatlh.AITRIPS@nhs.net

Study Design

• Study Type: interventional
• Phase: early phase 1
• Allocation: RANDOMIZED
• Masking: NONE
• Purpose: OTHER
• Intervention Model: SEQUENTIAL
• Sponsor Type: OTHER
• Responsible Party: SPONSOR

Model Details: Early-phase, formative, interventional study using a stepped-wedge cluster randomised design with an integrated process evaluation across an integrated regional trauma system. The intervention is a digital dashboard displaying real-time prehospital data and machine learning-based risk predictions to hospital trauma teams before patient arrival. Each site transitions from control standard care to intervention at a pre-specified time, dependent on randomisation. The primary outcome is clinician behaviour, assessed via the accuracy of risk estimation and clinician confidence. Secondary outcomes include clinician acceptability, care process metrics, patient outcomes, and safety endpoints

Study Record Dates

• First Submitted: May 6, 2026
• First Posted: June 8, 2026
• Study Start: June 1, 2026
• Primary Completion (Estimated): June 1, 2027
• Study Completion (Estimated): December 1, 2027
• Last Updated: June 8, 2026

Record last verified: 2026-06

Data Sharing

• IPD Sharing: Will not share