TPF Machine Learning Algorithms

Terminated

• Conditions: Tibial Plateau Fracture

• Registration Number: NCT04983316
• Lead Sponsor: Universitaire Ziekenhuizen KU Leuven

Brief Summary

To adopt a machine learning technique to decide whether operative or non-operative treatment will result in the best patient-outcome.

Detailed Description

The overall goal is to adopt a machine learning technique to decide whether operative or non-operative treatment will result in the best patient-outcome.

The primary objectives are to identify the most suitable machine learning algorithm to predict the best treatment for future patients. Whether conservative or operative treatment will lead to the best patient outcome, will be decided on the predicted KOOS score. Several input factors, such as treatment (conservative or operative), number of fracture fragments, location of the fracture, soft tissue involvement,...for each patient will be used as training data for the algorithm. Some of these input data will be derived from CT-scans. Therefore, the CT scans will be segmented in Mimics, for which UZ Leuven recently purchased licenses. The output variable of the training data will be the KOOS score of each patient. Based on the input and output variable, the algorithm will determine a relation between these. For future patients of which the input variable are known, the output variable (=KOOS score) will be predicted both in case of operative and conservative treatment. We hypothesize that the prediction will be improved by adding more input data over time.

To secondary objective is to identify clinical and radiological factors that help predicting the best treatment for future patients.

As an outlook, the machine learning technique could be implemented in the future in clinical practice and utilized as a patient-specific planning tool for tibial plateau fracture management by aiding the surgeon to select the best treatment for a given case. The collected data in this registry will be used to validate the machine learning model. Patients will not yet be treated based on the results of the developed model, the trauma surgeon is responsible to decide which treatment option is best for the patient.

Study Timeline

• Study Start: 2020-10-05
• Study First Submitted: 2021-07-16
• Study First Submitted QC: 2021-07-27
• Study First Posted: 2021-07-30
• Primary Completion: 2023-06-27
• Status Verified: 2024-12
• Study Completion: 2024-12-06
• Last Update Submitted: 2024-12-06
• Last Update Posted: 2024-12-12

Recruitment & Eligibility

• Status: TERMINATED
• Sex: All
• Target Recruitment: 70

Inclusion Criteria

• Age > 18 years
• Proximal tibia plateau fracture
• Patient is able to attend follow-up visits

Exclusion Criteria

• Age < 18 years
• Bilateral fractures
• Neurologic disorders (ie paraplegia, CVA, dementia etc.)
• Not understanding Dutch or English

Study & Design

• Study Type: OBSERVATIONAL
• Study Design: Not specified

Primary Outcome Measures

Name	Time	Method
Machine learning algorithm	1 year	To identify the most suitable machine learning algorithm that predicts the best treatment for future patients. The prediction will be improved over time by additional input.

Secondary Outcome Measures

Name	Time	Method
Clinical factors	1 year	To identify clinical factors that help predicting the best treatment for future patients
Radiological factors	1 year	To identify radiological factors that help predicting the best treatment for future patients

Trial Locations

Locations (1)

UZ Leuven; 🇧🇪 Leuven, Vlaams-Brabant, Belgium