Artificial Intelligence, Clinical Decision Support Systems, Deep Vein Thrombosis, Pulmonary Thromboembolisms, Superficial Thrombophlebitis
Conditions
Keywords
anticoagulant, algorithm, thrombosis
Brief summary
Study using a decision algorithm for the application of an oral anticoagulant calculator in vascular diseases, aimed at validating a clinical decision-support tool for conditions such as deep vein thrombosis, superficial thrombophlebitis, and pulmonary thromboembolism.
Detailed description
Cross-sectional, three-arm comparative validation study evaluating the accuracy and clinical utility of the DOACT algorithm versus standard clinical decision-making and large language model (LLM)-based decision tools.
Interventions
OTHERDOACT algorithm
Vascular and non-vascular physicians using DOACT (Dose-Oriented Anticoagulant Calculator for Evidence-Based Decision Tool) to recommend appropriate oral anticoagulant regimens-dose selection and duration responding 15 standardized clinical case vignettes representing patients with vascular diseases such as deep vein thrombosis (DVT), superficial thrombophlebitis, and pulmonary thromboembolism (PTE).
OTHERNo algorithm
Vascular and non-vascular physicians using standard clinical decision-making (no use of algorithm) to recommend appropriate oral anticoagulant regimens-dose selection and duration responding 15 standardized clinical case vignettes representing patients with vascular diseases such as deep vein thrombosis (DVT), superficial thrombophlebitis, and pulmonary thromboembolism (PTE).
OTHERLLM-based tools
Vascular and non-vascular physicians using large language model (LLM)-based tools to recommend appropriate oral anticoagulant regimens-dose selection and duration responding 15 standardized clinical case vignettes representing patients with vascular diseases such as deep vein thrombosis (DVT), superficial thrombophlebitis, and pulmonary thromboembolism (PTE).
Sponsors
ITALO EUGENIO SOUZA GADELHA DE ABREU
Study design
Allocation
RANDOMIZED
Intervention model
PARALLEL
Primary purpose
SUPPORTIVE_CARE
Masking
SINGLE (Investigator)
Masking description
Single-blind. Investigator was blinded to the intervention assignment.
Intervention model description
Three-arm comparative validation study evaluating the accuracy and clinical utility of the DOACT algorithm versus standard clinical decision-making and large language model (LLM)-based decision tools.
Eligibility
Sex/Gender
ALL
Age
18 Years to 89 Years
Healthy volunteers
Yes
Inclusion criteria
* Physicians with residency training in Vascular Surgery or official Board Certification in Vascular Surgery. * Currently practicing clinical and/or surgical vascular care in Brazil. * Completed the informed consent process (TCLE) and voluntarily agreed to participate.
Exclusion criteria
* Physicians without formal Vascular Surgery residency and without Board Certification. * Physicians not performing vascular clinical or surgical care (e.g., exclusively administrative, academic, or non-assistance roles). * Less than 1 year of professional experience after medical school graduation. * Did not sign or did not fully complete the TCLE. Large Language Models (LLMs) * Inclusion Criteria * Free-access LLMs available to the public at the time of data collection. * All responses generated using the same standardized prompt. * Capable of producing complete, text-based clinical answers relevant to vascular surgery decision-making.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Accuracy of the DOACT Algorithm in Guiding Oral Anticoagulant Therapy | Day 1 | Accuracy of anticoagulation recommendations Description: Proportion of correct responses generated by the four evaluated LLMs, vascular surgeons, and non-vascular physicians, with and without access to the DOACT algorithm, using standardized clinical vignettes. |
| Accuracy of anticoagulation recommendations | Day 1 | Proportion of correct responses generated by LLMs, vascular surgeons, and non-vascular physicians with and without access to the DOACT algorithm. All LLM outputs will be generated using the same standardized prompt, following methodological guidance recommended by IBM for evaluating large language models. |
Other
| Measure | Time frame | Description |
|---|---|---|
| 1. Identification of key clinical elements 2.Response time | Day 1 | Correct reporting of dosing adjustments, renal criteria, bleeding risks, reversal agents, and contraindications. Description: Time (seconds) from prompt submission to full answer generation for LLMs, and time to completion for physicians. |
Countries
Brazil
Outcome results
None listed