36 - Using OpenAI API to Simulate Clinical Outcomes in Interventional Radiology: A Feasibility Study

Purpose: Predicting clinical outcomes in interventional radiology (IR) is critical for optimizing patient care, improving procedural success, and reducing complications. Traditional clinical decision-making relies heavily on the experience of practitioners and retrospective studies. However, with advancements in artificial intelligence, tools like the OpenAI Application Programming Interface (API) can potentially simulate patient-specific clinical outcomes, helping clinicians make informed decisions in real-time. To explore the feasibility of using the OpenAI API to simulate and predict clinical outcomes for IR procedures, focusing on procedural success rates, patient-specific risk factors, and post-procedural complications.

Materials and Methods: The OpenAI API was employed to simulate clinical outcomes for various interventional radiology procedures, including angioplasty, venous access, and tumor embolization. Clinical variables such as patient demographics, comorbidities, imaging findings, and procedural details were input into the API to generate outcome predictions. The API produced simulations regarding procedural success, complication likelihood (e.g., post-procedural bleeding, stent migration), and personalized treatment recommendations. These AI-generated simulations were compared against actual clinical outcomes from patient data available in public clinical datasets and documented literature.

Results: The OpenAI API accurately predicted clinical outcomes in 86% of cases when compared to real-world patient data. For procedural success, the API achieved an accuracy rate of 89%, closely aligning with documented clinical results. Predictive modeling for complications, such as post-procedural bleeding and reintervention, was 80% accurate, effectively identifying high-risk patients based on input variables. The API provided clinically relevant recommendations, including suggestions for alternative procedural approaches and adjustments based on patient-specific risks.

Conclusion: The OpenAI API offers a promising tool for simulating clinical outcomes in interventional radiology, enabling real-time decision support based on patient-specific variables. By predicting procedural success and complication risks with high accuracy, the API could assist clinicians in optimizing patient management and improving procedural planning. Further development and integration of AI-based tools in clinical workflows could enhance personalized care in IR and potentially reduce adverse outcomes.