🚨 AI Driven Decision Making 🚨

Practical Application of the tool

Data to Decision: AI Workflows for Accurate Triage, Referral, and Diagnosis

Introduction

Clinical decision making is a complex task. There is huge void present in our health system which needs to be addressed urgently. A huge confusion is present among the patients in selecting the specialist for their disease while a proper record keeping and tracking the patients and their disease progress has been a cumbersome task to clinicians. Modern advanced Large Language Models (LLMs) offer the potential to bridge this gap by supporting decision making for both patients and healthcare providers which significantly aids symptoms assessments (i.e. Triage and referral) as well as increase the diagnostic accuracy (diagnosis).

Methods

Multiple versions of LLMs were evaluated by researchers which includes models enhanced with Retrieval Augmented Generation (RAG). A curated dataset of 2000 real world medical cases from Medical Information Mart for Intensive Care database was used. The LLMs were tested for their ability to help patients correctly and accurately choose their desired specialist for cure, assist in urgent care system (triage) and assist clinicians in predicting the possible diagnosis.

Findings:

• Aligning with the standard protocol and guidelines, LLMs performed well in identifying appropriate diagnosis, suggesting referrals, and during triage in emergency care.

• RAG integrated models provides a higher accurate and context aware responses.

• Used properly the LLMs showed a potential to reduce the errors related to triage- under/ over which significantly strengthen the emergency care and cure systems.

Results:

LLMs showed higher degree of agreement maintaining the clinical standards among wide range of clinical contexts. A more personalized and faster care decisions making was possible with use of LLMs.

Conclusion:

LLMs integrating with RAG can smoothly support the clinical decision making which eventually improved diagnosis, triage and referral workflow but more of the validation from the governing body is needed prior full scale deployment in the real world clinical setup to minimize and eventually nullify the errors.