Neuro-Symbolic AI Assesses Sepsis Treatment Compliance

Himanshu Tripathi, Kaushik Roy, Subash Neupane, Shahram Rahimi· August 17, 2026 View original

Key takeaways

  • A neuro-symbolic AI pipeline can assess clinical protocol compliance.
  • LLMs normalize clinical text, while fuzzy logic reasons over rules.
  • Sepsis treatment shows critical compliance gaps, especially in antibiotic timing.
  • This approach provides graded compliance scores, not just binary judgments.

Who benefits

HealthcareMedical TechnologyPharmaceuticalsInsurance

Summary

This paper introduces an expert-guided neuro-symbolic pipeline that combines a large language model for semantic normalization with a fuzzy inference system to evaluate compliance with sepsis treatment protocols. Applied to MIMIC-IV data, it identifies antibiotic timing and Hour-1 bundle underperformance as critical areas for improvement.

Ensuring clinical care adheres to evidence-based protocols is a complex challenge, particularly in critical conditions like sepsis. This research presents a novel neuro-symbolic pipeline designed to assess compliance with the Surviving Sepsis Campaign bundle rules. The system leverages a large language model (LLM) to perform semantic normalization, accurately mapping varied clinical text (like drug names) to a standardized vocabulary. Following normalization, a Sugeno fuzzy inference system takes over, reasoning about the events and assigning graded compliance scores between 0 and 1, rather than simple binary judgments. When applied to a large dataset of sepsis episodes from MIMIC-IV, the pipeline revealed significant compliance gaps, particularly in antibiotic timing (only 13% within one hour) and overall Hour-1 bundle adherence. It also highlighted differences in ICU stay durations based on compliance levels.

Why it matters

Healthcare professionals and administrators can use this AI-driven approach to systematically identify and address critical gaps in clinical protocol adherence, potentially improving patient outcomes and optimizing resource allocation in high-stakes medical scenarios.

How to implement this in your domain

  1. 1Explore integrating neuro-symbolic AI pipelines for automated compliance monitoring in clinical settings.
  2. 2Collaborate with medical experts to define and encode evidence-based clinical protocols into fuzzy inference systems.
  3. 3Utilize LLMs for semantic normalization of unstructured clinical data, ensuring consistency for downstream analysis.
  4. 4Focus on identified areas of low compliance, such as antibiotic timing in sepsis, for targeted quality improvement initiatives.

Original post by Himanshu Tripathi, Kaushik Roy, Subash Neupane, Shahram Rahimi

"arXiv:2608.13617v1 Announce Type: new Abstract: Verifying whether clinical care follows evidence-based protocols is a natural neuro-symbolic problem, yet the safety-critical setting defeats either paradigm alone. We present an expert-guided pipeline that constrains a large langua…"

View on X

Originally posted by Himanshu Tripathi, Kaushik Roy, Subash Neupane, Shahram Rahimi on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses