Open Nursing Core FHIR Implementation Guide (ONC-IG) - Local Development build (v1.0.0) built by the FHIR (HL7® FHIR® Standard) Build Tools. See the Directory of published versions

Relational Ai (AI Model)

Relational Ai for Nursing

The first open-source LLM fine-tuned on Foundation of Nursing Studies (FONS) literature for person-centred, equitable clinical documentation.

Overview

Relational Ai for Nursing is a specialized AI model developed as part of the Open Nursing Core IG. It is designed to assist nurses in writing high-quality, person-centred clinical notes that adhere to professional standards while reducing administrative burden.

Instead of generic medical text, this model is trained to prioritize:

• Relational Care: Focusing on the patient's experience and preferences.
• Health Equity: Specifically trained to ensure inclusive documentation (e.g., correct skin tone assessment).
• FONS Principles: Aligned with the Foundation of Nursing Studies' core values.

Key Capabilities

1. Equitable Skin Tone Assessment

Standard AI models often fail to describe pressure ulcer risks accurately for patients with darker skin tones. Relation Ai has been fine-tuned to capture these nuances, achieving an 8/10 score from expert judges on equity benchmarks.

2. Person-Centred Language

The model rewrites clinical jargon into language that respects the patient's dignity.

• Before: "Patient non-compliant with medication."
• Relation Ai: "Patient prefers to take medication with food to avoid nausea; discussed strategies to support adherence."

3. FONS Alignment

Trained on 6,698 instruction pairs from the International Practice Development Journal (IPDJ), the model understands concepts like "flourishing," "authentic partnership," and "values-based practice."

Evaluation Results

The model was evaluated using a rigorous multi-judge system (GPT-4o, GPT-5, Gemini 3 Pro).

Metric	Score	Note
Clinical Accuracy	6.6/10	Solid baseline for nursing interventions
Person-Centredness	7.6/10	Strong performance in respectful language
Equity (Skin Tone)	8.0/10	Best-in-class performance

Integration with FHIR

This AI model is designed to work alongside the FHIR profiles defined in this IG.

• Input: Structured data from Patient, Observation (e.g., Skin Tone), and Condition resources.
• Output: Narrative text for Composition or ClinicalImpression resources.

The "Super-Gold" Standard: Advancing Relational Care

The Open Nursing Core project aims to build upon rigorous clinical modeling (like openEHR) by making the "Human Elements" of nursing computable and mandatory.

1. The ONC Empathy Index

We have standardized the measurement of empathy. Documentation is no longer just "data"—it is scored on its therapeutic depth (1-5) helping nurses reflect on the quality of their engagement.

2. Mandatory Equity Invariants (The Fairness Gate)

Unlike static models, our IG includes executable safety rules. A wound assessment cannot be validated unless it accounts for the patient's specific skin tone (Fitzpatrick/Monk scale), ensuring no patient is overlooked due to biased clinical thresholds.

3. Real-Time Semantic Audits

The Relational AI performs a Super-Gold Audit on every note, identifying NANDA-I diagnoses and validating the note against our Relational Care Logical Model in real-time.

Usage & License

• License: CC BY-NC 3.0 (Non-Commercial)
• Base Model: Llama-3-8B (Unsloth)
• Disclaimer: This tool is for research and educational purposes. All clinical documentation must be verified by a registered nurse.