HPCSA AI Policy & Compliance

Compliance Summary

ASCLEPIUS has been designed in alignment with the Health Professions Council of South Africa (HPCSA) Ethical Guidelines on the Use of Artificial Intelligence (Booklet 20, September 2025). This document outlines how each AI feature in ASCLEPIUS complies with HPCSA requirements.

All AI features in ASCLEPIUS serve exclusively as clinical decision support tools. No AI output is used as a final clinical decision without explicit health practitioner review and approval.

Section 3 - Ethical Principles

HPCSA Requirement	ASCLEPIUS Compliance
3.1 Patient's best interests remain primary concern	All AI features are designed to assist health practitioners in providing better patient care. Patient confidentiality, privacy, and dignity are protected through POPIA-compliant data handling.
3.3 AI serves as a tool to support, not replace clinical decision-making	Every AI feature includes a "Doctor's Assessment" section where the practitioner makes the final clinical decision. AI outputs are clearly labelled as suggestions requiring verification.
3.5 Health practitioners shall always make the final decision on patient care	All AI-generated content (fracture classifications, image analyses, discharge summaries, clinical notes) requires explicit practitioner review before being finalised. No AI output is automatically applied to patient records without practitioner approval.
3.7 Express accountability for errors	All AI outputs are attributed to the reviewing health practitioner who approves them. The system maintains an audit trail of AI-generated content and practitioner modifications.
3.8 Protection of patient health information privacy	Patient data is processed within the application's secure environment. AI processing uses encrypted API connections. No patient data is stored by third-party AI providers.

Section 4 - Disclosure

HPCSA Requirement	ASCLEPIUS Compliance
4.1 AI tool must not be secret; limitations disclosed upfront	ASCLEPIUS openly discloses AI usage on every page (footer banner) and on each AI-powered feature with specific disclaimers. AI model names, capabilities and limitations are transparently documented.
4.3-4.4 Disclosure of learning type (continuous vs locked)	ASCLEPIUS uses "locked learner" AI models (Claude, Gemini) that do not automatically update with new patient data. The models are pre-trained and do not learn from individual patient interactions within the application.
4.5 Understanding training data quality, safety, and bias	AI providers' models are trained on diverse medical literature. ASCLEPIUS prompts include South African context-specific considerations (SA-prevalent conditions, local guidelines) to reduce bias in outputs.

Section 5 - Accountability

HPCSA Requirement	ASCLEPIUS Compliance
5.2 Health practitioners ultimately responsible for AI tool use	Every AI output requires practitioner review. The treating doctor's name and credentials are associated with all finalised documents. AI results cannot bypass the practitioner approval workflow.
5.3-5.4 Practitioners should not over-rely on AI	Prominent warnings on all AI features remind practitioners to exercise independent clinical judgment. AI outputs are presented as suggestions, not directives.

Section 6 - Equity & Transparency

HPCSA Requirement	ASCLEPIUS Compliance
6.1 AI must not exacerbate healthcare disparities	ASCLEPIUS provides the same AI-powered clinical support regardless of patient demographics. AI prompts include South African context to ensure relevance across diverse populations.
6.2 Patients must be informed when AI assists diagnosis/treatment	All AI-generated documents (PDFs, reports) clearly state that AI was used in the assessment process. Practitioners are reminded to inform patients per HPCSA requirements.
6.4-6.5 Informed consent is meaningful dialogue	ASCLEPIUS supports (not replaces) the informed consent process. AI outputs provide information to help practitioners have more informed discussions with patients.
6.7 Transparency about AI capabilities, limitations, safeguards	Each AI feature includes contextual information about what the AI can and cannot do. This policy page provides comprehensive transparency about all AI components.

Section 7 & 9 - Safety, Quality & Clinical Decision Making

HPCSA Requirement	ASCLEPIUS Compliance
7.1 Interventions must be in patient's interest and safe	AI features include safety checks, red flag alerts, and urgency indicators. Practitioners retain override authority on all AI suggestions.
9.1 Responsibility for decision making must remain with the practitioner	ASCLEPIUS enforces this through its workflow design: AI generates suggestions, the practitioner reviews, modifies if needed, and explicitly approves before any clinical action.
9.2 AI cannot be the final decision maker	No AI output in ASCLEPIUS is automatically applied. Every feature requires the health practitioner to actively confirm or modify AI suggestions before they become part of the clinical record.

Section 10 - Data Privacy & Protection

HPCSA Requirement	ASCLEPIUS Compliance
10.1 Patient information protected against improper disclosure	All data is encrypted in transit (TLS). Database access is restricted to authenticated, authorised practitioners. AI API calls use encrypted connections and do not store patient data.
10.2 Safeguard against unauthorized access	Role-based access control ensures practitioners only access their own patients' data. Session management, secure authentication, and POPIA compliance measures are implemented throughout.

AI Features in ASCLEPIUS

The following AI-powered features are available in ASCLEPIUS. Each operates as a clinical decision support tool with practitioner oversight:

Feature	AI Provider	Purpose	Practitioner Oversight
AI Doc Personal Assistant	Claude (Anthropic)	Clinical query assistance, medical information retrieval	Conversational tool; practitioner evaluates all responses
Fracture Classification	Claude / Gemini	AO/OTA fracture classification suggestions	Practitioner reviews classification, modifies treatment plan
Image Analysis (Ophthalmology & Dermatology)	Gemini Vision	Clinical image analysis with diagnostic suggestions	Doctor's Assessment form for final diagnosis
Discharge Summary Generator	Claude / Gemini	Automated discharge summary drafting	Full edit/review workflow before finalisation
Theatre Operation Notes	Claude / Gemini	Procedure description assistance	Practitioner edits and approves all content
Smart Clinical Notes	Claude / Gemini	AI auto-complete and note expansion	Suggestions only; practitioner accepts or rejects
Clinical Motivation Letters	Claude / Gemini	Motivation letter drafting for funders	Practitioner reviews and signs final document
Referral Letters	Claude / Gemini	Specialist referral letter generation	Practitioner reviews, edits, and approves
Voice Dictation	Speech Recognition	Clinical documentation transcription	Practitioner reviews and corrects transcription

Important Notices

AI Model Type: All AI models used in ASCLEPIUS are "locked learner" systems. They do not learn from or retain individual patient data from interactions within this application.
No Autonomous Decision-Making: No AI feature in ASCLEPIUS autonomously makes clinical decisions, prescribes treatments, or modifies patient records without explicit practitioner approval.
Patient Notification: Health practitioners are reminded of their obligation under HPCSA Booklet 20, Section 6.2 to inform patients when AI has been used to assist in their diagnosis or treatment planning.
Professional Responsibility: Per Section 9.1, the treating health practitioner retains full professional responsibility and accountability for all clinical decisions, regardless of AI involvement.
Limitations: AI systems may produce inaccurate, incomplete, or biased outputs. Practitioners must apply their clinical expertise and judgment to evaluate all AI-generated content.
Data Protection: All patient data processing complies with POPIA (Protection of Personal Information Act) and HPCSA Booklet 5 (Confidentiality) requirements.

HPCSA AI Compliance Policy