Explore example use cases for AI tools and systems in NSW Health, aligned with the AI Framework.
The following use cases illustrate practical applications of AI across clinical, operational and administrative settings. Each highlights key considerations, including risk, governance requirements and potential benefits for consumers, the workforce, services and the system.
Key considerations
- These use cases are provided for illustrative purposes only. They do not indicate that a use case or AI tool is approved for use within NSW Health.
- The information is presented at a high level and may not reflect all relevant risks, dependencies or implementation requirements.
- Any AI initiative within NSW Health must have the appropriate oversight, be registered where required, and may need completion of the AI Assessment Framework. Explore the AI landscape for guidance.
- AI tools and systems must be sourced from an approved supplier and be assessed through the eHealth NSW Privacy and Security Assessment Framework, including a Privacy Impact Assessment, where required.
- Higher-risk use cases, particularly those that are clinical or patient-facing, may require additional governance. Seek advice from the eHealth AI Advisory Service.
- Some risks identified in NSW guidance, such as human oversight, privacy, security and fraud, are not unique to AI. However, AI may amplify these risks or introduce them in new ways.
Example 1: Generative AI for drafting reports
A public hospital is proposing to deploy an approved generative AI system for the purpose of drafting reports and correspondence. Users will be required to anonymise information, by removing all personal, confidential or potentially sensitive information from all AI prompt materials.
How this aligns with NSW guidance
- As an approved system, this solution would already be assessed against the eHealth NSW Privacy and Security Assessment Framework. Closed enterprise AI systems are preferable for privacy and security reasons.
- All personal, sensitive and confidential information must be removed as a privacy and security protection measure.
- There is a risk for misinformation and hallucinations when using AI tools. All outputs need to be checked and verified by a human user for accuracy. Users must also disclose the use of AI tools for transparency.
Potential benefits
Consumer and community
- More timely correspondence and work deliverables, enabling operational efficiencies.
- Enhanced clarity and consistency in written communication, which may support consumer understanding if implicated.
Workforce
- Reduced administration time, leading to increased consumer contact time.
- Reduced administrative workload, leading to improved personnel wellbeing.
- Enhanced document quality and consistency across teams.
System
- Time and cost saving benefits, enabling staff to focus on priority tasks.
- Streamlined administrative processes leads to improved correspondence turnaround times and service efficiency.
- Improved data quality supports downstream reporting.
Example 2: AI-enabled ultrasound screening
The radiology department of a public hospital would like to use an ultrasound scanner that incorporates an AI system, which performs initial abnormality detection. The medical device is registered as a 'software as a medical device' on the Australian Register of Therapeutic Goods.
How this aligns with NSW guidance
- Confirm the ARTG registration is accurate and use the product in line with this approval. This will help to mitigate any harms associated with 'off label' use and the use of inappropriate non-medical tools for medical purposes.
- The scanner must be sourced from an approved supplier and assessed against the eHealth NSW Privacy and Security Assessment Framework.
- Gain informed patient consent to use this clinical tool. Any outputs of the AI system must have clinician oversight and approval to mitigate risk of inaccurate or bias information to inform clinical decision-making and negative health outcomes.
Potential benefits
Consumer and community
- Potential support for earlier identification of abnormalities.
- Improved patient confidence where AI is used transparently.
Workforce
- Reduced turnaround time for results.
- Enhanced initial abnormality diagnoses, enabling clinicians to deliver focused quality services.
- An additional layer of clinical review with clinician oversight.
System
- Automated initial image analysis leads to reduced costs of manual detection, enabling faster triage of priority cases and reduced wait times.
- Increased diagnostic sensitivity leading to enhanced safety outcomes.
- Reduced service delays if the AI tool is used and governed responsibly.
Example 3: Using AI to support invoicing
The finance department of a public hospital would like to use an AI-embedded invoice software system to support enhanced service payments.
How this aligns with NSW guidance
- The system must be sourced from an approved supplier and assessed against the eHealth NSW Privacy and Security Assessment Framework.
- Although this system is not clinical facing, it will still hold sensitive information, such as patient information. There is a risk of fraud and cyber-attack with financial systems. There may be a need for human approval or override mechanism to further mitigate risks of fraud, incorrect payments and malicious intent.
- The system should integrate with existing IT infrastructure to reduce manual handling and further risk of error.
Potential benefits
Consumer and community
- Increased accuracy and efficiency, leading to enhanced financial accounting and community confidence in financial stewardship.
- Reduced risk of billing or payment errors impacting patients or providers.
Workforce
- Reduced administrative burden and errors, with improved accuracy in processing invoices and stronger financial accounting.
- Increased focus on high-priority tasks by replacing more administrative tasks.
System
- Improved processing times and accuracy, strengthening financial accounting and overall service efficiency.
- Strengthening auditability and potential fraud risk through detection and flagging systems.