How AI is changing the way doctors treat their patients | 7.30

Key Concepts

• AI scribes in healthcare
• Differential diagnosis
• Patient privacy and data security
• AI-generated misinformation, particularly regarding vaccines
• AI in cancer research (AAnet)
• Tumor heterogeneity and hypoxic niches

AI Scribes in Clinical Practice

• Introduction: AI is being implemented in healthcare to streamline workflows and improve patient care, but also presents challenges related to misinformation and privacy.
• AI Scribe Demonstration: Dr. Grant Blashki, a GP in Melbourne, demonstrates an AI scribe during a mock appointment. The AI scribe records the consultation and generates notes.
• Functionality: The AI scribe offers differential diagnoses based on the patient's symptoms. In the example given, the AI suggests "tension type headache" and "cervicogenic headache" for a patient complaining of morning headaches.
• Heidi Health: Dr. Blashki uses Heidi Health AI. He emphasizes that doctors should view the AI's suggestions as guidance, not definitive answers.
• Clinician Perspective: The AI scribe aims to reduce administrative burden and free up clinicians' time.
• Heidi Health's Statement: Heidi Health CEO Dr. Thomas Kelly states that the software summarizes clinical encounters using appropriate terminology but does not provide a differential diagnosis absent the clinician. The clinician is responsible for reviewing the documentation for accuracy.

Patient Privacy and Data Security

• Data Deletion: Kai Vanlashout, CEO of Lieird Health, another AI scribe company, states that patient notes are automatically deleted after 7 days.
• Encryption: Patient data is encrypted to the same level as highly sensitive bank data, rendering it "completely indiscernible" if accessed without authorization.
• Regional Compliance: Heidi Health states that its software complies with healthcare regulations and privacy policies in Australia, New Zealand, Canada, US, and the UK, storing data within each region.
• Inherent Risks: John Lawler notes that there is always some risk when dealing with digital data.

AI-Generated Misinformation

• Data-Driven Models: AI models improve with more data, but this also raises concerns about the potential for misuse.
• Bot Activity: AI bots can generate and post content on social media, making it difficult to distinguish them from human users.
• Vaccine Misinformation Example: Researchers at the Murdoch Children's Research Institute found AI-generated comments on Reddit spreading misinformation about vaccines. Examples include claims about high vaccine injury rates in the US and questioning the safety of multi-dose vaccines.
• Reddit's Response: Reddit stated that the thread was an openly labeled bot project and that deceptive bots are against their policies.
• Real-World Impact: Brett Sutton, former chief health officer of Victoria, says vaccine misinformation online is contributing to vaccine hesitancy and declining vaccine uptake. He points to measles outbreaks in the US as an example of the potential harms.
• Call to Action: Sutton advocates for social media companies to actively identify and delete misleading medical information, emphasizing that it poses a threat to science, democracy, national security, and trust in institutions.

AI in Cancer Research

• AAnet Tool: Associate Professor Christine Schaffer and Associate Professor Smitta Krishna Swami are using an AI tool called AAnet to analyze tumor structure.
• Tumor Heterogeneity: AAnet helps researchers understand the heterogeneity of cancer cells within a tumor, identifying different groups of cells with varying characteristics.
• Process: Christine collects tumor samples in Sydney and sends them to Smitta at Yale University, who processes them through AAnet.
• Color-Coded Imaging: AAnet generates color-coded images of tumor samples, highlighting different cell populations. Orange cells are rapidly proliferating, while blue cells indicate a hypoxic niche where cancer cells may be hiding.
• Eradication Strategies: The goal is to use this information to develop strategies to eradicate all cancer cells within a primary tumor, preventing recurrence.
• Hypoxic Niche: The discovery of hypoxic niches is significant because these areas are often resistant to treatment, allowing cancer cells to survive and potentially lead to relapse.

Synthesis/Conclusion

AI offers significant potential benefits in healthcare, from streamlining clinical workflows with AI scribes to advancing cancer research through tools like AAnet. However, the technology also presents challenges, including the need to protect patient privacy and combat the spread of AI-generated misinformation. Responsible implementation, robust data security measures, and proactive efforts to address misinformation are crucial to harnessing the benefits of AI while mitigating its risks. The future of healthcare will likely involve increased integration of AI, requiring clinicians and the public to understand both its capabilities and limitations.