Everyday AI Podcast – EP 427: The Role of GenAI in Modern Healthcare - Challenges and Opportunities

Date: December 20, 2024
Host: Jordan Wilson
Guest: William Horton, Staff Machine Learning Engineer at Included Health

Episode Overview

This episode explores the intersection of generative AI (GenAI) and modern healthcare, focusing on both the challenges and opportunities that AI presents to the industry—particularly in the US. Host Jordan Wilson and guest William Horton discuss regulatory hurdles, privacy concerns, integration issues, the physician burnout crisis, and the promise of AI-powered personalized virtual care. The conversation balances optimism about empowering patients with caution around data privacy and the need for validation.

Key Discussion Points & Insights

1. The Current State of AI in Healthcare

[08:57]

• Adoption Lag: Healthcare in the US is about a decade behind other sectors in AI adoption due to privacy concerns and regulatory challenges like HIPAA.
• Progress Areas:
  • Back Office/Admin: Most progress seen in areas such as medical scribing, administrative automation, and streamlining operations.
  • Patient Care: Early forays into patient-facing uses (chatbots, smarter WebMD alternatives), with research ongoing for diagnostic support.
• Quote:

  "GenAI has made a lot of progress in certain areas, I think mostly in kind of back office or administrative tasks." — William Horton [08:57]

2. Privacy Concerns & Data Security

[11:05]

• HIPAA & BAAs: Strict data protection laws; organizations require a Business Associates Agreement (BAA) to use most cloud-based AI.

• Cloud Providers Catching Up: Major clouds (AWS, Google, Microsoft, OpenAI) now offer BAAs, which eases integration for compliant projects.

• Quote:

  "The good thing for the US consumer is that the government has strong protections for your data. And that comes in the form of HIPAA..." — William Horton [11:05]

3. Opportunities for GenAI in Healthcare

[13:23]

• Physician Burnout: AI can reduce administrative burden, freeing doctors for more patient interaction—a critical need amid physician shortages.

• Integration Challenges:

  • Complex Software Ecosystems: Many incompatible systems (Salesforce Health Cloud, Athena, internal apps).
  • Data Interoperability: Standards like FHIR help, but inconsistently adopted; matching patient identities across systems remains tough.

• Quote:

  "If we could free up more of their time... the doctor can be face to face working with you instead of behind a screen." — William Horton [13:23]

4. Physician and Nursing Shortages

[18:04]

• Reality of Shortages: Genuine and increasing, especially acute in "health deserts."

• Role for AI:

  • Virtual Triage: Smart bots could guide patients on when to seek emergency or routine care, optimizing doctors' time.
  • Telemedicine: Expanding service reach to underserved locations.

• Quote:

  "Physician shortage problem is real and we definitely should be concerned about it... I do think AI has a role to play moving forward." — William Horton [18:04]

5. Patient Trust & Acceptance of AI

[19:21], [21:53]

• AI Outperforms in Some Diagnostics: Studies show LLMs (e.g., ChatGPT) occasionally outperform as diagnostic tools.

• Psychological Barriers: Resistance remains due to need for human warmth and empathy, similar to hesitancy around self-driving cars.

• Hybrid Approach Favored: Near-term future is likely to combine AI-driven insights with human doctors for interpretation and patient interaction.

• Quote:

  "Part of what a doctor provides is warmth... I think in the immediate future, say the next couple years it's probably going to be some kind of hybrid system." — William Horton [21:53]

6. The Next Frontier: Multimodal and Personalized Care

[24:39]

• Multimodal AI: Future models will interpret not just text but X-rays, videos (e.g., gait analysis for physical therapy), and patient-generated health data.

• Self-Collected Data: Patients increasingly use wearables and at-home tests; interest in actionable insights from these datasets is rising.

• Quote:

  "What if I could take my Fitbit data and like blood testing data and get the AI to interpret that and explain it to me in layman's terms?" — William Horton [24:39]

7. DIY AI Diagnoses: Promise & Pitfalls

[26:55]

• Risks: Consumer use of ChatGPT for personal diagnosis is rising but lacks validation; outcomes can range from helpful to potentially dangerous.

• Need for Oversight: Opportunity exists for regulated platforms to offer LLM-based advice validated by clinicians, upholding privacy standards.

• Quote:

  "I've gone to ChatGPT, put in symptoms or put in the results from a test... but I think the dangers are like it's not really validated to do that, right?" — William Horton [26:55]

8. Building Patient Trust & Consent

[29:26]

• Transparency Essential: Users must consent to AI involvement and recordings.

• Trust as Currency: Patients need a clear benefit in exchange for sharing sensitive data; healthcare orgs must prove privacy can be upheld.

• Quote:

  "The first thing is definitely putting the power in the patient's hands and getting explicit consent for some of these things... And then part of it, I think, is just building trust with your users." — William Horton [29:26]

9. The Future of “All-In” AI Healthcare

[31:32]

• Continuous, Integrated Care: Rise of always-on, data-rich, AI-enhanced healthcare ecosystems, combining medical records, wearables, and rapid AI consultations.

• Escalation Pathways: AI provides primary triage and routine answers; humans handle complex or sensitive situations—full loop integration is the goal.

• Quote:

  "There's a healthcare company that can actually integrate both traditional medical data versus this data that people are starting to collect on their own... something like that is coming in the future, if it's not already here." — William Horton [31:32]

10. Final Takeaway: Power to the Patient

[33:16]

• Optimism Tempered by Caution: AI can and should empower patients, provided ethical guardrails and validation are maintained.

• Quote:

  "Ultimately we want to use AI to give more power to the patient. And that's kind of my mission." — William Horton [33:16]

Notable Quotes & Memorable Moments

• “[GenAI] has made a lot of progress in certain areas, I think mostly in kind of back office or administrative tasks.” — William Horton [08:57]
• "The good thing for the US consumer is that the government has strong protections for your data. And that comes in the form of HIPAA..." — William Horton [11:05]
• "If we could free up more of their time... the doctor can be face to face working with you instead of behind a screen." — William Horton [13:23]
• "The physician shortage problem is real and we definitely should be concerned about it... I do think AI has a role to play moving forward." — William Horton [18:04]
• "Part of what a doctor provides is warmth... it's probably going to be some kind of hybrid system." — William Horton [21:53]
• "What if I could take my Fitbit data and like blood testing data and get the AI to interpret that and explain it to me in layman's terms?" — William Horton [24:39]
• “I've gone to ChatGPT, put in symptoms... but the dangers are like it's not really validated to do that, right?” — William Horton [26:55]
• “The first thing is definitely... getting explicit consent... And then part of it... is just building trust with your users.” — William Horton [29:26]
• "There's a healthcare company that can actually integrate both traditional medical data versus this data that people are starting to collect on their own..." — William Horton [31:32]
• “Ultimately we want to use AI to give more power to the patient. And that's kind of my mission.” — William Horton [33:16]

Important Timestamps

• [06:27] — William Horton introduction and Included Health’s mission
• [08:57] — Healthcare AI adoption snapshot
• [11:05] — Deep dive into privacy, HIPAA, and cloud AI integration
• [13:23] — Physician burnout and GenAI’s role in reducing admin burdens
• [16:28] — Engineering and interoperability challenges in healthcare AI
• [18:04] — Doctor and nurse shortages; telemedicine and virtual triage
• [19:21], [21:53] — Trust, patient acceptance, and psychological barriers
• [24:39] — Multimodal AI and the future of personalized, data-driven care
• [26:55] — DIY AI diagnosis: benefits, risks, and the future role of validated apps
• [29:26] — Patient consent and data privacy in AI-facilitated healthcare
• [31:32] — Vision for all-in, always-on, integrated AI healthcare
• [33:16] — Key takeaway: AI as a tool for empowering patients

Summary

This episode offers a concise yet thorough dive into how generative AI is gradually reshaping US healthcare. While regulatory and integration hurdles slow wide adoption, the future holds immense promise—especially in combating physician shortages, streamlining admin tasks, and moving towards personalized, always-on healthcare powered by patient data and multimodal AI. William Horton emphasizes both the optimistic vision of patient-empowering AI and the ongoing need for validation, consent, and trust.

Listeners walk away with a grounded understanding of where GenAI is today in healthcare, the systems holding it back, and the actionable opportunities for patients and practitioners as the technology and regulatory frameworks improve.