The ROI of AI, Doctors using ChatGPT, and AI Call Centers

The AI Clinic #1 | Subscribe

Welcome to The AI Clinic, your go-to source for exploring the future of healthcare through the lens of AI. In each issue, we delve into how artificial intelligence is reshaping patient care, medical decision-making, and the role of healthcare professionals as these tools become more prevalent.

Our goal is to spark thoughtful discussions on AI usage in patient care, keep you informed on what’s new, and support your ideas for using AI in your own medical practice.

Today We’re Covering:

• 76% of physicians using Conversational AI in patient care

• Google and Mayo Clinic explore the ROI of AI

• CHS using AI in their call centers

• An Iowa health system expanding their ambient AI assistant

• Open AI’s DevDay and potential healthcare implications

Survey Shows 76% of Doctors using Conversational AI in Patient Care

Fierce Healthcare released an article this week highlighting the increased use of ChatGPT and other generative AI (genAI) tools amongst physicians, despite concerns about their safety and reliability. In the absence of standardized regulations and formalized guidelines, doctors are turning to publicly available AI tools for clinical decision-making, raising questions about the trade-off between convenience and risk as AI tools are rapidly integrated into clinical workflows.

Key Takeaways:

1. Rising Adoption Among Clinicians: According to a survey conducted by Fierce Healthcare and Sermo, 76% of physicians reported using general-purpose AI models like ChatGPT for tasks such as checking drug interactions, supporting diagnosis, and patient education. However, the lack of formal vetting means physicians must independently verify outputs.

2. Patient Perception and Trust: Some patients are growing skeptical of doctors’ reliance on AI, feeling they can access similar information independently. Will AI reduce the trust patients place in their doctors if these tools are seen as replacing expertise rather than enhancing it?

3. Safety Concerns and Lack of Regulation: AI tools like ChatGPT, which were not designed for clinical use, present significant risks. The article discusses how AI-generated outputs can be incomplete or misleading, potentially leading to unsafe recommendations. For instance, AI can miss critical context, such as a patient’s pregnancy, when suggesting treatment options. Confabulation (AI generating plausible but incorrect answers) and the use of outdated or unverified sources are major issues.

4. AI in Clinical Workflows: While AI shows promise in streamlining tasks like drafting notes or brainstorming diagnoses, there is growing consensus that it is not ready for fully autonomous clinical decision-making.

5. The Need for Standards and Training: The article points out the regulatory gap in the U.S., where formal standards for AI use in healthcare are still years behind current utilization rates. Without proper training, many doctors may struggle to use these tools safely. The American Medical Association (AMA) advises caution, recommending that AI tools in healthcare be rigorously tested and proven before adoption. Physicians are ultimately responsible for the outcomes of their decisions, even when using AI, which raises concerns about legal liability and ethical considerations.

Why This Matters

This piece highlights the growing tension between AI’s potential to enhance healthcare delivery and the risks associated with its premature use. As AI adoption increases, healthcare providers are faced with a dilemma: how to integrate AI tools into clinical workflows without compromising patient safety or trust. The lack of regulatory oversight and clinical validation for tools like ChatGPT leaves both physicians and patients in a precarious position. The article underscores the pressing need for comprehensive guidelines, physician education, and robust safeguards to ensure that AI contributes meaningfully to clinical decision-making, rather than introducing new risks.

Unfortunately, any such guidelines are unlikely to appear anytime soon, so clinicians and patients are left to decide for themselves AI's role in active patient care. We are likely to see organizations address this challenge with varying degrees of effectiveness as the industry attempts to find an equilibrium that satisfies patients, providers, payers, and regulatory bodies. (Full Article)

Google, Mayo Clinic want to know the ROI of AI

Digital Medicine Society (DiMe), Google, and the Mayo Clinic are partnering together to create a ‘playbook’ to help health systems maximize the benefits of AI investments. The initiative seeks to address real-world challenges in AI implementation, providing tools such as a return on investment (ROI) calculator and guidance on forming AI strategies. This collaboration involves various stakeholders, including patient advocates and AI developers, with the goal of overcoming barriers that limit AI adoption in clinical settings.

Key points of the partnership include:

1. Big-Picture Focus: Instead of merely assessing the safety and effectiveness of AI models, the playbook hopes to help systems understand how to implement AI in ways that improve operational workflows and care delivery.

2. AI's Business Case: DiMe's CEO, Jennifer Goldsack, stresses the importance of equipping decision-makers with the right tools to build a business case for AI. Without such tools, AI's promised value may remain unrealized, potentially stalling adoption.

3. Practical Resources: The ROI calculator will allow health systems to assess whether AI investments are delivering tangible returns. This resource aims to prevent AI from falling into a 'hype cycle,' similar to what happened with digital therapeutics, where early excitement did not translate into long-term success.

4. Strategic Use of AI: Goldsack noted that health systems must identify their biggest opportunities, such as detecting health issues early to reduce emergency department visits, rather than implementing AI tools haphazardly.

Why This Matters

This initiative addresses a critical challenge facing large health systems: how to justify AI investments in practical, financial terms. With limited budgets and resources, health systems are cautious about adopting new technology that doesn’t quickly show measurable returns.

This puts most health systems in a difficult bind. The most ‘transformational’ uses of AI are likely to be those needing long incubations and are unlikely to show short-term positive ROI. However if they merely focus on the “easy-wins”, they’re unlikely to unlock the deeper value that most AI analysts claim is the ‘true’ value in integrating AI systems.

In an industry where AI adoption is often hindered by skepticism, regulatory hurdles, and fragmented implementation, resources to help organizations tackle these big questions could be pivotal. (Full Article)

CHS Announces 'Conversational AI' Powered Call Centers

Community Health Systems (CHS) and Denim Health have partnered to scale conversational AI across CHS's Patient Access Center (PAC). The PAC is a centralized call center that handles over 25,000 inbound calls daily, serving nearly 1,000 CHS-affiliated primary care providers. This partnership intends to enhance the patient call experience by integrating AI-driven tools to improve efficiency and patient satisfaction.

Key aspects of the partnership include:

1. AI-Powered Efficiency: Denim Health’s conversational AI, initially deployed to authenticate patient identities and manage call workflows, has resulted in reduced call times and smoother agent interactions. The AI bot captures preliminary information from patients before routing them to human agents, thereby streamlining operations and reducing bottlenecks.

2. Scaling Capabilities: Beyond handling authentication, the system is designed to expand into areas like patient self-scheduling, care gap closures, and handling other routine tasks. This allows human agents to focus on more complex cases, enhancing overall patient care quality.

3. Human-AI Balance: Although AI is handling more transactional tasks, patients retain the ability to speak to a live agent when needed, ensuring that more nuanced or sensitive cases are dealt with by humans.

Why This Matters

This partnership highlights a growing trend amongst health systems who have deep interest in the uses of conversational AI but are unsure of how to 'eat the elephant'. These systems don't have the flexibility or desire to rebuild their operations around these tools, so are instead focusing on using these technologies to address significant operational inefficiencies that exist across most US healthcare systems.

By offloading routine tasks like patient authentication and appointment scheduling to AI, health systems hope to differentiate their brand as being 'tech-forward' while also reducing costs and relieving operational bottlenecks.

The article underscores what we'll continue to see from other AI announcements especially at large healthcare organizations - initial AI integrations will be augmenting human roles, not replacing them. We'll see many such announcements in the next twelve months as systems seek to explore the benefits of these AI enhancements while moderating the risk to existing operations. (Full Article)

Iowa Health System Expands AI Assistant Usage

University of Iowa Health Care (UI Health Care) has rolled out Nabla's 'ambient AI assistant' to all clinicians following a successful pilot program. The AI tool aims to reduce clinician burnout by streamlining clinical documentation processes, allowing physicians to focus more on patient care. During the five-week pilot, the use of Nabla led to a 26% reduction in clinician burnout, underscoring its potential to improve both efficiency and provider well-being.

Key elements of the initiative include:

1. Addressing Physician Burnout: A key driver of this rollout is the significant administrative burden clinicians face, particularly in managing electronic health records (EHRs). Studies suggest that for every hour spent with patients, physicians spend almost two additional hours on EHR and desk work. The announcement points to Nabla's AI-generated directly addressing this imbalance, helping to alleviate stress and improve work-life balance for physicians.

2. Pilot Success and Adoption: UI Health Care’s pilot program saw 100% adoption of Nabla by participating clinicians, with a high engagement score of 75% week-over-week. Clinicians rated the AI-generated clinical notes highly (4.3 out of 5), which was a major factor in the decision to implement the tool across the system.

3. EHR Integration: Nabla integrates directly with Epic, UI Health Care’s EHR system, allowing clinicians to review, edit, and finalize AI-generated notes with a single click. This smooth integration is crucial for seamless adoption and reducing time spent on manual tasks like copying and pasting.

Why This Matters

The UI Health Care-Nabla partnership is emblematic of a larger trend in healthcare: the outsourcing of routine tasks to AI. And while the healthcare industry desperately needs solutions to tackle clinician burnout and operational inefficiencies, it’s essential to recognize that AI cannot be a one-size-fits-all solution. Tools like Nabla may significantly reduce documentation time, but they must fit into a larger strategic shift that addresses workflow redesign, clinician engagement, and holistic care delivery.

As AI becomes more prevalent, the question will be whether adopting these cutting-edge tools leads to larger rethinking on the structural issues that make such tools necessary. If AI becomes a crutch for inefficient systems, we may miss the opportunity to reimagine healthcare delivery in a way that truly benefits both clinicians and patients. (Full Article)

Open AI DevDay

OpenAI, developer of ChatGPT, hosted their 2024 DevDay on October 1st. The company unveiled updates on the direction of tools that will be available to developers that utilize the OpenAI platform for building AI tools.

As healthcare organizations increasingly explore AI-driven solutions to enhance patient care and streamline operations, these new capabilities offer a glimpse into how AI might reshape everything from diagnostics to telemedicine.

The four core announcements were:

1. Realtime API: This API supports low-latency speech-to-speech experiences, making it possible to create natural, real-time conversational AI systems. This could be valuable for virtual telephone operators, virtual triage services, scribing, or other services. assistants or telemedicine platforms, enabling smoother, more natural interactions between patients and AI-driven care providers.

2. Vision Fine-Tuning: OpenAI now allows developers to fine-tune models using both text and images, which could significantly improve medical image analysis. This could enhance diagnostics and object detection in medical imaging, allowing for more accurate and faster identification of conditions through AI-driven tools.

3. Prompt Caching: By reducing costs and latency for repeated tasks, prompt caching is particularly relevant for healthcare applications that require continuous monitoring or regular follow-up queries, such as patient monitoring systems or decision support tools for clinicians. This feature reduces the time and cost of repetitive requests, which is especially important in data-heavy environments like hospitals.

4. Model Distillation: This allows healthcare developers to fine-tune smaller, cost-efficient models while maintaining high performance. For healthcare organizations aiming to build specialized models—such as for niche medical conditions or specific patient demographics—this could enable faster development cycles and more affordable implementation of AI solutions.

Why This Matters

As healthcare systems continue to face rising operational costs and the need for more streamlined care delivery, these advancements allow developers to create highly specialized, affordable AI solutions. Each iteration of technological functionality nudges AI tools one step closer to seamless integration into everyday healthcare activities — addressing the industry's need for scalable virtual care options.

OpenAI's (and their competitors’) focus on cost reduction and increased customization will continue to lower the barriers for developers to bring new tools to market and hopefully make basic AI tools more affordable for smaller medical practices who are interested in AI but don't have the budgets of large health systems.

If we consider the sorts of technological advancements smartphones have been through in the last ten years, it's not a stretch to imagine the current AI technological capabilities only just beginning to make a dent in what's possible as more of these tools are brought to market. (More Here)

Other Things Worth Checking Out

Here are some other developments that might be worth your time.

How AI could tame the charting madness for nurses

2024 Nobel Prize winners: Physics award goes to Hopfield, Hinton for machine learning discoveries

A 'multibillion-dollar bottleneck': 3 top AI issues for health systems

The number of AI medical devices has spiked in the past decade

ChatGPT overprescribed emergency treatment: UCSF study

Healthcare startup Suki raises $70 million to build AI assistants for hospitals