The Importance of Handoff Communication in Healthcare
At the end of each hospital shift, the outgoing nurse has the crucial task of bringing the incoming nurse up to speed about the patients under their care. This handoff process is critical for ensuring patient safety and continuity of care. However, it can be prone to errors and information gaps, putting patients at risk.
Recognizing this, healthcare provider HCA Healthcare saw an opportunity to improve handoff communication using generative artificial intelligence (AI). Large language models, such as OpenAI’s GPT-4, have shown promise in summarizing and organizing data, making them ideal for this application.
Partnering with Google Cloud’s Vertex AI
HCA Healthcare turned to Google Cloud’s Vertex AI suite to build and deploy their AI model for handoff communication. While Google offers its own foundation model called PaLM, the platform is model agnostic, allowing customers to use any model of their choice, including OpenAI’s GPT-4.
In a bid to cater to the healthcare industry specifically, Google is developing a healthcare-focused large language model called Med-PaLM 2. HCA Healthcare, along with other healthcare organizations, have had early access to this model.
The Competitive Landscape in Healthcare AI
Google’s venture into healthcare AI comes as Microsoft and Amazon also make their own AI-powered advancements in the sector. Each company is vying for dominance in the healthcare AI economy, offering their own solutions and partnering with other companies to provide choice to customers.
While the competition is fierce, none of the cloud providers have fully articulated a compelling vertical story for healthcare AI. The industry’s complexity and regulations pose challenges, and customers are looking for complete solutions rather than building blocks.
The Pilot Phase and the Role of Composite AI
HCA Healthcare is taking a cautious approach to AI adoption in healthcare. They are currently piloting the AI handoff tool at UCF Lake Nona hospital to evaluate its efficacy. The tool processes patient data from the past 12 hours and generates a transfer summary, including recommendations for the incoming nurse.
HCA Healthcare recognizes the value of using multiple models to solve complex problems, also known as composite AI. They are testing both generic models like PaLM and medically-trained models to determine which works best in different scenarios.
Addressing Challenges and Ensuring Transparency
Large language models like GPT-4 raise concerns about their performance degradation over time and lack of transparency. To address these concerns, Google is experimenting with niche models trained on specific sets of data, such as healthcare data. This allows for domain adaptation and rigorous quality evaluation.
Another challenge is keeping AI models up to date with recent knowledge in healthcare. Google is working on determining a cutoff date for training data and ensuring that users are aware of the information’s currency.
The Journey Towards AI Empowered Healthcare
Despite the challenges, HCA Healthcare remains optimistic about the potential of AI in healthcare. They see AI as a tool that can alleviate administrative burdens and improve the daily lives of clinicians. By slowly introducing AI and gaining clinician acceptance, HCA Healthcare believes they can guide the appropriate and ethical use of AI in more complex healthcare applications.
