AI has the potential to impact every industry by expanding the capabilities of IT vendors to develop novel solutions, and the ability of users to better leverage their data.
Healthcare is buzzing with excitement about the possibilities of AI, but at the same time, there is regulatory uncertainty around how best to oversee its use while optimizing its potential.
The need for safe application of AI is universal – it’s a standard that individuals and organizations in every sector should expect.
Healthcare is in the early stages of understanding AI and how it can function in the medical field. Having a checklist to evaluate your organization’s readiness will be essential for ensuring an effective introduction of AI into workflows, and maximizing its benefits.
MEDITECH has been exploring the use of AI in our intelligent EHR platform, Expanse, for several years now. While we can’t say that we have all the answers yet, here are five steps your organization should take before embarking on your AI journey.
- Establish your AI strategy and ethical guidelines
Many organizations have not developed an AI strategy yet. A strategy should be built upon your existing mission and vision, and should outline both the potential benefits and risks you see from AI (along with the resources you have to capitalize on the former and mitigate the latter).
Establish an AI governance board to monitor the regulatory environment and evaluate technology safety, security, and ethical matters. There are many resources that can help you develop responsible guidelines and accompanying policies. A board dedicated to AI oversight will strengthen your ability to hone in on the most meaningful way to integrate AI solutions into your organization.
Once you’ve established these internal guidelines and principles, evaluate how the priorities of technology vendors align with your own. Your technology vendor should be able to clearly articulate its position on AI, and show how that philosophy is expressed in its solutions. At MEDITECH, we established and shared our philosophy early on.
- Understand the limits and potential of AI
As you begin your AI journey, understanding AI’s vocabulary will be key to framing expectations. A challenge with Generative AI is that it can produce “hallucinations” – realistic output that is not based in factual reality. Another challenge is matching the right model to the use case, and ensuring consistent control of the output, depending on the context.
Large Language Models (LLMs) are built on probability, which means that output can vary based on how the model is prompted to respond and all the possible word choices and sequences that can be used when generating a response. For creative purposes, you may want responses with a lot of variability – meaning the model has some leeway in choosing the next most likely word in a sequence.
In healthcare, factual accuracy and reliable output are paramount. There are several steps developers and end users can take to address these limitations. Prompt engineering involves forming the explicit questions and instructions necessary to elicit comprehensive, context-specific, and accurate responses. This will become a critical skill set for organizations. In addition to determining the best prompt for output type and format, adjusting something called the model’s temperature provides better user control. Setting the model’s temperature to zero ensures the model provides the most likely (highest probability) response given the prompt and input data. This means results are more consistent and accurate.
Beyond that, organizations still need to maintain accountability through development of policies that support “human-in-the-loop” (HITL) review. Especially during early stages of Generative AI development, it is important to remember that Generative AI is a tool, not a solution in and of itself.
- Match AI advantages to specific workflows
At the end of the day, technology has to work for the people using it. In healthcare, that means both healthcare professionals and the patients they serve. There are a myriad of opportunities to build solutions – the most dynamic uses will not just automate but optimize workflows.
At MEDITECH, we’re looking at workflow challenges that impact total experience, whether directly or indirectly. Challenges like more efficient clinical documentation, better navigation of medical records, and improved provider handoffs. We’re finding ways to elevate relevant data and enhance care team communication of necessary information, increasing the quality of care and patient-care team interactions. To accomplish this, we work with the people using the software – doctors, nurses, specialists, and other staff – so that solutions are not only designed for them, but with them.
- Establish effective data governance
In order to make the most meaningful use of data, everyone in the organization needs to have a similar baseline understanding of what data is and how it can be leveraged – this is data literacy. Even if an individual isn’t responsible for the direct analysis of information, there needs to be an understanding of how data flows throughout the organization, and how their role ultimately supports the collection, curation, or utilization of data. Palo Pinto General Hospital has done this effectively by using a culture-driven approach to the use of data.
Increasing your organization’s data literacy will also allow you to more effectively prioritize AI use cases that fit your health system’s needs, and have an appropriate framework to measure progress towards meeting your goals. This process takes a high level of focus and discipline, which your organization will need when considering and deploying AI.
The other aspect of data governance is a commitment to appropriate data access and use. Healthcare organizations are already familiar with data privacy requirements, but new uses of data reinforce the need to reevaluate your policies and partnerships. This is where transparency is key.
At MEDITECH, security and privacy concerns are top of mind in every project, with additional processes and reviews in place for any third-party component in a solution. This begins during vendor vetting, ensuring that vendors have certified security practices and meet HIPAA guidelines if we will be exchanging Protected Health Information (PHI). This continues as we define use cases and determine the exact nature of shared data.
Data may be anonymized in some uses, ephemeral in others, and in those instances where identifiable data needs to be persistent, all parties must meet the highest levels of security protocols and execute business associate agreements which codify all data protection expectations.
- Assess your cloud resources
Preparing for AI begins with establishing your organization’s approach to the cloud, since integrating AI will rely on the use of some cloud services. There are just too many specialized functions to run entirely on-premise. A hybrid cloud solution can enable organizations to keep their current system of record, but interface with cloud components.
This also means evaluating how your data is available for various AI use cases. Consuming a pre-built model that looks at a very small footprint of instance-specific data can be almost a black box approach; deep learning may require a larger, more significant data set and training your own LLM will require ingestion of millions of documents. Starting with a workflow that can easily be augmented, not replaced, by a pre-built model is a great way to begin your journey.
MEDITECH has been successfully guiding healthcare organizations across the country and around the world along the path to digital transformation. Our thoughtful and deliberate approach to the use of AI is just the next chapter. Already today, Mile Bluff Medical Center is using AI-powered search and summarization within Expanse delivered via the cloud-based MEDITECH as a Service (MaaS) subscription model, and Fraser Health is piloting our Generative AI functionality to help automatically draft hospital course discharge summaries.
We are committed to supporting health systems in this next era of healthcare. I hope these starting points will help your organization identify the way forward.
Connect with us to learn more about MEDITECH’s vision for AI at HIMSS and ViVE 2024.
Join MEDITECH at ViVE 2024, Feb. 25-28 in Los Angeles.