Healthcare organizations are increasingly adopting artificial intelligence to streamline operations and enhance patient care. Applications range from ambient clinical documentation to predictive modeling, offering extensive possibilities.
However, maximizing these benefits involves more than selecting appropriate AI tools. According to Dr. Angela Shippy, a senior physician executive at Amazon Web Services, success hinges on establishing a strong data foundation that ensures reliable, secure, and scalable implementation.
Data Challenges in Healthcare AI
While healthcare systems hold vast amounts of data, the difficulty lies in making this information accessible, consistent, and usable across the organization. Data governance and security are also vital concerns.
Shippy observes that many organizations recognize the importance of organizing their data. A common sentiment is, “We just need it to be in the right place at the right time and in the right format.”
Evaluating AI Data Readiness
Instead of overhauling their entire data environment, healthcare organizations should focus on assessing the specific data needed for their intended AI use case. This involves critical questions about data accessibility, reliability, governance, and performance monitoring.
For example, is the data siloed or can it be shared across systems? Are there missing fields, duplicate records, or inconsistent definitions? Who owns the data, and how is its use managed? These questions help identify potential weaknesses in the data foundation.
Scaling is often where AI projects encounter difficulties. A successful pilot, typically limited in scope, may not scale effectively to a larger, more complex implementation. As organizations expand AI applications across departments, they face challenges related to differing systems, data definitions, and integration needs.
Shippy notes a change in approach among AWS’s healthcare provider customers. Rather than treating pilots as isolated experiments, they increasingly consider how initiatives can be scaled across the enterprise from the start. This involves a phased approach, with a clear understanding of the eventual organization-wide impact.
Early identification of data foundation weaknesses helps organizations prioritize improvements. While a perfect data environment is not required for AI deployment, continuous enhancement is essential. Organizations can start AI projects while simultaneously improving their data infrastructure, provided they address the specific needs and risks of each use case.
For instance, a health system aiming to automate an administrative workflow would first identify the necessary data, establish access controls, and verify data quality. As integrations, governance, and monitoring improve, the initiative can be expanded. The regulatory environment introduces complexity to AI implementation in healthcare. Organizations must adhere to various reporting standards and definitions when sharing information with payers, regulators, patients, and specialty societies. AI governance must comply with these requirements while protecting sensitive clinical data.
Human oversight is essential, particularly with agentic AI applications. As Shippy stresses, “There’s always going to be a human in the loop at some point to really ensure that we’re staying on track.” Post-launch evaluation, performance monitoring, and adjustments are key to ensure the technology functions as intended.
AI Readiness Assessment Tools
CDW offers an AI and Data Readiness Checklist to help organizations evaluate their preparedness for AI implementation. This tool provides a framework covering strategy, data readiness, operations, workforce adoption, infrastructure, and security.
Expanding AI Initiatives Across Healthcare Organizations
Healthcare providers are adopting a phased approach, considering the broader organizational impact from the beginning. This strategy involves breaking implementation into stages, ensuring each phase builds on the last, and addressing the unique needs of different departments and systems.
Ongoing Enhancement and Human Supervision
For example, automating administrative tasks involves identifying required data, setting access controls, and validating data quality. As these processes improve, the initiative can expand, providing a solid foundation for future AI applications.
