How To Assess Healthcare AI Data Readiness
Rather than attempting to overhaul the entire data environment, healthcare organizations should evaluate the specific data they need for the AI use case they’re looking to roll out. Begin with several questions:
- How accessible is the data we need? Identify the sources required for the intended application, including data found in clinical and administrative systems. Can it be shared across systems, or is it trapped in silos?
- Is the data reliable and consistent? Look for missing fields, duplicate records, outdated information and conflicting definitions. Healthcare organizations may need to reconcile differences in how departments, regulators and specialty societies define and report clinical measures.
- How well is data governance established? Who owns critical data assets? Who can access them? How is their use controlled? Review privacy requirements, security safeguards and policies governing AI applications.
- Can the organization monitor performance? Establish processes to evaluate data quality, validate AI outputs and identify problems after deployment. Define where human review is necessary, particularly when AI supports clinical or other consequential decisions.
- Are business and clinical goals aligned? Identify the outcomes the organization wants to achieve and the metrics it will use to measure success. Assign responsibility for resolving data issues and overseeing implementation.
These questions provide an initial screening, not a comprehensive assessment. CDW’s AI and Data Readiness Checklist offers a broader framework covering strategy and governance, data readiness, AI operations, workforce adoption, infrastructure and security.
Where AI projects often fail is in the scaling phase, after a seemingly successful pilot. Pilot projects tend to succeed precisely because they’re small, with a data set carefully selected by a motivated project champion.
When the organization tries to expand that application across departments, problems arise, because at that point different systems, data definitions and integration requirements are involved.
Shippy says that AWS’s healthcare provider customers have been increasingly telling her that rather than treating a pilot as an isolated experiment, they are working to consider how an initiative could eventually extend across the enterprise from the outset.
“Maybe it’s phase one, phase two, phase three, understanding from the beginning that it’s going to be organizationwide,” she says.
Close the Gap Between Data Readiness and Healthcare AI Goals
Identifying data foundation weaknesses early makes it clearer how organizations should prioritize improvements. The better the data foundation, the more effectively — and safely — the AI application will run. But organizations need not wait until everything is perfect before deploying AI.
“You don’t have to have every single part of it in place, but you’re working on it continuously to get to that foundation,” Shippy says. Organizations can begin implementing selected AI projects while improving the underlying data environment, provided they address the requirements and risks of each use case.
For example, suppose a health system is looking to use AI to automate a specific administrative workflow. After identifying the data needed, it would establish appropriate access controls and validate data quality. Then, it can expand the initiative as integrations, governance and monitoring mature.
“The opportunity to continue to improve and ensure that their data is ready, and they have the right foundation — that’s going to be ongoing, continuous improvement,” Shippy says.
