What We’ve Learned About AI and Registry Management
Artificial intelligence has changed considerably over the past four years, and so has the way healthcare organizations are thinking about its role in registry management. At Registry Partners, our perspective on AI comes from working alongside hospitals, registry teams, technology vendors, and AI-powered solutions as these tools have been evaluated, tested, and incorporated into real registry environments. That experience has given us the opportunity to see where AI can create meaningful efficiencies, where challenges emerge, and what needs to be in place for successful implementation.
Successful AI adoption depends on understanding the registry, identifying the right opportunities for technology, and combining AI with the people and processes needed to use it effectively.
Here are some of the most important things we have learned.
1. Successful AI Implementation Starts Before the Technology
When a registry is struggling with staffing shortages, backlogs, inefficient workflows, or increasing reporting demands, AI can look like an immediate solution. But technology cannot correct every operational problem that already exists within a registry.
If workflows are inconsistent, processes are inefficient, or data quality issues already exist, introducing AI into that environment may not produce the improvements an organization expects. Before implementing new technology, it is important to understand how the registry is currently operating. Where are the bottlenecks? Which processes consume the most time? Where are quality issues occurring? What is creating the greatest burden for the registry team?
Sometimes AI may be part of the answer. Other times, improving a workflow, strengthening data quality, or addressing another operational issue may need to happen first. This is one reason Registry Partners approaches AI as part of the larger registry operation rather than as a separate technology initiative.
2. No Single AI Solution Is Right for Every Registry
There is no universal AI solution for registry management. Registry programs vary in their specialties, reporting requirements, staffing structures, workflows, existing technology, and operational challenges. AI solutions vary as well, both in what they are designed to accomplish and how they fit into existing registry processes. The most important place to begin is not with a particular product or vendor. It is with the problem that the registry is trying to solve.
Is the goal to improve case finding? Reduce time spent reviewing documentation? Help prioritize workloads? Identify missing information? Strengthen quality assurance? Create additional capacity for an understaffed team?
Clearly defining the need makes it easier to evaluate whether a solution is actually a good fit. Instead of beginning with “What can this AI solution do?” we have found that a more useful question is “What does this registry need AI to do?”
3. Implementation Is Only the Beginning
Selecting and implementing an AI solution is an important milestone, but it is not the end of the process. Once AI becomes part of a registry workflow, its performance needs to be evaluated in the context of real cases and real registry requirements.
Where is it performing well? Where are errors occurring? Are certain types of cases consistently requiring additional review? Is the technology producing the efficiency the organization expected? Is the resulting data meeting registry standards?
This is where experienced registry professionals become particularly important. They can evaluate AI-generated information, identify patterns in errors, validate results against registry standards, and help determine where workflows or processes need adjustment.
We have seen the evolution of the traditional abstractor role into an abstraction/quality analyst role. The abstractors review the patient data presented by AI for accuracy and completeness. This review process requires experienced abstractors who are confident in data standards to effectively review the retrieved data and make a determination if it can be accepted into the case or if additional patient information needs to be reviewed. The abstractors still need to seek information from the EHR if the AI tool isn’t able to pull all required patient data in for review. An example would be treatment performed at referring facilities through shared portals like Epic’s Care Everywhere. We believe that, from our experience to date, as AI becomes more accurate in what it is capable of collecting, the abstractor’s review process will be reduced. However, currently significant human oversight is required to ensure current-day cases are complete and to assist in the development and training of AI tools to advance their capabilities.
AI implementation should be viewed as an ongoing process of evaluation and refinement rather than a one-time technology launch.
4. AI Changes the Work. It Does Not Eliminate the Need for Expertise.
One of the greatest opportunities we see with AI is the ability to use registry professionals’ expertise more effectively. AI can assist with portions of case finding, documentation review, workflow prioritization, data validation, and other information-heavy tasks. As technology takes on more of that work, registry professionals can direct more of their time toward complex cases, clinical interpretation, quality assurance, data analysis, and other responsibilities that require their expertise.
That changes the role of the registry professional rather than eliminating it. It also means organizations need to think about how their teams will work alongside AI. Training, workflow development, quality processes, and clear expectations for human review all become part of successful implementation.
One issue we identified with AI retrieved information is the completeness and organization of the information presented. An example is the medication doses administered to a radiation patient. The information presented to the abstractor did not add the phased doses of radiation, so it had to be manually tallied by the abstractor. This example is representative of the time the abstractor still has manual requirements while working alongside the AI tools.The strongest results we have seen come from treating AI as another resource available to the registry team, not as a replacement for the expertise that makes registry data reliable.
5. AI Should Fit Into the Registry, Not the Other Way Around
Perhaps the most important lesson is that AI should support the registry’s goals. Organizations should not have to redesign successful processes to accommodate technology. The technology should address a clearly identified challenge and make the registry stronger, more efficient, or better equipped to meet its responsibilities. That requires looking beyond the technology itself.
Staffing, workflows, data quality, education, quality assurance, governance, and ongoing registry management all influence whether an AI implementation ultimately succeeds. This is why we see AI as a natural extension of the work Registry Partners already does with healthcare organizations. Assessing registry health, improving operations, mentoring teams, managing registry programs, strengthening data quality, and helping organizations implement new solutions are all part of building a registry that can use emerging technology effectively.
Integrating AI Into Your Registry
Organizations considering AI do not have to begin by choosing a product. A better starting point is to understand what you want to improve, what your registry needs, and what a potential solution must accomplish.
Registry Partners works alongside organizations throughout that process, from assessing current registry operations and identifying opportunities for AI to evaluating potential solutions, supporting implementation, and providing the registry expertise needed to monitor performance over time.
As AI continues to evolve, there will be more options available and more ways to incorporate the technology into registry workflows. Asking the right questions before selecting a solution can help organizations make decisions based on their actual needs rather than the capabilities of a particular product.
