Human Expertise Still Matters in the Age of AI

Although AI offers substantial promise for efficiency and scale, experts continue to emphasize that data quality, clinical context, and patient safety still depend on experienced human oversight. In the latest episode of Exploring AI in Oncology, Waqas Haque, MD, MPH, sat down with Erin Tobaison, Senior Clinical Data Abstraction Manager at Carta Healthcare, to discuss how AI is reshaping registry workflows while reinforcing the need for strong clinical judgment, specialty expertise, and institutional support.

Clinical Experience as a Foundation for Data Quality

Ms. Tobaison brings a distinctive perspective to the conversation. Before transitioning into tumor registry operations and broader clinical data management, she worked as a radiation therapist for 15 years. That frontline experience continues to shape how she approaches abstraction and quality oversight today. “I treated the patients in real time. I worked on their treatment plan. I saw them day in and day out,” she said, underscoring the depth of understanding that comes from direct patient care.

That clinical background is especially valuable in oncology, where treatment courses may evolve in response to toxicity, disease progression, or logistical factors. Ms. Tobaison described how such changes may be obvious to clinicians involved in care delivery, yet far less clear to abstractors reviewing the record retrospectively. “I knew exactly what I needed to do there,” she said, but others without that practical experience “didn’t have that real-world kind of knowledge.” In her view, the role of a clinical data manager is not simply to oversee abstraction, but to help teams understand the "why" behind the documentation so they can more accurately capture the full patient story.

Why Data Abstraction Is Also a Patient Safety Issue

A central message in this discussion is that data abstraction should not be viewed solely as a compliance or reporting task. Ms. Tobaison argues that high-quality abstraction is “a patient safety issue,” pointing to the downstream consequences of incomplete documentation and poor data capture.

This concern is particularly relevant when patients receive treatment across multiple sites. If information from outside institutions is not incorporated into the electronic medical record, abstractors may not have access to the complete history needed to document care accurately. That gap can affect future treatment planning, public reporting, and longitudinal analyses used to guide best practices. Ms. Tobaison warned that faulty or incomplete abstraction can ultimately misinform clinical decision making. In radiation oncology, for example, a missed prior treatment history could create the dangerous impression that a patient is eligible for additional radiation in an area that has already been treated. “You can see where it would be a very big issue to over-radiate a patient just because of poor data,” she said.

For healthcare professionals across the multidisciplinary team, this framing is important: registry quality is not separate from care delivery. Rather, it contributes to the accuracy of the information clinicians rely on both now and in the future.

Where AI Adds Value in Registry Work

Ms. Tobaison sees AI as a powerful operational tool, particularly when used to reduce the burden of repetitive manual review. She acknowledged that many experienced abstractors initially approach AI with skepticism, especially those who have spent decades performing careful chart abstraction by hand. She described one veteran nurse on her team with more than 20 years of abstraction experience who was initially doubtful about AI’s usefulness. Over time, however, she began to appreciate its practical benefits.

“AI is great at pulling data and pulling data quickly,” Ms. Tobaison said. Structured and semi-structured information—such as laboratory values, demographics, and key elements from pathology reports—can often be retrieved far faster through AI-assisted workflows than through manual chart review. What may take an abstractor 10 to 15 minutes to locate can often be identified almost instantly by AI.

She also noted meaningful time savings in practice. For relatively straightforward cases, AI-supported abstraction reduced completion time from approximately 30 minutes to 15 minutes. Even in highly complex cases that previously required five or six hours, AI helped abstractors focus their energy on the areas that truly required expertise, reducing the total time to three or four hours.

Why Human Oversight Remains Essential

Despite these gains, Ms. Tobaison is clear that AI has important limitations. “What it’s not great at doing is using clinical knowledge to read that path report,” she said. Ambiguous language, conflicting documentation, and nuanced clinical scenarios still require expert review. In those circumstances, abstractors must interpret the available evidence, determine whether clarification is needed, and sometimes return to the pathologist or treating clinician to ask, “What did you really mean here?”

For that reason, Ms. Tobaison repeatedly stresses a core principle for safe AI adoption: “You’re always going to need a human in the loop.” In her view, AI should augment and not replace the clinical reasoning that experienced abstractors bring to the process. The future of abstraction is therefore not automation alone, but hybrid intelligence, in which technology accelerates data retrieval while human experts preserve quality and context.

Operational Challenges for Healthcare Leaders

Beyond the AI conversation, Ms. Tobaison also highlighted several organizational issues that leaders often underestimate. One is the assumption that a data abstractor can move easily between specialties such as oncology, cardiovascular, and surgical registries. In practice, each domain has its own rules, workflows, and clinical logic. “An abstractor’s an abstractor and it’s really not the case,” she said, explaining that even a highly experienced oncology abstractor may become “a novice again” when reassigned to another registry area.

She also cautioned against rigid productivity benchmarks that fail to account for case complexity. Some cases can be abstracted efficiently, while others require substantially more time due to fragmented documentation, multiple sites of care, or unusual clinical circumstances. When leaders evaluate performance without understanding that variation, they risk creating pressure to prioritize speed over accuracy.

Investing in People to Strengthen Data Teams

When asked what healthcare leaders should prioritize, Ms. Tobaison’s answer was unequivocal: “They need to invest in their people.” She emphasized the importance of education, regular auditing, and continuous feedback. Monthly audits that compare AI-assisted and manual abstraction can help identify mismatches, reinforce best practices, and support both senior and newer abstractors. In her view, strong data teams are built not only on technology, but on knowledgeable professionals who receive the training and support needed to use that technology well.

Key Takeaways

• Patient Safety Link: Data abstraction is closely tied to patient safety and longitudinal care quality.

• AI Efficiency: AI dramatically improves efficiency, especially for routine data retrieval and structured reports.

• Human Criticality: Expert human oversight remains essential for navigating clinical ambiguity and nuance.

• Specialty Nuance: Registry work is specialty-specific; abstractors are not easily interchangeable across different medical domains.

• Workforce Investment: Building strong data teams requires ongoing education, audits, and leadership support.

About the Speakers

Erin Tobaison is the Senior Clinical Data Abstraction Manager at Carta Healthcare, with experience spanning oncology, cardiovascular, and surgical registries. She previously worked as a radiation therapist for 15 years, bringing valuable frontline clinical insight to her leadership role in tumor registry operations, data quality, and abstraction management.

Waqas Haque, MD, MPH, is a second-year Hematology/Oncology Fellow at the University of Chicago. He recently completed his Internal Medicine Residency at New York University (NYU) in a Clinical Investigator Track. Dr. Haque has balanced his patient care work with a variety of research projects. During his fellowship training at the University of Chicago, he plans to further his work in innovative clinical trial design, value-based care delivery to cancer patients, and clinical investigation.