AI can help healthcare organizations address workforce shortages by taking over time-consuming data work while keeping trained professionals responsible for important decisions. It uses cancer registries as a case study. Cancer registrars must transform information scattered across pathology reports, imaging, surgery, genomic testing and treatment records into standardized registry data. With fewer than 6,000 ODS-certified cancer registrars worldwide, the growing volume of cancer cases is creating a significant capacity problem.
The approach described by Brent Dover, CEO of Carta Healthcare, is deliberately “registrar-in-the-loop” rather than fully autonomous. AI reads the patient's record, identifies relevant information, proposes answers to registry questions and connects those answers to the source material. The credentialed registrar then reviews, corrects and validates the information before submission. This is particularly important because errors in staging or histology can affect reported outcomes, quality measurements and cancer research cohorts.
The model changes the registrar's role from searching to checking. Instead of spending much of the day locating information across hundreds of pages of records, professionals can concentrate on resolving contradictions, applying specialized rules and making judgments that require clinical nuance. Carta reports that, in registries where its approach is being used, abstraction time has fallen by as much as 66%, costs by half or more, while inter-rater reliability has remained above 98%. These figures are company-reported results rather than an independent evaluation.
The broader lesson is that healthcare AI may deliver its greatest value not by replacing clinical professionals, but by increasing the capacity of scarce specialists. The article recommends evaluating AI through operational outcomes such as turnaround time, backlog, cost per case and reliability rather than generic AI benchmarks. In high-stakes areas such as oncology, the goal should be to automate information retrieval and repetitive work while keeping an accountable expert responsible for the final decision.
