Why registry work keeps a person in the loop
This job sits between clinical notes and the databases that hospitals, states, and researchers depend on. Software can read a chart quickly. It cannot be accountable for what ends up in a cancer registry. Abstracting a case means reading pathology reports, treatment dates, and notes that disagree with each other, then deciding what the record actually says. That decision is the heart of the work.
Two other duties pull the same way. Following up on patient outcomes often means calling a clinic, chasing a missing report, and asking a physician to clarify what was written. Training clinical staff on documentation and electronic health record workflows is persuasion as much as instruction. Neither task is really about typing data; both are about getting other people to produce usable records.
So the honest answer to the question of whether AI will replace health information technologists is that it keeps eroding the clerical half of the role while leaving the accountable half in place. Our headline figure, Still needs a human, reads 62 out of 100 (higher is safer), and you can see how that is built on the methodology page.
What software handles, what it assists, and what lands on you
Some of this work is now routine for software. Compiling standard statistical reports from registry data and turning dictated or written notes into coded entries are both jobs that systems do with little supervision. The share of task time in that group is 25%, and the task list above shows which items sit there.
A second group is shared. Screening charts for missing or inconsistent fields and producing a first-pass case abstract are faster with a model doing the reading and a registrar doing the checking. That assisted share is 44%. The measure behind these splits is explained on the coverage method page.
The rest stays with people: completing case records by working with providers, teaching documentation standards to clinical teams, and signing off submissions to state and national registries. That block is 31% of task time. It is the part employers are still hiring for, even as the keystroke work shrinks.
How strong is the evidence?
Thin, and we say so. Our evidence grade for quality parity is D, which means no study has tested AI against qualified registrars on this job’s own tasks. Because of that, we publish no parity number here. Guessing one would be worse than leaving it blank.
What would settle it is specific and testable: a blind comparison on the same set of real cases, with model-produced abstracts and human-produced abstracts scored against a gold-standard reference, plus an error count on the records actually submitted to a registry. Until that exists, treat confident claims in either direction with care. The quality parity method explains the grades, and what the AI assistants say shows how the models answer this question about themselves.
Labor market data is firmer. The Bureau of Labor Statistics counts about 38,100 US jobs in this occupation, with median pay of $68,020 and projected employment growth of 15.9% between 2025 and 2035 (BLS, 2025). Demand for the function is rising, not falling, even while the mix of daily tasks changes.
When the balance could shift
Most likely between 2040 and 2050 (8 in 10 of our scenarios). How that window is built is set out on the replacement year method page.
Two things could pull it earlier. Health records are already structured and digital, so there is no hardware step to wait for; our robotics requirement for this job is none. And running the software is cheap next to staffing a registry, which makes pilots easy to justify.
Two things hold it back. Registry and reporting programs expect a credentialed person to be answerable for what is submitted, and an audit trail with a name on it. Legacy records are also messy, with scanned documents, free-text notes, and local abbreviations that break automated extraction. Hospital IT procurement is slow for the same reasons.
How to stay needed
Lean into the work the task list leaves with people. First, own the hard abstraction cases where documentation conflicts. Second, run provider follow-up and query processes, because that is relationship work. Third, take responsibility for data quality audits and registry sign-off.
Two skills travel well. One is validation: knowing how to sample AI output, measure its error rate, and document the result. The other is data analysis beyond reporting, so you can answer questions from quality teams and researchers instead of only filling forms.
What to do: ask to lead the quality check on your employer’s next documentation or coding automation pilot.
If you want adjacent options, look at Health Informatics Specialists, Medical Records Specialists, and Medical and Health Services Managers. You can put any two of them side by side on our compare tool, see the wider health technologists and technicians family, or read how the question plays out across the healthcare sector.