Medical coder — tasks, one by one

Coding short, routine charts

Short, standardised documents are where coding software first codes charts by itself: some health systems send the charts the software is confident about straight to billing and route the rest to people. The work that remains is the exceptions — the unusual or poorly documented charts the software sends back.

No primary source measures how many charts are coded without a person; the published figures come mostly from the vendors selling the software, and they are not used as evidence here.

Coding complex inpatient stays

Software increasingly reads the record and proposes codes, and the coder decides. Kaohsiung Medical University Chung-Ho Memorial Hospital in Taiwan deployed and tested an AI system in January 2023 that offers ICD-10-CM code suggestions to its certified coding specialists for diagnosis-related-group assessment; the coders review the suggestions and select the final codes.

That is a three-month evaluation in one hospital, under Taiwan's diagnosis-related groups; it does not measure time saved or show coding done without a coder.

Querying doctors when the record is unclear

Software can flag a gap in the documentation, but deciding what to ask, asking it without leading the doctor, and getting an answer that holds up in an audit is a conversation between people with professional rules on how a query may be worded.

This rests on how documentation queries work, not on a measurement of how many queries coders now send.

Reviewing codes that software suggested or assigned

Where software proposes codes, reviewing them becomes a standing part of the job rather than an exception. In the Taiwan hospital's evaluation the certified coding specialists reviewed every suggestion and chose the final codes, and the authors report that the system also helped detect coding errors in 1.9% of cases (50 of 2,632).

One hospital's evaluation shows review as part of that workflow; it does not show how widespread this work is or how many coders do it.

Audits, denials and appeals

Software now samples charts for audit, flags claims likely to be denied and can draft appeal letters. Judging whether a code was right under the payer's rules, and making the case to a reviewer, stays with people who know both the record and the rules.

This rests on what audit and denial software can do, not on a measurement of how audit work has changed.

Keeping up with code sets and payer rules

Coding software is updated with new code sets and can surface the rule that applies to a chart. Knowing when a rule applies, and noticing when the software's version is wrong or out of date, is still part of a coder's professional competence.

This rests on how coding software is maintained, not on a measurement of how coders now spend their training time.