アプリケーションガイド
E/MレベルコーディングのためのAI
AI for E/M level coding reads a visit note or transcript and suggests an evaluation and management code, such as office visit codes 99202 through 99215.
このページでは4 分で読めます
概要
It bases the suggestion on medical decision making (MDM) or on the clinician's total time, following the AMA rules in effect since 2021. E/M visits are among the most frequently billed services, so small errors add up across thousands of visits. AI tends to go wrong in judging complexity, such as which problems were actually addressed and how much risk the management plan carried.
ディープダイブ
In 2021 the AMA changed the rules for office and outpatient E/M visits. History and exam no longer set the level. They only need to be documented as medically appropriate. The level now depends on either medical decision making or the total time the physician or other qualified clinician spent on the date of the visit. Code 99201 was deleted. In 2023 the same approach was extended to most other E/M families, including hospital, emergency department, nursing facility and home visits. Emergency department visits cannot be billed on time. MDM has three elements: the number and complexity of problems addressed; the amount and complexity of data reviewed and analyzed; and the risk of complications, morbidity or mortality from patient management. Each is rated straightforward, low, moderate or high, and the MDM level is the highest level that at least two of the three elements meet. For established patients, straightforward through high correspond to 99212 through 99215. AI tools pull these elements out of the note and map them to that scale. Errors tend to follow a few patterns. A problem counts only if it was addressed, meaning evaluated or treated, not simply listed. Data credit follows specific rules. Ordering a test includes reviewing its result later, so the review is not counted a second time. Risk depends on the management options chosen or considered, such as prescription drug management or a documented decision about hospitalization. It does not depend on how sick the patient sounds. Social determinants of health that significantly limit diagnosis or treatment can support moderate risk. A common misconception is that longer notes support higher levels. Ambient scribes produce long, detailed notes, but length says nothing about MDM. A related risk is that a tool trained on physicians' past choices copies their existing overcoding or undercoding.
戦略的影響
ビルドの選択
AI が実際の成果を向上させるかどうかは、アプリケーション レベルの設計によって決まります。
チームとワークフロー
ワークフローを適切に統合すると、ユーザーが信頼できる生産性が向上します。
リスクと安全性
適切な範囲のユースケースにより、変更の疲労と実装のリスクが軽減されます。
The Future of AI for E/M Level Coding
As ambient scribes spread, suggesting an E/M level is becoming a standard feature of documentation tools rather than a separate product. That makes independent auditing more important, because the same system is both writing the note and scoring it. Payers already compare level distributions across clinicians and are likely to notice practices whose levels jump after adopting these tools. The most useful tools will explain their reasoning one element at a time and leave the final choice to the clinician. Opaque tools that output only a code are harder to defend in an audit.
現実世界の実装
An ambient scribe drafts the note for a hypertension follow-up. Blood pressure is above goal and the physician raises the lisinopril dose. The tool suggests 99214: a chronic illness that is progressing counts as a moderate problem, and prescription drug management counts as moderate risk.
A tool sees six diagnoses listed in a stable patient's assessment and suggests 99215. An auditor points out that the physician addressed only two of them. The others were carried forward from the problem list and do not count.
A practice bills some visits on time. Its AI checks that the physician documented total time personally spent on the date of the visit, and flags notes where the time statement includes time spent by nurses or medical assistants.
A compliance team runs the AI over 500 visit notes and compares its level distribution with the levels physicians chose. The AI picks 99215 far more often, which leads to a review of how it counts problems.
リスクとガードレール
壊れたプロセスを自動化すると、既存の問題がさらに拡大する可能性があります。
チームが過剰に自動化し、必要な人間の判断を排除してしまう可能性があります。
出力が継続的に評価されないと、品質が変動する可能性があります。
実装ロードマップ
現在のワークフローをマッピングし、最も摩擦が大きいステップを特定します。
完全自動化の前に人間によるチェックポイントを定義します。
プロンプト、エスカレーション パス、品質基準についてユーザーをトレーニングします。
タスクレベルの結果を追跡して、持続的な価値を確認します。
探検を続けましょう
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the AI for E/M Level Coding quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
よくある質問
What is AI for E/M Level Coding?
AI for E/M level coding reads a visit note or transcript and suggests an evaluation and management code, such as office visit codes 99202 through 99215. It bases the suggestion on medical decision making (MDM) or on the clinician's total time, following the AMA rules in effect since 2021. E/M visits are among the most frequently billed services, so small errors add up across thousands of visits. AI tends to go wrong in judging complexity, such as which problems were actually addressed and how much risk the management plan carried.
Under the 2021 office visit rules, what determines the E/M level?
Since 2021, office and outpatient E/M levels are based on MDM or total time. History and exam only need to be documented as medically appropriate.
How many of the three MDM elements must meet or exceed a level for that MDM level to apply?
The MDM level is the highest level that at least two of the three elements (problems, data and risk) meet or exceed.
An AI suggests 99215 because six diagnoses appear in the assessment, but only two were evaluated or treated. What error did it make?
A problem counts toward MDM only if it was addressed, meaning evaluated or treated at the visit. Conditions carried forward from the problem list do not count.
For the hypertension follow-up where the physician raised the lisinopril dose, what supports moderate risk?
Prescription drug management is a standard example of moderate risk in the MDM table. Together with a chronic illness that is progressing, it supports 99214.
Why is a longer ambient-scribe note not evidence of a higher E/M level?
Note length has no bearing on MDM. A detailed note of a straightforward visit is still a straightforward visit.
学び続ける
関連ガイド
このトピックのために選ばれたその他のガイド