AGENT SKILL

abstracting-clinical-documentation

Extracts codeable diagnoses and procedures from clinical notes with specificity capture. Use when abstracting medical records, identifying codeable conditions, or capturing documentation specificity.

License 许可
Apache-2.0
最近更新
2026-11-09
Stars
★ 0
Forks
0

01 前言 Front Matter

name
abstracting-clinical-documentation
tags
[ "medical-coding-and-billing", "clinical" ]
source
https://github.com/CaseMark/skills
license
Apache-2.0
language
en
metadata
{ "author": "casemark", "skill_modes": [ "Analysis" ], "document_types": [ "Report" ], "practice_areas": [ "Medical Coding", "Revenue Cycle", "Health Information Management" ] }
description
Extracts codeable diagnoses and procedures from clinical notes with specificity capture. Use when abstracting medical records, identifying codeable conditions, or capturing documentation specificity.
imported_at
2026-09-20
source_path
skills/med/abstracting-clinical-documentation/SKILL.md
source_repository
CaseMark/skills

02 资源文件 1

03 SKILL.md 正文

Abstracting Clinical Documentation

Transforms bulk medical records into structured, chronologically ordered clinical abstracts. The output supports downstream use in medical-legal review, insurance audit response, CMS quality reporting, peer review, and clinical research.


Why This Skill Exists

Clinical documentation abstraction is the bridge between raw medical records and every downstream process that depends on them:

Consumer Why They Need a Structured Abstract
Billing / Revenue Cycle Accurate code assignment (ICD-10-CM/PCS, CPT, HCPCS) requires every codeable condition and procedure surfaced with laterality, severity, and linkage.
Quality Reporting (CMS/HEDIS) Measure numerators and denominators depend on discrete data elements extracted from narrative notes.
Litigation Support Med-legal reviewers need a defensible chronological timeline with source-page citations to assess causation, standard of care, and damages.
Insurance Audit RAC, ZPIC, and MAC auditors compare billed codes against documentation; the abstract is the first line of defense.
Peer Review / Credentialing Case abstracts drive mortality and morbidity conferences, sentinel event reviews, and provider profiling.
Clinical Research Chart abstraction is the primary data-collection method for retrospective studies and registry submissions.

A poor abstract produces incorrect codes, failed audits, unsupported legal opinions, and missed quality measures. Precision here compounds everywhere.


Checkpoint A: Pre-Draft Intake (Mandatory)

Before touching a single record, confirm the following with the requesting party. Do not proceed until every row is resolved.

Item Question to Resolve Why It Matters
Purpose Litigation, audit defense, quality reporting, coding review, research, or peer review? Determines depth, element focus, and whether de-identification is required.
Record Scope Which facilities, providers, and date ranges are in scope? Prevents abstracting irrelevant episodes and controls volume.
Timeframe Exact start and end dates, or "all available"? Anchors the chronological spine of the abstract.
Target Audience Coders, attorneys, physicians, quality analysts? Drives vocabulary level and output format.
PHI Handling Full PHI retained, Limited Data Set, or Safe Harbor de-identified? Determines whether the 18 HIPAA Safe Harbor identifiers must be stripped.
Source Inventory Has a document inventory or index been provided? If not, the first deliverable is the source document manifest (see Step 1).
Known Gaps Are any records known to be missing or outstanding? Gaps must be flagged, not silently skipped.

Workflow

Step 1: Source Document Manifest

Catalog every record before abstracting. Use the following classification:

Document Type Typical Content Abstraction Priority
History & Physical (H&P) Chief complaint, HPI, ROS, exam, assessment/plan High -- anchors encounter context
Progress Notes Daily/interval updates, reassessments High -- tracks clinical trajectory
Operative Reports Pre/post-op dx, procedure narrative, findings, specimens High -- procedure detail source
Discharge Summary Hospital course, discharge dx, follow-up plan High -- episode closure
Consultation Notes Specialist assessment and recommendations Medium-High
Lab / Pathology Reports Results, reference ranges, interpretations Medium -- supports dx specificity
Imaging Reports Findings, impressions, recommendations Medium
Medication Administration Records (MAR) Drug, dose, route, frequency, duration Medium
Nursing Notes / Flowsheets Vitals, I&O, wound care, fall risk Low-Medium -- ancillary but relevant for quality/litigation
Therapy Notes (PT/OT/ST) Functional status, goals, progress Low-Medium
Referral / Authorization Forms Requested services, clinical justification Low -- administrative context
Unsigned / Addended Notes Late entries, amendments, corrections Critical to flag -- often missed

Deliverable: A completed Source Document Manifest table listing every document with: document type, date, author/provider, page range or file reference, and a completeness flag (Complete / Partial / Missing).

Step 2: Chronological Data Extraction

Walk the records in strict date order. For each encounter or clinical event, extract the following data elements:

Data Element Capture Requirements
Encounter Date Exact date (and time if relevant, e.g., ED visits).
Setting Inpatient, outpatient, ED, ambulatory surgery, telehealth, SNF, home health.
Provider(s) Name, credentials, specialty. Note attending vs. consulting vs. referring.
Chief Complaint / Reason Patient-stated reason or referral indication.
Diagnoses Each diagnosis with: ICD-10-CM-level specificity, laterality, acuity (acute/chronic), onset context (new vs. established), and clinical status (active, resolved, rule-out).
Procedures Each procedure with: CPT/HCPCS-level description, approach, laterality, qualifier, and outcome.
Medications Name, dose, route, frequency, start/stop dates, indication if documented.
Lab Results Test name, result value, units, reference range, abnormal flag.
Imaging Findings Modality, body region, key findings, impression.
Referrals / Orders To whom, for what, urgency.
Hospitalizations Admit/discharge dates, principal and secondary dx, procedures, disposition.
Functional Status ADLs, mobility, cognitive status (especially for quality/litigation).

Use the chronological abstract table format in references/ABSTRACTION-TEMPLATE.md.

Step 3: Terminology Normalization

  • Map all diagnoses to ICD-10-CM descriptors. Do NOT assign codes; capture the natural-language equivalent at code-level specificity (e.g., "acute ST-elevation myocardial infarction of the left anterior descending artery" not just "heart attack").
  • Map procedures to CPT/HCPCS descriptors where identifiable.
  • Standardize medication names to generic (brand in parentheses).
  • Normalize lab units where source uses non-standard reporting.

Step 4: Distinguish Fact from Inference

This is the most common source of abstraction error. Apply these rules without exception:

Category Rule Example
Documented Fact Appears in the clinical narrative authored by the treating provider. "Assessment: Type 2 diabetes mellitus with diabetic nephropathy, stage 3 CKD."
Inference / Implication Suggested by data but not explicitly stated by a provider. Lab shows HbA1c 9.2% but no diabetes diagnosis documented.
Assessment vs. Diagnosis An assessment is a clinical impression; a confirmed diagnosis requires explicit provider documentation or supporting evidence. Do not conflate. "R/O PE" is an assessment, not a diagnosis of pulmonary embolism.
Primary Source The treating provider's own documentation. Attending physician's progress note.
Secondary Reference Another provider quoting or summarizing. Consultant note stating "per Dr. Smith, patient has CHF."

Mark inferences with [INFERRED -- not explicitly documented]. Mark secondary references with [PER: source provider/document].

Step 5: CDI Query Generation

When documentation is ambiguous or incomplete, generate Clinical Documentation Improvement (CDI) queries rather than guessing. Format:

CDI QUERY #[n]
Record:    [Document type, date, provider]
Issue:     [What is ambiguous or missing]
Clinical Indicators: [Supporting data that suggests the undocumented condition]
Suggested Clarification: [Specific question for the provider]
Impact:    [Coding, quality measure, or legal relevance]

Common CDI triggers:

  • Lab/imaging findings inconsistent with documented diagnoses
  • Medications prescribed without a corresponding documented indication
  • Severity or acuity not specified (e.g., "heart failure" without systolic/diastolic, EF, or NYHA class)
  • Cause-and-effect linkage absent (e.g., "CKD" and "diabetes" documented but no causal relationship stated)
  • Unsigned or contradicted documentation

Step 6: PHI Handling

If de-identification is required, apply HIPAA Safe Harbor method. The 18 identifiers to strip or generalize:

  1. Names
  2. Geographic data smaller than state
  3. Dates (except year) for dates related to an individual
  4. Phone numbers
  5. Fax numbers
  6. Email addresses
  7. SSN
  8. Medical record numbers
  9. Health plan beneficiary numbers
  10. Account numbers
  11. Certificate/license numbers
  12. Vehicle identifiers and serial numbers
  13. Device identifiers and serial numbers
  14. Web URLs
  15. IP addresses
  16. Biometric identifiers
  17. Full-face photographs
  18. Any other unique identifying number or code

For Limited Data Sets, retain dates and geographic data at zip-code level.


Output Structure

The final deliverable includes these sections in order:

  1. Abstract Header -- Patient identifier (or de-identified placeholder), DOB/age, sex, date range, purpose, abstractor, date of abstraction.
  2. Source Document Manifest -- Table from Step 1.
  3. Chronological Case Abstract -- Table from Step 2 (see template).
  4. Problem List Summary -- Consolidated active/resolved problem list with onset dates and current status.
  5. Medication Summary -- Current medications with start dates and indications.
  6. CDI Queries -- Any generated queries from Step 5 (or "None identified").
  7. Abstraction Notes -- Gaps, limitations, assumptions, and flags.
  8. Attestation -- Statement that the abstract reflects only documented facts, with inferences and gaps explicitly marked.

Checkpoint B: Post-Draft Alignment (Mandatory)

Before delivering the final abstract, verify each item:

Check Pass Criteria
Chronological integrity Every entry is in strict date order. No undated entries unless flagged.
Source traceability Every data element cites its source document (type, date, provider, page/section).
Fact vs. inference marking All inferred data elements carry the [INFERRED] tag. All secondary references carry [PER:].
Completeness against manifest Every document in the manifest has been abstracted or explicitly marked as "Not abstracted -- [reason]".
Terminology consistency Diagnoses use ICD-10-CM-level language throughout. Medications use generic names.
PHI compliance If de-identification was required, all 18 identifiers have been scrubbed. Spot-check 5 random entries.
CDI queries generated Ambiguous documentation has been flagged, not silently resolved.
No silent gaps Missing records, illegible sections, and unsigned notes are explicitly called out.
Purpose alignment Output depth and element focus match the stated purpose from Checkpoint A.

Quality Audit

Common Abstraction Errors (Ranked by Impact)

# Error Impact Prevention
1 Conflating assessment with confirmed diagnosis Overcoding, audit liability, legal exposure Apply fact-vs-inference rules strictly
2 Missing unsigned or late-addended notes Incomplete clinical picture, missed dx/procedures Check for addenda on every note; include document status in manifest
3 Skipping ancillary records (nursing, therapy, MAR) Missing functional status, medication errors, quality measure gaps Manifest must include all document types, not just physician notes
4 Losing chronological order Causal relationships become unclear; legal timelines break Sort by encounter date before writing a single line
5 Copying secondary references as primary facts Attribution errors; circular documentation Always trace to the originating provider's note
6 Omitting laterality, severity, or acuity Codes default to "unspecified," triggering audits Extract every specificity qualifier the documentation supports
7 Inconsistent terminology Same condition described differently across entries Normalize at Step 3; use a single descriptor per condition throughout
8 PHI leakage in de-identified output HIPAA violation Run Safe Harbor checklist against final output

Audit Sampling

For abstracts exceeding 50 entries, perform a minimum 10% random sample audit against source documents. For litigation-purpose abstracts, audit 100% of entries related to the contested clinical issues.


Key Rules

  • Never present inferred clinical data as documented fact.
  • Every data element must trace to a specific source document, date, and provider.
  • Chronological order is non-negotiable. If a date is uncertain, flag it; do not omit the entry.
  • When documentation is ambiguous, generate a CDI query -- do not resolve ambiguity silently.
  • Mark with [VERIFY] any data point that cannot be confirmed from available records.
  • Include appropriate disclaimers: the abstract is a summary tool, not a substitute for the complete medical record.
  • Escalate to human review when clinical complexity exceeds abstraction confidence.

Guidelines

  • Clinical documentation abstraction is a high-stakes activity. Errors propagate to coding, billing, quality reporting, litigation support, and clinical research. Treat abstraction with the same rigor as clinical documentation itself
  • Always prioritize the treating provider's own documentation over secondary references. A consultant quoting another provider is not equivalent to the original provider's own note for abstraction purposes
  • Chronological integrity is the backbone of a defensible abstract. Every data point must be anchored to a specific date, source document, and provider. Undated entries or entries without source attribution undermine the entire abstract
  • CDI queries are a feature, not a flaw. Generating queries when documentation is ambiguous demonstrates rigor and prevents the silent propagation of assumptions through the coding and billing pipeline
  • For litigation-purpose abstracts, apply a higher standard: 100% audit of entries related to contested clinical issues, explicit notation of documentation gaps, and clear distinction between documented facts and clinical inferences
  • PHI handling is not optional for any abstraction activity. Verify de-identification requirements before beginning work and validate compliance before delivering the final product
  • When abstracting records from multiple facilities or providers, maintain consistent terminology normalization across all sources. The same condition described differently by different providers must be reconciled to a single standardized descriptor throughout the abstract
  • Escalate to a clinical subject matter expert when the abstraction requires interpretation of complex clinical scenarios (e.g., distinguishing between related diagnoses, interpreting ambiguous operative findings, or determining whether a documented condition meets coding criteria)

04 同仓库更多 Skills