Coding / CDS
Code the encounter while you document it. The Coding & Clinical Decision Support module sits between Clinical Documentation and RCM, listening to the encounter as it is built and returning coded suggestions, E/M level proposals, HCC capture hints, and CDS Hooks 2.0 alerts — all clinician-in-the-loop, all evidence-linked, all auditable.
What Makes This Different
Every EHR vendor ships an ambient scribe that listens and types. That is table stakes. REV.health's differentiator is the Question Learning Loop: every encounter produces a set of clinical questions that become clickable, reusable prompts for the next encounter of the same type. The system gets smarter with every visit — not because an LLM was retrained, but because real clinicians accepted, edited, or rejected real questions during real encounters, and those signals feed back into the encounter-type template.
How the Loop Works
- Encounter runs. During an acute visit, follow-up, AWV, or any encounter type, the clinician asks questions, orders labs, adjusts meds, and signs the note. The ambient scribe captures what was discussed; the coding module captures what was coded.
- Questions are extracted. After sign-off, the system extracts the clinical questions that drove the encounter — "Are you still taking lisinopril?", "Any chest pain on exertion?", "Has the cough resolved since the Z-pack?" — and tags each with the encounter type, the ICD-10/CPT codes involved, and the accept/edit/reject signal from the clinician.
- Questions surface as clickable prompts. The next time a clinician opens an encounter of the same type (e.g., "HTN follow-up"), the template presents the highest-signal questions as one-tap prompts. The clinician clicks a question to ask it; the answer flows into the SOAP note automatically. No typing, no dictation, no free-text guesswork.
-
A task is created to update the encounter type. When the system detects
new questions that were asked but are not yet in the encounter-type template, it
automatically creates a review task assigned to the practice manager or lead
clinician. The task title is explicit:
"Review 3 new questions from acute-visit encounters this week". The task links to the questions, their source encounters, and the accept/reject stats. No question enters the template without human review.
Where This Is Already Demoed
- Clinical Documentation demo — The ambient scribe fills bounded SOAP fields; the clinician edits, accepts, or rejects each field. This accept/edit/reject signal is the raw input to the Question Learning Loop. (Clinical Doc walkthrough)
- Coding / CDS demo — CPT and ICD-10 suggestions appear with confidence scores and Accept / Override buttons. Each action feeds back: accepted codes reinforce the question-to-code mapping; overrides signal that the template needs refinement.
Why Competitors Cannot Copy This
The loop requires three things competitors lack: (1) encounter-type templates that are versioned and practice-configurable, not static forms; (2) a global patient record so questions can reference cross-org clinical facts (medications from pharmacy A, labs from lab B); and (3) a task-management system that enforces human review before any question enters the template. Scribes transcribe. REV.health learns.
Key Capabilities
Real-Time CPT / ICD-10-CM Suggestions with Evidence
As the ambient scribe writes the SOAP note, the coding service emits a ranked list
of ICD-10-CM problems and CPT procedure candidates. Each suggestion is pinned to
the moment-in-transcript that triggered it — "Code suggested because: 'patient
reports chest pain on exertion radiating to the left arm' in HPI." The clinician
accepts, edits, or rejects each suggestion. Every suggestion carries an
EvidenceUriString pointing at the supporting transcript fragment or
document.
E/M Level Rationale
When the note reaches signing, the service proposes an E/M level (e.g., 99214) with an MDM-based justification: number and complexity of problems, amount of data reviewed, risk of complications. The clinician sees the proposed level and a downcode option (e.g., 99213) with rationale for each — defensible E/M levels, not upcoded ones. The E/M downcode rate target is ≤ 5% audit-traceable.
Code Optimizer
The code optimizer surface surfaces Dx/Proc pairings, payer-specific coding rules, and combo history to maximize legitimate reimbursement without upcoding. NCCI edit checking runs at code-acceptance time, flagging mutually exclusive or column-one/column-two code pairs. Modifier suggestions (-25, -59, -95, etc.) are offered with clinical context so the clinician understands why a modifier is needed before accepting it. Up to four modifiers per procedure are supported.
Clinical Decision Support Alerts
The CDS Hooks 2.0 service evaluates rules at chart-open,
problem-add, and order-sign trigger points. Four HTI-1 §170.315(b)(11) DSI
baseline checks fire on order-sign and
medication-prescribe:
- Drug-drug interaction — against the patient's global medication list.
- Drug-allergy — against the patient's global allergy list.
- Drug-disease — against active conditions across all contributing orgs.
- Dose-range — for the patient's renal function and weight.
Because clinical facts are global in REV.health, the CDS engine sees the union of every contributing org's medications, allergies, and conditions — not a single-org silhouette of the patient. Rule precision goes up; alert fatigue comes down. Override rate target: ≤ 60% per rule. Sustained overrides above threshold auto-enter the rule into managed review.
Audit-Ready Coding Documentation
Every coding suggestion, every override, and every accepted code carries full
provenance. CdsHookEvent rows capture the trigger, context, rule
version, suggestion cards returned, clinician action (accepted / overridden /
dismissed), structured override reason, and end-to-end evaluation latency. Overrides
are captured with structured reason codes (e.g., "patient already taking,
well-tolerated"), not free text. The audit trail traces every code back to the
clinical evidence that supports it.
Persona Connections
- Doctor — Primary coding reviewer. Accepts or overrides ICD-10/CPT suggestions with one click. Signs off on the E/M level. Wants accurate codes proposed inside the chart with evidence links, and CDS alerts that fire only when they should.
- RCM — Claim scrubbing downstream. The codes accepted in this module are the input to RCM charge capture, 837P generation, and remittance posting. The coding audit trail flows directly into the claim audit trail.
- Practice Manager — Owns the financial outcome. Tracks HCC capture rates, E/M distribution, MIPS Promoting Interoperability and Quality measures, and denial trails that trace back to the coding decision. Reads the dashboard, not the rule engine.
Technical Highlights
- NCCI edit checking. At code-acceptance time, the coding service validates CPT code pairs against the CMS National Correct Coding Initiative edits, flagging mutually exclusive and column-one/column-two pairs before the clinician signs. Modifier suggestions are context-aware and tied to the specific NCCI edit that triggered them.
-
Modifier suggestions. Up to four modifier slots per
Procedureentity. The coding service proposes modifiers based on the clinical scenario — e.g.,-25for a significant, separately identifiable E/M on the same day as a procedure,-59for a distinct procedural service. Each suggestion includes the supporting rationale. -
E/M level calculation engine. The service analyzes the encounter
content — problem count and complexity, data reviewed, risk of complications —
and maps to the AMA/CMS E/M leveling framework. The MDM rationale text is
persisted on the
Procedureentity asMdmRationaleText, available for audit defense. -
CDS Hooks 2.0. The service implements
patient-view,order-select,order-sign,medication-prescribe,problem-list-item-create, andencounter-dischargehook types per the CDS Hooks 2.0 spec. Rules are versioned withbreak.feature.bug.buildtimestampand tracked byRuleFamilyIDfor lineage. Active version is selected at evaluation time per(OrganizationID, RuleFamilyID, EffectiveDateTimeUTC).
Domain Entities at a Glance
-
Diagnosis— Global clinical fact. ICD-10-CM + SNOMED-CT dual-coded condition. CarriesSourceOrganizationID, HCC category mapping, and evidence URI. -
Procedure— Global clinical fact. CPT/HCPCS coded procedure with up to four modifiers, E/M level, MDM rationale, and AI-augmented flag for CY2026 CPT variants. -
OrderSet— Org-scoped. Practice-authored bundles of orders triggered by condition. Versioned, withTriggerIcd10CodeandTriggerSnomedCodefor auto-suggestion. -
CdsRule— Org-scoped, versioned. One immutable row per rule version. Hook type, severity, expression (CQL/JSON-Logic), DSI flag, and model-card URI. -
CdsHookEvent— Org-scoped audit log. One row per CDS evaluation: trigger, rule version, suggestion cards, clinician action, override reason, evaluation latency. -
CodeSet/CodeSetVersion/CodeMapping— Global reference data. Licensed code systems (ICD-10-CM, CPT, HCPCS, SNOMED-CT, LOINC, RxNorm) with dated versions and deprecated→current mappings.
2026–2027 Regulatory Window
- CY2026 AMA AI-augmented CPT codes. The first AI-augmented CPT codes ship January 1, 2026. The module supports the new variants natively, including the documentation language an auditor expects when an AI scribe or AI image-analysis tool was the basis for a billable element.
- HTI-1 §170.315(b)(11) DSI. Drug-drug, drug-allergy, drug-disease, and dose-range checks ship with model cards, training-data summaries, performance metrics, and bias notes per the DSI transparency framework.
- HTI-1 USCDI v3 baseline (March 2026). Care Plans, Problem List, Goals, and Health Concerns are USCDI v3 elements; CDS rules read and write USCDI-compliant structures.
- CDS Hooks 2.0. REV.health implements 2.0 directly; no 1.0 legacy.
Delivery Phases
Real-time ICD-10-CM and CPT suggestions emitted as the encounter is built, with evidence links. E/M level prediction with MDM rationale at note signing.
CdsRule versioning and the four DSI baseline checks
(drug-drug, drug-allergy, drug-disease, dose-range) are live on
order-sign. NCCI edit checking and modifier suggestions
fire at code-acceptance. Every fired hook produces a
CdsHookEvent audit row within 5 seconds. CDS evaluation
latency target: p95 ≤ 400 ms.
HCC capture hints surface at chart-open when a chronic condition documented in a prior year is missing from the active problem list. Order-set application lets the clinician apply a condition-tied bundle of orders in one action, with each order linked to the source
OrderSetID and version. CY2026 AI-augmented CPT
support detects when an AI scribe or AI image-analysis result was the basis for a
billable element and proposes the AI-augmented CPT variant. Override-rate alarm
triggers managed review when per-rule override exceeds
OverrideThresholdDecimal over a rolling 30-day window.
Specialty-specific CDS rule packs beyond primary care (pediatric immunization schedules, women's health screening, geriatric polypharmacy). Code-set versioning at write time — every
Diagnosis and Procedure stores
the CodeSetVersionID in force at coding time, with automatic
deprecated→current mapping on read. Retrospective coding-compliance analytics
and clinician coding-accuracy dashboards (tied to
Payer Optimization). DSI transparency surface
expanded with model-card drill-down and performance-metric visualizations.
Success Metrics
- Coding suggestion acceptance rate: ≥ 70% — below this threshold the model is noisy and ignored.
- HCC capture rate: ≥ 90% vs. prior-year baseline — risk-adjusted contracts pay correctly when documented HCCs are re-surfaced annually.
- E/M downcode rate: ≤ 5% audit-traceable — defensible levels, not upcoded.
- CDS alert override rate: ≤ 60% per rule — sustained overrides above threshold trigger automatic rule review.
- CDS evaluation latency: p95 ≤ 400 ms from hook-fire to suggestion-card render.
- DDI / drug-allergy true-positive rate: ≥ 95% on labeled DDI test set — HTI-1 DSI baseline.
Module Dependencies
- Upstream — Clinical Documentation emits encounter content the coding service consumes.
- Lateral — eRx / EPCS orders fire the
order-signCDS hook; the medication list is read by drug-drug and drug-allergy checks. - Downstream — RCM consumes accepted codes for claim scrubbing and 837 generation.
- External — Terminology server for code-set hosting and licensing (ICD-10-CM, CPT, HCPCS, SNOMED-CT US Edition, LOINC, RxNorm).