Home / Platform / Coding & CDS

Module 04 · Coding & clinical decision support

Code the encounter while you document it.

The right ICD-10, CPT, modifier, and HCC hint — the moment the evidence appears, linked to its source. You sign, not the system.

The signature mechanic

The E/M level shows its work. The safety alert shows its source.

Two things every coding tool gets wrong: it asserts an E/M level without the medical-decision-making derivation, and it fires drug alerts you can't see the basis for. rev.health derives the 2021 MDM grid in the open — problems, data, risk — and every CDS Hooks card names its rule, its evidence, and offers a structured override.

E/M level prediction with MDM rationale and a CDS Hooks suggestion card On the left, the three medical-decision-making elements (number and complexity of problems, amount of data reviewed, risk of management) each scored at a level, combining to a predicted E/M code of 99214 moderate, with a 99213 downcode option shown. On the right, a CDS Hooks 2.0 suggestion card naming the rule id, the drug interaction, the evidence, and an override-with-reason control. E/M LEVEL — MDM DERIVATION (2021 guidelines) Problems addressed 2 chronic illnesses with progression (HTN, T2DM-CKD) Moderate Data reviewed & analyzed HbA1c + home BP log reviewed; external records reconciled Moderate Risk of management Prescription drug management (titration) Moderate 2 of 3 elements at Moderate → PREDICTED E/M 99214 moderate complexity DOWNCODE OPTION 99213 if risk reads as low — clinician decides, logged CDS HOOKS 2.0 — order-sign HOOK ! Hyperkalemia risk lisinopril + spironolactone RULE cds.rh.drug-drug.k-sparing v2.4.1 SEVERITY warning — not hard-stop EVIDENCE global med list (spironolactone added by another practice, Apr 2026) · K+ 5.1 last draw SOURCE / MODEL CARD rule provenance + version shown in-context Override ▾   "patient tolerates, K+ trended, monitoring q2wk" Adjust order Override w/ reason Override reasons are structured, not free-text — they feed rule tuning. Rendered in 312 ms, before your hand returns to the keyboard.

Because the medication list is global across every practice the patient visits, the interaction check sees the drug another clinic prescribed last month — the difference between a useful warning and a noisy one. Override rates are tracked per rule, per version, and a rule that exceeds its threshold enters automatic review.

The problem

Revenue and time leak at the coding boundary.

E/M downcoding from missed MDM elements is routine. HCC capture is uneven, so risk-adjusted contracts underpay. Bolt-on coding tools sit outside the chart and add reconciliation work. And legacy drug-checking pop-ups are tuned so coarsely that clinicians override nearly every alert — eroding the safety value of all of them.

Coding and CDS as one surface

The module listens to the encounter as it is built — scribe text, problem-list edits, vitals, labs, orders — and returns coded suggestions: ICD-10-CM, CPT, HCPCS, an E/M level proposal with MDM-based rationale, and HCC capture hints. Every suggestion carries an evidence link (“suggested because: 'chest pain on exertion radiating to the left arm' in HPI”). Accept, edit, or reject — the disposition is logged.

Whole-patient CDS, not single-org CDS

Because clinical data in rev.health is global across every practice the patient visits, the CDS engine sees the full medication list, problem list, allergy list, and care plans — not one org's silhouette of the patient. That's the difference between a useful warning and a noisy one: rule precision goes up, alert fatigue comes down, and override rates are tracked per rule, per version, with automatic review when they climb.

Key capabilities

Clinician-in-the-loop, evidence-linked, auditable.

Real-time coding assistant

Ranked ICD-10-CM and CPT candidates emitted while the note is written, each pinned to the transcript moment or chart fact that triggered it.

E/M with MDM rationale

A defensible E/M proposal at signing — problems, data reviewed, risk — with the downcode option and its rationale shown alongside. You choose; the audit trail remembers.

HCC capture hints

Conditions documented in prior years that map to HCC categories surface for re-documentation. Nothing is auto-coded; the clinician confirms or dismisses.

CDS Hooks 2.0 safety checks

Drug-drug, drug-allergy, drug-disease, and dose-range checks at chart-open, problem-add, and order-sign — rendered in under 400 ms with rule ID, source evidence, and one-click structured override.

Primary-care order sets

Add “Type 2 diabetes” to the problem list and the matching order bundle offers itself: HbA1c q90d, annual eye exam, foot exam, microalbumin. Apply, edit, or skip.

CY2026 AI-augmented CPT

The first AMA AI-augmented CPT codes require documentation of AI involvement. rev.health detects when an AI result underpins a billable element and proposes the right variant with auditor-ready language.

DSI transparency by default

Every AI suggestion carries an in-context “why”: model name, version, training-data summary, performance metrics, known limitations — the HTI-1 §170.315(b)(11) fields, not a buried docs page.

Versioned, override-aware rules

Rules that exceed configured override thresholds enter a managed review state automatically. Alert fatigue is an engineering metric here, not a fact of life.

Structured override reasons

Overrides capture structured reasons (“patient already taking, well-tolerated”) that feed rule tuning — no free-text-only prompts shouting into the void.

Workflow

From spoken word to defensible code.

The scribe writes; the coder listens

As the structured A/P forms, ranked diagnosis and procedure candidates appear with their evidence links.

Safety checks fire at order-sign

The new prescription is cross-checked against the patient's global med list — including the one another practice prescribed last month.

E/M proposed at signing

99214 with the MDM math shown, 99213 offered as the downcode with its own rationale. The clinician picks; the choice and the basis are logged.

HCC hints close the year

The diabetes-with-CKD documented last September resurfaces for annual re-documentation before the risk-adjustment window closes.

Codes flow to the claim

Accepted codes feed the RCM scrubber directly — the same data, no re-keying, no retrospective coding pass.

Who benefits

Clinician

Accurate codes proposed inside the chart with evidence, an E/M level they can defend, and alerts that fire only when they should.

Coder / billing specialist

A full audit trail of every suggestion accepted, edited, or rejected — and a one-screen queue of encounters needing a coder touch before scrubbing.

Practice owner

HCC capture at parity with risk-adjusted contracts and an E/M distribution that matches case-mix without upcoding exposure.

Compliance lead

Versioned rules, override-reason analytics, and model-card transparency for every AI-driven suggestion.

Incumbents code retrospectively from the signed note. rev.health codes alongside the note, with each suggestion pinned to the moment in the transcript that triggered it.

Performance targets

The numbers this module is built to hit.

MetricTarget
Coding suggestion acceptance rate≥ 70%
HCC capture rate vs. prior-year baseline≥ 90%
E/M downcode rate (audit-traceable)≤ 5%
CDS alert override rate per rule≤ 60% — above triggers automatic rule review
Order-sign to CDS card render≤ 400 ms p95
Drug-drug / drug-allergy true-positive rate≥ 95% on labeled test sets — release-gating

Standards & code systems

The terminologies and the transparency rules.

ICD-10-CM CPT / HCPCS E/M 2021 MDM grid CMS-HCC risk adjustment SNOMED-CT CDS Hooks 2.0 FHIR Clinical Reasoning CY2026 AI-augmented CPT HTI-1 DSI transparency §170.315(b)(11)

CDS fires at three hook points — chart-open, problem-add, and order-sign — with rule provenance and a model card surfaced in-context. The CY2026 AMA AI-augmented CPT codes require documentation of AI involvement; rev.health detects when an AI result underpins a billable element and proposes the right variant with auditor-ready language.

Defensible codes, captured in the room.

Evidence-linked coding and CDS that clinicians actually trust.

Join the waitlist