Scribing.io vs. Nabla: The Clinical Accuracy Audit

How Specialty-Tuned Pertinent Negative Capture Closes the DVT Documentation Gap That Generic AI Scribes Miss

TL;DR: Generic ambient AI scribes—including Nabla Copilot—benchmark "accuracy" as transcript fidelity, but they systematically ignore the documentation mechanics that drive claim denials and audit failures. For DVT workups specifically, missing pertinent negatives tied to the Wells score (no calf tenderness, no asymmetric swelling, no pitting edema) cause 20–25% denial rates on venous duplex (CPT 93970/93971) and same-day E/M + ultrasound pairs billed with Modifier 25. Scribing.io's specialty-tuned clinical logic explicitly prompts, records, and maps each pertinent negative as discrete EHR data linked to ICD-10 symptom codes—reducing denials below 3%, defending Modifier 25 on post-pay review, and creating a $40K+ annual revenue recovery for a typical 4-site urgent care group. This playbook details the exact clinical logic, ICD-10 mapping, and audit defense architecture that CMIOs need to evaluate.

Contents

• The Accuracy Illusion: Why Transcript Fidelity Is Not Clinical Accuracy

• Why Competitors Benchmark Accuracy but Ignore Denial Mechanics

• Scribing.io Clinical Logic: DVT Workup From Denial Exposure to Audit Defense

• Step-by-Step Logic Breakdown: How Scribing.io Solves the DVT Documentation Gap

• Technical Reference: ICD-10 Documentation Standards for DVT Workup

• Modifier 25 Defense Architecture: Structural Separation That Survives Post-Pay Review

• EHR Integration: Discrete Data Flow From Encounter to Claim

• Head-to-Head: Scribing.io vs. Nabla on DVT Documentation Mechanics

• Book a 15-Minute Workflow Audit

The Accuracy Illusion: Why Transcript Fidelity Is Not Clinical Accuracy

The ambient AI scribe market in 2026 has consolidated around a single benchmark: transcription accuracy. Heidi Health publishes 2.5 million weekly patient visits. Nabla Copilot reports 380,000. Both platforms cite speed metrics and adoption rates as proxies for quality. Heidi's own comparison page notes that Nabla sometimes generates "clinical details, such as findings that weren't actually discussed during the encounter."

Both Heidi and Nabla—and the broader market—share a blind spot that neither addresses in public documentation: the absence of a finding is clinically as important as its presence, and generic AI models are structurally incapable of prompting for what was never said. Scribing.io exists to close that exact gap—not by transcribing faster, but by understanding what a note requires based on the clinical context and ensuring it gets there.

A clinician performing a DVT workup who states "patient has left calf pain and swelling" has documented two positives. What they have not documented—and what a generic scribe cannot infer or prompt—are the remaining Wells criteria findings: Was there pitting edema? Was there asymmetric swelling greater than 3 cm? Was there tenderness along the deep venous system? Was there a history of prior DVT? Were alternative diagnoses equally likely? The modified Wells criteria require documentation of both positive and negative elements to produce a valid, defensible score.

The JAMA clinical guidelines on venous thromboembolism establish that risk stratification scores must be supported by individually documented criteria—not summary assertions. A note containing "Wells score 3" without the underlying element-by-element documentation is an unsupported assertion that fails audit. This principle applies whether the documentation is produced by a human scribe, a generic AI, or a physician typing directly.

Current clinical benchmarks from the CMS Medical Record Integrity program indicate that pertinent negatives are omitted from AI-generated notes at rates between 18% and 30% depending on specialty and clinical scenario. In DVT workups specifically, the Wells score requires documentation of both positive and negative criteria to establish medical necessity for venous duplex imaging and to support the E/M level billed alongside the ultrasound.

This is not a transcription problem. It is a clinical logic problem. A scribe that perfectly transcribes what was said will still produce a note that fails audit if the clinician never verbalized the pertinent negatives—because no one prompted them to. For organizations running athenahealth or Epic Integration workflows, the scribe must write discrete, structured data into the EHR—not just narrative text that looks correct on screen but is invisible to payer adjudication engines.

Why Competitors Benchmark Accuracy but Ignore Denial Mechanics

The foundational insight that separates Scribing.io's approach from every other ambient AI scribe: the real cost of documentation failure is not measured in transcript errors—it is measured in claim denials, audit clawbacks, and downstream liability exposure.

Consider the revenue cycle anatomy for a suspected DVT encounter at an urgent care facility:

When a payer audits a 99214-25 + 93971 claim pair, the reviewer asks two questions:

1. Was the E/M service separately identifiable from the procedure? This requires documented clinical reasoning—including the pertinent negatives that informed the decision to order imaging rather than treat empirically or discharge. The AMA's Modifier 25 guidance is explicit: the E/M must stand on its own documentation, not borrow from the procedure note.

2. Was venous duplex medically necessary? The Wells DVT score is the de facto standard. If the note shows a Wells score of ≥2 but the underlying criteria are not individually documented, the payer can—and routinely does—deny the imaging claim or downcode the E/M.

Here is where Nabla's generic model creates a specific, quantifiable gap:

• Nabla captures what is said. If the clinician says "Wells score is 3," Nabla may transcribe it. But without discrete documentation of which criteria were positive and which were negative, the Wells score is an unsupported assertion—not auditable evidence.

• Nabla does not prompt for omissions. If the clinician forgets to verbalize "no pitting edema" or "no history of prior DVT," Nabla does not flag the gap. The note ships incomplete.

• Nabla stores findings as free text. Even when pertinent negatives are verbalized, they are embedded in narrative paragraphs—not mapped to discrete data fields that payer systems, audit algorithms, and denial management tools can parse.

The result: venous duplex claims denied for insufficient medical necessity documentation. Modifier 25 stripped on post-pay review because the E/M note lacks the separately identifiable clinical decision-making. The audit trail—when it matters most—consists of unstructured prose that requires manual review to defend.

Scribing.io Clinical Logic: DVT Workup From Denial Exposure to Audit Defense

Before: The Cost of Generic Documentation

A 4-site urgent care group evaluates approximately 70 suspected DVT visits per month. With a generic ambient scribe (Nabla, Heidi, or similar), the documentation workflow produces the following outcomes:

The $4,300 monthly figure is conservative. It excludes staff time on denial appeals, clinician minutes rewriting notes, and the compounding risk of a targeted audit triggered by a pattern of unsupported 93970/93971 claims. The OIG Work Plan has flagged same-day E/M + imaging pairs at urgent care facilities as a focus area for multiple consecutive years.

After: Scribing.io Specialty-Tuned DVT Logic

Step-by-Step Logic Breakdown: How Scribing.io Solves the DVT Documentation Gap

This is the granular clinical logic that separates a specialty-tuned documentation engine from a passive transcript generator. Each step addresses a specific failure mode observed in generic AI scribe outputs.

Step 1: Context Detection and Wells Criteria Prompt Activation

Scribing.io's clinical logic engine monitors the encounter in real time for DVT workup context signals: chief complaint mentions of leg pain, leg swelling, or calf tenderness; verbal differential including "rule out DVT" or "suspected deep vein thrombosis"; or order entry for venous duplex (93970/93971). Any of these triggers activates the Wells-specific documentation pathway.

Once activated, each of the eight modified Wells criteria appears as a discrete documentation field within the scribe interface. This is not a static template dropped into the note—it is a dynamic checklist that persists until every criterion is addressed.

Why this matters: Nabla and other generic scribes have no concept of "what should be documented for this clinical scenario." They process audio and produce text. Scribing.io processes audio and applies clinical documentation rules to identify what is missing.

Step 2: Pertinent Negative Prompting for Unverbalized Findings

As the encounter progresses, Scribing.io auto-populates Wells criteria that the clinician verbalizes. Criteria that are not mentioned—the pertinent negatives—are surfaced as prompts. Depending on the EHR integration mode:

• Screen-based prompt: An unobtrusive notification appears on the scribe dashboard: "Wells criterion not documented: pitting edema (present/absent)?"

• Audio cue (ambient mode): A brief, clinician-configurable audio prompt during a natural pause: "Pitting edema status?"

• Post-encounter checklist: Before note finalization, any undocumented criteria are presented as a one-screen attestation requiring explicit positive/negative selection.

Why this matters: The CMS documentation integrity standards require that medical necessity be documented at the time of service. A retrospective addendum stating "Wells score components were assessed" does not carry the same evidentiary weight as contemporaneous, discrete documentation. Scribing.io ensures the documentation happens during the encounter, not after a denial.

Step 3: Discrete Data Capture With Laterality and Specificity

Every Wells criterion—positive and negative—is recorded as a structured data element, not free text. The difference is fundamental:

"No calf tenderness" becomes a codeable finding linked to the physical exam section with laterality. "No asymmetric swelling >3 cm" is captured with laterality and measurement context. These discrete elements flow directly into the EHR's structured data layer, where they are accessible to billing, compliance, and analytics systems without manual extraction.

Step 4: Order Linkage — The Medical Necessity Chain

The documented Wells score and its constituent findings are automatically linked to the venous duplex order in the Assessment & Plan section. This creates an unbroken chain: clinical finding → Wells criterion → aggregate score → imaging order → CPT code → ICD-10 diagnosis code.

When a payer requests documentation supporting a 93971 claim, this chain is instantly auditable. There is no need to read through paragraphs of narrative to find the clinical justification. The linkage is structural, not inferential.

Step 5: One-Tap Attestation and Read-Back Optimization

The clinician reviews the structured Wells summary—every criterion displayed with its positive/negative status, the calculated score, and the linked order—and attests with a single interaction. Read-back time drops to under 30 seconds because the note is pre-organized around the exact elements payers audit.

Compare this to the generic scribe workflow: the clinician must read the entire narrative note, mentally verify that each Wells criterion was captured, check for laterality errors, confirm that the score matches the documented criteria, and ensure that the A/P links the findings to the order. At 90–180 seconds per note across 70 DVT visits per month, that is 105–210 minutes of clinician time monthly—purely on verification of documentation that should have been correct from the start.

Step 6: Modifier 25 Defense Architecture

The E/M documentation is automatically structured to demonstrate separately identifiable service. The clinical evaluation (history, exam, medical decision-making including Wells scoring) is documented as a distinct section from the procedural component (ultrasound performance and interpretation). This structural separation is not cosmetic formatting—it is the architectural basis for surviving post-pay review.

Scribing.io inserts a clear documentation boundary: the E/M note contains the clinical reasoning that led to the decision to order imaging; the procedure note contains the imaging execution and results. Per the AMA CPT Editorial Panel guidance, this separation is what defines a "separately identifiable" E/M service when billed with Modifier 25.

Technical Reference: ICD-10 Documentation Standards for DVT Workup

Precise ICD-10 coding is central to the documentation problem, not peripheral to it. When pertinent negatives are captured as discrete data, they can be mapped to symptom-level ICD-10 codes that support medical necessity and enable automated compliance checks. When they exist only as free text, they are invisible to payer adjudication engines.

The following codes represent the core ICD-10 documentation framework for DVT workup encounters. Scribing.io's specialty-tuned logic ensures each code reaches maximum specificity—including laterality, acuity, and anatomic site—to prevent denials caused by unspecified or insufficiently specific coding:

Symptom-Level Codes Supporting Medical Necessity for Imaging

M79.661 - Pain in right lower leg; M79.662 - Pain in left lower leg; M79.669 - Pain in unspecified lower leg; R22.41 - Localized swelling — These symptom codes document the presenting complaint with laterality. Scribing.io's discrete capture ensures that "left calf pain" maps to M79.662, not the unspecified M79.669. The unspecified code triggers payer edits at higher rates because it fails to demonstrate that the clinician assessed and documented laterality—a basic quality indicator that payers use as a proxy for documentation thoroughness.

mass and lump — Localized swelling and mass findings require anatomic specificity. When a clinician documents "swelling in the left calf," Scribing.io maps this to the laterality-specific code, not a generic finding.

right lower limb; R22.42 - Localized swelling and mass and lump — Right-sided findings receive equivalent laterality-specific mapping. The discrete data capture ensures that bilateral presentations are coded with both laterality codes rather than a single unspecified code.

Diagnosis-Level Codes for Confirmed or Suspected DVT

left lower limb; I82.401 - Acute embolism and thrombosis of unspecified deep veins of right lower extremity; I82.402 - Acute embolism and thrombosis of unspecified deep veins of left lower extremity; Z86.718 - Personal history of other venous thrombosis and embolism — These codes represent the diagnostic endpoint. I82.401 and I82.402 are used when DVT is confirmed; Z86.718 is critical for the Wells criteria element "previously documented DVT" and must be captured as a discrete historical finding, not buried in the narrative social/medical history.

Scribing.io's logic ensures that Z86.718 is queried during the Wells criteria assessment, not left to the clinician's memory. If the patient's problem list or prior encounter data contains a VTE history, it is surfaced during the documentation workflow. If the patient denies prior DVT, that denial is captured as a discrete negative finding linked to the Wells scoring element—supporting a lower Wells score and, in turn, the clinical rationale for the chosen management pathway.

The CMS ICD-10-CM Official Guidelines require that codes be assigned to the highest level of specificity supported by the documentation. Generic scribes that capture "leg pain" without laterality force coders to select unspecified codes, which increases denial risk and signals documentation deficiency to payer audit algorithms.

Modifier 25 Defense Architecture: Structural Separation That Survives Post-Pay Review

Modifier 25 is the single most audited modifier in outpatient medicine. The OIG and commercial payers systematically review same-day E/M + procedure pairs, particularly in urgent care settings where imaging is performed on-site. The audit question is always the same: does the documentation support that the E/M was a separately identifiable service?

Scribing.io's architecture addresses this at the structural level:

• Section isolation: The E/M components (HPI, ROS, physical exam, MDM including Wells scoring and differential diagnosis) are documented in sections that are structurally distinct from the procedure note (ultrasound indication, technique, findings, interpretation).

• MDM complexity anchoring: The Wells score documentation—with its discrete positive and negative criteria—directly supports moderate medical decision-making complexity. The number of diagnoses addressed, the data reviewed (prior imaging, D-dimer results, Wells calculation), and the risk of the management option (anticoagulation initiation vs. discharge) are each documented as discrete MDM elements.

• Temporal separation markers: Scribing.io timestamps the clinical evaluation and the procedure as separate workflow events, providing temporal evidence that the E/M occurred independently.

Generic scribes produce a single narrative that blends the clinical evaluation with the procedure. On audit, the reviewer cannot distinguish where the E/M ends and the procedure begins. This is the documentation pattern that causes Modifier 25 to be stripped.

EHR Integration: Discrete Data Flow From Encounter to Claim

The documentation architecture described above is only valuable if it flows into the EHR as structured data. Scribing.io integrates with major EHR platforms to ensure that discrete findings, Wells criteria, and linked orders populate the correct fields:

For organizations on athenahealth, the discrete data elements map to athena's structured clinical fields and flow through to the claim without manual coding intervention. For Epic environments, Scribing.io's integration uses the structured data API to populate SmartData Elements that are visible to Epic's charge capture and coding workflows. Detailed integration steps are available in our athenahealth integration guide and Epic integration guide.

Head-to-Head: Scribing.io vs. Nabla on DVT Documentation Mechanics

Book a 15-Minute Workflow Audit

Denial rates on DVT imaging and same-day E/M pairs are not abstract—they are measurable on your current claims data. Scribing.io offers a 15-minute Workflow Audit designed specifically for CMIOs and revenue cycle leaders evaluating ambient AI scribes:

• Payer-mapped Pertinent Negative coverage report: We analyze your current DVT documentation against every Wells criterion and show exactly which negatives your notes miss.

• Field-level denial mapping: We identify which discrete data fields are absent or free-text-only in your current EHR workflow, and which ones are blocking clean submission of 93971 and 99214+25 claims.

• Live fix demonstration: We run a DVT workup scenario in your EHR environment and show the discrete negative capture, Wells score linkage, and Modifier 25 defense architecture writing structured data into your system in real time.

This is not a sales demo. It is a diagnostic audit of your documentation workflow with a deliverable report you keep regardless of vendor decision. Book your 15-minute Workflow Audit at Scribing.io and get the payer-mapped report within 48 hours.

The question for CMIOs is no longer "which AI scribe transcribes fastest." It is: does the note get paid, and does it survive audit? Every other metric is vanity. Scribing.io is built to answer the question that matters.