AI Documentation for OB-GYN Providers: Global OB Accuracy and the Split-Visit Billing Engine

Author: Lead Clinical Consultant, Scribing.io | Audience: OB-GYN Medical Directors, Coding Compliance Officers, Revenue Cycle Leaders | Last Updated: January 2026

TL;DR — Why This Guide Exists

Most AI scribes treat a prenatal visit as a single note. When your 22-week patient mentions dysuria during a routine check, that single-note approach bundles the UTI workup into the global OB fee — and payers claw back the E/M, deny the culture, and trigger a recoupment averaging $1,250 per incident. This guide is the definitive clinical-library reference for OB-GYN Medical Directors who need AI documentation that automatically enforces the split-visit distinction between prenatal wellness and acute problem management, exports dual FHIR R4 claim lines with the correct -25 modifier, and passes NCCI edits on first submission. If you manage a global OB panel and want zero split-visit denials, read every section — then see Scribing.io Pricing to run a proof of concept in your practice.

Contents

• The Split-Visit Trap: Why Global OB Billing Fails Without AI-Enforced Segmentation

• What Competitors Miss: Payer OB Global Accumulators, FHIR Claim-Line Modifiers, and the Information-Gain Gap

• Scribing.io Clinical Logic: Handling the 22-Week Prenatal Visit with Acute UTI

• Technical Reference: ICD-10 Documentation Standards for OB-GYN Split Visits

• NCCI/PTP Edits, Modifier -25, and the Medicaid TH Routing Problem

• FHIR R4 Dual-Encounter Export Architecture

• Implementation Checklist for Medical Directors

• Book a 15-Minute Demo: Global OB Split-Visit Guardrail

The Split-Visit Trap: Why Global OB Billing Fails Without AI-Enforced Segmentation

The economic backbone of obstetric care is the global OB package — a single bundled fee covering approximately 13 antepartum visits, delivery, and postpartum care. Payers track this bundle through an OB global accumulator: a running tally of visits mapped to supervision-of-pregnancy codes (Z34.xx, Z3A.xx) that counts toward the all-inclusive fee. The ACOG coding guidance outlines this structure clearly, yet its implications for AI documentation are universally ignored by scribe vendors.

The trap is deceptively simple. A patient arrives for a routine 22-week prenatal visit. She also mentions burning on urination and a low-grade fever. The clinician documents both concerns in a single encounter. The coder assigns Z34.90. The note disappears into the global accumulator. No separate E/M is generated. No -25 modifier is appended. The urine culture, the sensitivity panel, and the antibiotic prescription are absorbed into the global fee — services the payer never agreed to cover under that bundle.

When the payer audits — and post-payment audits of OB global claims have increased over 30% since 2023 according to HHS OIG work plan priorities — the result is predictable:

• Recoupment of the separately billable E/M and lab charges, averaging $1,250 per incident across commercial and Medicaid payers.

• Denial of the urine culture and any associated diagnostics retroactively linked to the global line.

• Potential compliance flags for unbundling errors if the practice later tries to retroactively append -25, which the AMA CPT guidelines expressly prohibit as post-hoc modifier assignment.

The core failure is not clinical. The physician performed a perfectly appropriate evaluation of an acute problem. The failure is documentary architecture: the AI scribe — or the human scribe — produced one note where two were required.

The Anchor Truth: AI must distinguish between "Prenatal Wellness" and "Acute Problem" discussion (e.g., UTI) to justify the -25 modifier and prevent global billing rejections. Any documentation platform that treats a prenatal encounter as a monolithic artifact is structurally incapable of protecting OB-GYN revenue. This is the foundational design principle behind Scribing.io's split-visit engine — and it is the reason we built this playbook.

What Competitors Miss: Payer OB Global Accumulators, FHIR Claim-Line Modifiers, and the Information-Gain Gap

Reviewing the competitive landscape in AI scribe solutions marketed to OB-GYN providers reveals a consistent pattern: platforms emphasize visit-type flexibility ("procedures, evals, or well-woman exams"), EHR push capabilities, and time savings. These are table-stakes features. What they do not address — and what constitutes the largest single source of preventable OB revenue loss — is the interaction between three technical systems:

Most guides and competitor platforms treat "ICD-10 code generation" as the endpoint. In reality, code generation without claim-architecture awareness is the root cause of split-visit denials. A platform that generates O23.42 but packages it on the same claim line as Z34.90 — or fails to enforce -25 at the FHIR Claim.item.modifier level — has not solved the problem.

Scribing.io's original insight is that the AI documentation engine must operate at three layers simultaneously:

1. Clinical segmentation — real-time detection that the conversation has shifted from wellness to acute problem.

2. Documentary bifurcation — generation of two structurally independent notes, each satisfying current E/M documentation rules per the AMA E/M descriptors.

3. Claim-line export — FHIR R4 output where each note maps to a distinct Claim.item with correct primary diagnosis sequencing and modifier injection.

No other platform in the OB-GYN AI scribe market publicly documents this three-layer architecture. That is the information gap this guide exists to fill.

For a parallel example of how Scribing.io handles complex modifier logic in other specialties, see our deep dives on Psychiatry (where -25 surfaces in combined therapy/E/M sessions) and Cardiology (where modifier -59 and NCCI edits govern stress-test/E/M bundling). The underlying engine is the same; the specialty-specific rules differ.

Scribing.io Clinical Logic: Handling the 22-Week Prenatal Visit with Acute UTI

The scenario: A 28-year-old G2P1 at 22 weeks arrives for a routine prenatal check but reports dysuria and fever.

The failure state: The chart is saved as a single global OB visit with Z34.xx and no -25. The UTI workup, antibiotic, and E/M are bundled into the global, triggering a $1,250 recoupment and denial of the urine culture.

The Scribing.io resolution: The conversation is auto-segmented in real time. A prenatal wellness note feeds the global accumulator, while a distinct problem E/M note is generated with -25, primary O23.42 and Z3A.22, linked labs and Rx, and a payer-ready justification paragraph. The claim exports as dual lines, passes NCCI edits, and pays on first submission. Here is exactly how the engine processes this encounter, step by step.

Step 1 — Real-Time Conversational Segmentation

Scribing.io's ambient listener maintains a running classification model that tags conversational segments into two buckets:

• Global OB Wellness — Fundal height, fetal heart tones, weight, blood pressure, routine lab review, anticipatory guidance, fetal movement inquiry.

• Acute Problem — Any new symptom with its own HPI arc (onset, location, duration, severity, associated symptoms), physical exam findings, diagnostic orders, or treatment decisions.

The moment the patient says "I've been having burning when I urinate for the last three days, and I had a fever last night", the engine opens a parallel documentation stream. This is not a retroactive tag applied after the visit. It is a real-time fork in the documentation graph. The segmentation model is trained on OB-specific conversational patterns and weighted toward high-recall detection of acute complaints — missing a segmentation trigger is far more costly than generating a false positive that a clinician reviews and merges back.

The technical distinction matters: generic NLP models classify sentences. Scribing.io classifies clinical episodes — contiguous spans of dialogue that share a chief complaint, an assessment arc, and a plan. An episode boundary is triggered by onset language, new-symptom language, or diagnostic-order language that cannot be mapped to the existing global OB plan.

Step 2 — Dual Note Generation with Distinct E/M Elements

Two exportable, structurally independent artifacts are generated. Each note satisfies the CMS E/M documentation requirements independently:

Each note is a self-contained clinical document. An auditor reviewing Note 2 in isolation — without ever seeing Note 1 — can identify a distinct chief complaint, a complete HPI, a focused exam, moderate-complexity MDM, and a treatment plan. This is what "separately identifiable" means under the AMA's -25 modifier definition, and it is the standard that post-payment auditors apply.

Step 3 — Payer-Ready Justification Paragraph

Scribing.io auto-generates a medical-necessity justification block appended to Note 2:

"This encounter included a separately identifiable evaluation and management service for an acute genitourinary infection complicating pregnancy (O23.42). The patient presented with a distinct chief complaint (dysuria, fever), a separately documented history of present illness, focused physical examination, and moderate-complexity medical decision-making including prescription drug management with pregnancy-specific antibiotic selection. This service is distinct from the routine prenatal care visit documented concurrently and meets the criteria for modifier -25 per CMS and ACOG guidelines for services beyond the scope of the global OB package."

This paragraph is not decorative. It is the single most effective defense against post-payment recoupment. Auditors from UnitedHealthcare, Aetna, and Medicaid MCOs have published denial rationale templates that specifically look for language establishing "separately identifiable" service. The justification paragraph maps directly to those criteria.

Step 4 — FHIR R4 Dual Claim-Line Export

The claim exports as two Claim.item resources:

• Claim.item[0]: CPT 59426 (antepartum care, global increment), Dx Z34.90 + Z3A.22, no modifier. Routed to the payer's OB global accumulator.

• Claim.item[1]: CPT 99214 (established patient, moderate complexity), Dx O23.42 (primary) + Z3A.22, Claim.item.modifier = -25. Routed as a separately payable line.

Critical guardrail: The engine prevents Z34.xx from appearing as the primary diagnosis on the problem E/M line. If Z34 is primary on the problem line, the accumulator absorbs the charge and the E/M is denied as a duplicate. This single logic rule — enforced at the claim-export layer, not the note layer — eliminates the most common cause of split-visit denials. The urine culture (CPT 87086) and urinalysis (CPT 81001) are linked to Claim.item[1] via Claim.item.diagnosisSequence, ensuring lab payment is tied to the problem diagnosis, not the global.

Technical Reference: ICD-10 Documentation Standards for OB-GYN Split Visits

Accurate ICD-10 sequencing is the difference between a paid claim and a denial. This section provides the reference framework for the codes most commonly involved in split-visit OB encounters and documents how Scribing.io ensures maximum specificity.

Z34.90 — Encounter for Supervision of Normal Pregnancy

Z34.90 — Encounter for supervision of normal pregnancy is the default code for routine prenatal care when the pregnancy is uncomplicated. Key documentation rules:

• Usage: Primary diagnosis for all routine prenatal visits feeding the OB global accumulator. Must never appear as primary on a separately billed problem E/M line.

• Trimester specificity: Z34.90 is the unspecified trimester variant. Best practice — and Scribing.io's default behavior — is to use trimester-specific variants (Z34.01 for first trimester, Z34.02 for second, Z34.03 for third). The engine calculates trimester from the EDD or LMP stored in the patient's OB flowsheet and auto-selects the highest-specificity code. Unspecified trimester codes trigger audit flags at multiple commercial payers and should only appear when gestational age is genuinely unknown.

• Pairing with Z3A.xx: Per CMS ICD-10-CM Official Guidelines, Z3A.xx (weeks of gestation) should be listed as an additional code whenever the gestational week is documented. Scribing.io calculates Z3A.xx from the same EDD/LMP source and auto-appends it as a secondary diagnosis on both the global and problem lines.

O23.40 and O23.42 — Infections of the Genitourinary Tract in Pregnancy

O23.40 — Infections of genitourinary tract in pregnancy, unspecified trimester is the parent code. For the split-visit scenario, Scribing.io enforces the following specificity cascade:

1. Site specificity: O23.4x is the "unspecified" urinary tract site code. If the documentation supports urethral involvement (dysuria without upper-tract signs), the engine prefers O23.42 (infections of urethra in pregnancy). If flank pain or systemic signs suggest pyelonephritis, O23.02 (infections of kidney in pregnancy, second trimester) is selected instead.

2. Trimester specificity: The fifth character encodes trimester. O23.40 (unspecified trimester) will be rejected or flagged by payers when gestational age is documented. Scribing.io maps the calculated trimester to the correct fifth character: 1 (first), 2 (second), 3 (third), or 9 (unspecified). For our 22-week patient, this resolves to O23.42 (second trimester) without coder intervention.

3. Sequencing: On the problem E/M line, O23.42 must be primary. Z3A.22 is secondary. Z34.xx does not appear on this line. This sequencing prevents the accumulator from absorbing the charge and ensures the claim reads as a problem visit to the adjudication engine.

Hyperlipidemia and Other Comorbidities on OB Encounters

Occasionally, OB patients have pre-existing conditions managed at the same visit. Codes like E78.5 — Hyperlipidemia, unspecified may appear in the problem list. Scribing.io handles these with the same split-visit logic: if a pre-existing condition is actively managed (medication adjustment, new lab order, counseling documented), it generates a distinct problem E/M line. If it is merely listed in the problem list without active management during the encounter, it is excluded from the claim to avoid audit exposure. This prevents the common "code-stuffing" pattern that triggers NCCI edit failures.

NCCI/PTP Edits, Modifier -25, and the Medicaid TH Routing Problem

The CMS NCCI (National Correct Coding Initiative) Procedure-to-Procedure (PTP) edit pairs define which CPT codes can and cannot be billed together on the same date of service. For OB split visits, the relevant edit pairs are:

The Medicaid TH Routing Problem

Medicaid Managed Care Organizations (MCOs) in 23 states require the TH modifier (obstetric treatment/services, prenatal or postpartum) on global OB claim lines to route the visit into the correct maternity case rate. If TH is omitted from Claim.item[0] (the global wellness line), many MCOs deny the visit as "not part of maternity episode." If TH is applied to Claim.item[1] (the problem E/M line), the MCO bundles it into the maternity case rate and denies separate payment.

Scribing.io maintains a payer-specific modifier matrix that maps each Medicaid MCO's TH requirements to the correct claim line. The engine:

• Appends TH to Claim.item[0] when the payer requires it for global accumulator routing.

• Withholds TH from Claim.item[1] to prevent the problem E/M from being absorbed into the maternity case rate.

• Alerts the billing team if the patient's Medicaid MCO has TH requirements not yet mapped, preventing blind submission.

This logic is invisible to the clinician but eliminates a denial pattern that costs multi-site OB practices $15,000–$40,000 annually in unrecovered Medicaid E/M charges.

FHIR R4 Dual-Encounter Export Architecture

Documentation is only as valuable as its interoperability. Scribing.io exports split-visit encounters as FHIR R4-compliant resources, ensuring that downstream systems — EHRs, clearinghouses, payer portals — receive structurally distinct encounters rather than a monolithic blob.

Resource Mapping

The dual-Encounter architecture ensures that even EHR systems that flatten claim data into a single visit record can reconstruct the split at the clearinghouse layer. Each Claim.item references its own Encounter resource, creating an unambiguous audit trail. For practices using Epic, Cerner, or athenahealth, Scribing.io's EHR integration layer maps the dual-encounter structure to the platform's native multi-encounter-per-appointment model.

Trimester and Gestational-Week Validation

Before claim export, the engine performs a final validation pass:

1. Z3A.xx week matches documented gestational age. If the note says "22 weeks" but the EDD-derived calculation yields 21w6d, the engine rounds per ACOG convention and flags the discrepancy for clinician review.

2. Trimester character matches Z3A.xx range. Z3A.22 (22 weeks) falls in the second trimester. If the selected O-code uses a first-trimester character (e.g., O23.41), the engine blocks export and presents the correct option.

3. Z34.xx trimester matches the same gestational calculation. A claim with Z34.01 (first trimester) and Z3A.22 (22 weeks) will be denied for internal inconsistency. The engine prevents this mismatch automatically.

These validation rules run in under 200ms and are invisible to the clinician. They prevent the class of denials that coders typically catch only in retrospective audits — days or weeks after the visit, when correction requires chart amendment and resubmission.

Implementation Checklist for Medical Directors

Deploying Scribing.io's split-visit engine in a multi-provider OB practice requires coordination across clinical, coding, and IT teams. This checklist covers the critical path:

Practices that complete this checklist typically see a measurable reduction in OB E/M denials within the first 30 days of full deployment, with the largest gains concentrated in Medicaid-heavy panels where TH routing errors were previously undetected.

Book a 15-Minute Demo: Global OB Split-Visit Guardrail

If you manage a global OB panel, every prenatal visit where an acute problem surfaces is a potential $1,250 recoupment event. The question is not whether your AI scribe generates notes quickly — it is whether it generates the right claim architecture at the moment of documentation.

Book a 15-minute demo to see the Global OB Split-Visit Guardrail in action:

• Real-time -25 justification — Watch the engine detect an acute problem mid-conversation and fork the documentation stream live.

• Medicaid TH routing — See payer-specific modifier logic applied automatically based on the patient's MCO.

• Trimester/gestational-week validation — Observe the engine block a claim with mismatched trimester codes before export.

• FHIR dual-encounter export — Inspect the raw FHIR R4 resources and Claim.item structure tuned to current NCCI/PTP edits for first-pass payment.

Schedule your demo at Scribing.io →

Your documentation engine should protect your revenue at the claim-line level, not just the note level. That distinction is the entire point.