Operations Playbook — Table of Contents

• 1. Why Summary Scores Alone Fail Outpatient Psychiatry

• 2. The Information-Gain Gap — What Existing Guidance Misses and What Scribing.io Delivers

• 3. Scribing.io Clinical Logic — Handling the 90837 Telehealth Desk Audit

• 4. Technical Reference: ICD-10 Documentation Standards

• 5. FHIR R4 Data Model — Persisting Item-Level Observations

• 6. The Pre-Sign-Off Validator — Enforcement Logic

• 7. Telehealth Parity Documentation Requirements

• 8. Implementation Workflow — From Install to First Audit-Proof Note

• 9. See the Item-to-HPI Mapper Live

Clinical Update — June 2026
This guide has been revised for June 2026 to reflect CMS's updated CY2026 OPPS/ASC Final Rule telehealth place-of-service audit criteria, the APA's 2026 Practice Guidelines for Major Depressive Disorder documentation standards, and FHIR R4 Observation profile updates from HL7's May 2026 ballot cycle. If you read a prior version of this playbook, the clinical logic walkthrough in Section 3 and the FHIR component mappings in Section 5 have been substantially rewritten.

Automating PHQ-9 & GAD-7 Documentation: The Clinical Operations Playbook for Outpatient Psychiatry

Most EHRs store PHQ-9 and GAD-7 results the same way they store a hemoglobin A1c: one number, one date, one row in a flowsheet. That works for lab values. It does not work for psychotherapy billing. A hemoglobin A1c of 9.2 justifies intensifying metformin without additional narrative. A PHQ-9 of 18 does not justify 90837 without additional narrative—and the narrative must connect individual elevated items to specific interventions, time expenditure, and patient response. This playbook codifies the operational logic that Scribing.io uses to generate that narrative automatically from ambient session capture, persist it as structured FHIR R4 data, and enforce medical-necessity completeness before a note is signed.

The target reader is an outpatient psychiatrist (MD/DO) billing 90837 for sessions exceeding 53 minutes of face-to-face psychotherapy time, particularly via telehealth where audit scrutiny is highest. The playbook is equally relevant to group practice administrators responsible for compliance and revenue integrity, and to clinical informaticists integrating Scribing.io's psychiatry-specific AI scribe into EHR workflows. Clinicians in Family Medicine who perform integrated behavioral health screening will find the FHIR data model and item-level documentation principles directly transferable.

1. Why Summary Scores Alone Fail Outpatient Psychiatry

A PHQ-9 total of 18 tells a payer one thing: the patient exceeds the moderate-severity threshold. It does not tell the payer which of the nine symptom domains are driving that score. Two patients can both score 18 with entirely different clinical profiles—one dominated by suicidality and psychomotor retardation, the other by anhedonia and appetite dysregulation—requiring completely different intervention sets and session durations. Summary scores erase this distinction.

Payer audit logic exploits that erasure. The desk reviewer's checklist for 90837 medical-necessity review (as operationalized by major commercial payers and Medicare Administrative Contractors) requires four documentation elements:

1. Specific symptoms addressed — not diagnostic categories, but discrete presenting problems tied to the session's clinical focus.

2. Specific interventions matched to those symptoms — the psychotherapy modality or technique applied to each problem, not a generic "CBT provided."

3. Rationale for extended time — an explicit clinical argument for why 53+ minutes was necessary rather than 38–52 minutes (the 90834 range).

4. Patient response — observable or reported change during the session, documenting engagement with the intervention.

A note containing "PHQ-9: 18, moderate-to-severe depression, CBT psychotherapy provided, 58 minutes" satisfies zero of those four requirements. The score is summary, the diagnosis is categorical, the intervention is generic, and the time is unsubstantiated by narrative detail. Current clinical benchmarks suggest up to 20% of 90837 claims in outpatient behavioral health face post-payment review, with insufficient medical-necessity documentation—not incorrect CPT selection—as the leading denial rationale (AMA CPT E/M Documentation Guidelines).

The AMA's own BHI Workflow Examples reference PHQ-2, PHQ-9, and GAD-7 as screening instruments within a collaborative care model. The guidance specifies when to administer: initial visit, follow-up thresholds, care transitions. It does not specify how to decompose item-level results into the psychotherapy note's HPI. That decomposition is the documentation step where revenue is protected or lost, and it is the step that Scribing.io automates at the transcript level.

2. The Information-Gain Gap — What Existing Guidance Misses

The Core Problem in One Sentence

Existing workflow guides treat PHQ-9 and GAD-7 as binary screening gates (positive/negative) and leave the documentation burden entirely to the clinician's free-text typing after a 58-minute session.

Five Critical Gaps and How Scribing.io Closes Each

The Anchor Truth

AI must map individual response items from PHQ-9/GAD-7 directly into the HPI narrative to support the "Medical Necessity" of High-Intensity Psychotherapy codes (90837) and prevent downcoding to 90834.

This is not a productivity feature layered onto documentation. It is a revenue-protection and clinical-integrity mechanism. The difference between a note that says "PHQ-9 total 18, moderate-to-severe depression, psychotherapy provided" and a note that says "PHQ-9 #9 = 2, indicating suicidal ideation without active plan; C-SSRS Level 2 confirmed; 22 minutes of safety planning including means restriction counseling; patient verbalized commitment to safety plan and identified three protective factors" is the difference between a downcode and a defended claim. Scribing.io's PHQ-9/GAD-7 auto-linker produces the second version from a live session transcript—without requiring the psychiatrist to type, click, or template anything beyond conducting the session.

3. Scribing.io Clinical Logic — Handling the 90837 Telehealth Desk Audit

This section provides a granular, step-by-step walkthrough of the single most important use case this playbook addresses: a post-payment desk audit targeting a telehealth 90837 session. Every decision point is annotated with the specific Scribing.io function that resolves it.

The Scenario

A psychiatrist documents a 58-minute 90837 telehealth session for a patient carrying F33.1 - Major depressive disorder, recurrent, moderate; F41.1 - Generalized anxiety disorder. The EHR flowsheet captures only PHQ-9 = 18 and GAD-7 = 16 as summary totals. Three months post-service, a payer desk audit moves to downcode to 90834 citing "lack of medical-necessity detail."

Step-by-Step: How the Audit Is Defeated

Step 1 — Ambient Capture and Transcript Generation. Scribing.io's ambient microphone captures the full 58-minute telehealth session (audio/video, HIPAA-compliant, BAA-covered). The speech-to-text engine generates a time-stamped transcript with speaker diarization (clinician vs. patient). Timestamp precision: ±15 seconds. This transcript becomes the evidentiary backbone of the note.

Step 2 — Item-Level Instrument Parsing. The auto-linker cross-references the PHQ-9 and GAD-7 responses captured during the session against the transcript. Rather than recording "PHQ-9 = 18," the system decomposes the total into its nine constituent items with individual scores. It identifies which items are clinically elevated (score ≥2) and flags them for mandatory HPI inclusion:

• PHQ-9 #9 (Suicidal ideation) = 2 — SI without active plan; passive thoughts of death reported 3×/week.

• PHQ-9 #3 (Sleep disturbance) = 3 — Sleep-onset insomnia, 90+ minute latency, daytime fatigue impairing occupational function.

• PHQ-9 #1 (Anhedonia) = 2 — Loss of interest in social activities, withdrawal from previously enjoyed hobbies.

• GAD-7 #5 (Restlessness) = 3 — Psychomotor agitation observable on video; unable to sit still during opening 10 minutes.

• GAD-7 #7 (Catastrophic fear) = 3 — "If I go out, something terrible will happen" — verbatim from transcript at 00:14:22.

Step 3 — Intervention-to-Item Mapping with Time Anchors. The transcript reveals four distinct clinical interventions. Scribing.io maps each intervention to the elevated item it addresses and calculates minutes from transcript timestamps:

Step 4 — Auto-Generated Medical-Necessity Statement. From the intervention map, Scribing.io composes and inserts the following HPI paragraph:

"Extended psychotherapy time (58 minutes, exceeding the 38-minute threshold for 90837) was medically necessary due to: (1) active suicide risk assessment requiring full C-SSRS administration and comprehensive safety planning; (2) multiple targeted psychotherapy interventions tied to individually elevated PHQ-9 and GAD-7 items across four functional domains (safety, sleep, somatic anxiety, social functioning); (3) documented patient engagement and in-session behavioral response to each intervention; (4) total face-to-face psychotherapy time: 58 minutes, start 10:02 AM ET, stop 11:00 AM ET, audio/video telehealth via HIPAA-compliant platform, POS-02."

Step 5 — Pre-Sign-Off Validation. Before the psychiatrist can sign the note, the validator confirms all four required elements are present: (1) start/stop timestamps ✓, (2) ≥2 item-linked interventions ✓ (four present), (3) explicit "why 90837 not 90834" sentence ✓, (4) patient response documented for each intervention ✓. The note signs. The claim submits.

Step 6 — Audit Response. Three months later, the desk audit arrives. The practice submits the note. The auditor's checklist—specific symptoms, matched interventions, time rationale, patient response—is satisfied on every line. 90837 stands. No recoupment.

Revenue Impact at Scale

A psychiatrist seeing 25 patients per week with 60% billed at 90837 files approximately 780 high-intensity claims annually. At a 10% downcode challenge rate and a $55 average differential between 90837 and 90834 reimbursement, that represents $4,290 in annual risk per clinician from downcoding alone—before accounting for the administrative cost of appeals (estimated 45–90 minutes per appeal) or the cascade effect when a payer extends the audit to other dates of service. A five-clinician group faces $21,000+ in annual exposure. Practices deploying item-level documentation automation report downcode challenge rates approaching zero for sessions where clinical time genuinely met the 90837 threshold.

4. Technical Reference: ICD-10 Documentation Standards

Scribing.io's diagnostic coding engine enforces maximum specificity for every ICD-10 code attached to a psychiatric encounter. Generic codes invite denials; specific codes corroborate the clinical narrative.

Specificity Enforcement for Common Psychiatric Diagnoses

Consider the patient in our audit scenario. An underspecified note might assign "F33 — Major depressive disorder, recurrent" without a severity qualifier. This fourth-character truncation triggers automatic rejection by many payers because CMS requires coding to the highest level of specificity documented in the record (CMS ICD-10-CM Official Guidelines for Coding and Reporting).

Scribing.io resolves this by cross-referencing the PHQ-9 total and item-level data against ICD-10 severity thresholds:

• F33.1 - Major depressive disorder — The system selects F33.1 (not F33.0 mild, not F33.2 severe) based on PHQ-9 total 18 falling within the 15–19 "moderately severe" range. It appends the specifier recurrent by verifying prior episode documentation in the problem list. The resulting code reaches fifth-character specificity: F33.1, recurrent, moderate — maximum granularity for this clinical presentation.

• moderate; F41.1 - Generalized anxiety disorder — GAD-7 total 16 (≥15 = severe range per Spitzer et al., 2006) is cross-referenced against clinical documentation. F41.1 is assigned as a secondary diagnosis, linked to the specific GAD-7 items driving session interventions.

How Specificity Prevents Denials

When the ICD-10 code matches the instrument score range and the HPI narrative describes symptom-specific interventions consistent with that severity level, the payer's automated claims adjudication system finds no discrepancy. The triad of (1) maximally specific ICD-10 code, (2) item-level instrument data, and (3) intervention-matched HPI narrative creates a closed evidential loop. Scribing.io constructs this loop automatically. The clinician's only responsibility is to conduct the clinical session.

For practices managing complex comorbidity patterns, the system supports up to 12 linked ICD-10 codes per encounter, each validated against the note's clinical content. Codes that appear on the problem list but receive no session-specific documentation are flagged rather than automatically attached—preventing the common compliance error of "diagnosis list carry-forward" without clinical re-attestation (AMA CPT Documentation Guidance).

5. FHIR R4 Data Model — Persisting Item-Level Observations

Narrative documentation alone is insufficient for longitudinal outcome tracking, quality reporting (e.g., MIPS Improvement Activities for behavioral health), and interoperability. Scribing.io persists every instrument administration as a FHIR R4 Observation resource conformant with the HL7 FHIR Observation specification.

Observation Resource Structure

The same structure applies to GAD-7 (LOINC 70274-6 total, with component codes 70275-3 through 70281-1 for items 1–7).

EHR Failover Logic

Not every EHR supports FHIR R4 Observation components natively. Epic's FHIR API accepts components; certain legacy systems flatten Observations to a single value. For those environments, Scribing.io's failover logic writes:

1. Discrete HPI synopsis — A structured text block enumerating each item→score→intervention link, inserted as a parseable section within the HPI.

2. Embedded signed JSON evidence block — A cryptographically signed JSON object containing the full component-level Observation, stored as a note attachment. This preserves item-level auditability even when the EHR's data model cannot represent it natively. The signature prevents post-hoc modification.

6. The Pre-Sign-Off Validator — Enforcement Logic

Documentation quality degrades at the end of a clinic day. A psychiatrist finishing their eighth 90837 session at 6:30 PM is not going to manually verify that every note contains a medical-necessity rationale. The validator does it for them.

Four Required Fields — Non-Negotiable

The validator is not optional. It cannot be overridden by the clinician. This is a deliberate design decision grounded in compliance engineering: the most common documentation failure mode is not inability to document—it is forgetting to document under time pressure. The validator eliminates the forgetting.

7. Telehealth Parity Documentation Requirements

Telehealth 90837 sessions face disproportionate audit scrutiny. CMS and commercial payers have flagged telehealth behavioral health claims for focused medical review since 2023, with the OIG Work Plan specifically targeting telehealth psychotherapy time verification. Scribing.io addresses this with three automatic insertions for every telehealth encounter:

• POS-02 Tagging — Place of Service code 02 (Telehealth Provided Other than in Patient's Home) or 10 (Telehealth Provided in Patient's Home) is auto-applied based on patient-reported location at session start.

• Audio/Video Confirmation — The note auto-documents that the session was conducted via synchronous audio/video telecommunications, satisfying CMS's interactive telecommunications system requirement. Audio-only fallback (modifier 93/FQ) is flagged when video drops are detected.

• Transcript-Derived Timestamps — Start and stop times are extracted directly from the ambient capture system's clock, not entered manually by the clinician. This eliminates the most common telehealth audit vulnerability: unsupported self-reported session duration.

8. Implementation Workflow — From Install to First Audit-Proof Note

Total time from contract to first fully autonomous audit-proof note: typically 10–14 business days. No workflow disruption during calibration—the clinician conducts sessions identically to their current practice. Scribing.io observes, learns, and drafts. The clinician reviews, adjusts, and signs.

9. See the Item-to-HPI Mapper Live

Reading about item-level documentation automation is one thing. Watching your own clinical scenario processed in real time is another.

See a live demo of our item-to-HPI mapper with FHIR export and the 90837 audit-defense validator that flags downcoding risks before you sign the note.

The demo walks through a complete 90837 telehealth session—from ambient capture to signed note to simulated desk audit—using your specialty's instrument set and intervention vocabulary. You will see exactly how PHQ-9 #9 = 2 becomes 22 minutes of documented safety planning in the HPI, how the validator catches a missing patient-response field, and how the FHIR R4 Observation persists every item score for longitudinal tracking.

Request your live demo at Scribing.io →

This Operations Playbook is maintained by the Clinical Documentation Intelligence team at Scribing.io. Last substantive revision: June 2026. Clinical references verified against AMA CPT 2026 codebook, CMS CY2026 OPPS Final Rule, HL7 FHIR R4 (v4.0.1 + May 2026 ballot updates), and DSM-5-TR diagnostic criteria. For corrections or clinical feedback, contact the editorial team via scribing.io.