Why Behavioral Health Coding Denials Keep Rising in 2026
Claims denials and automatic downcoding made the top five list of administrative burdens in MGMA’s 2026 Regulatory Burden Report, released in April from a survey of more than 230 medical group practices. Ninety percent of respondents said prior authorization burden had increased over the past year, and 40% of practices now dedicate three or more full-time administrative staff per physician just to keep up with payer rules, authorizations, and audits. For behavioral health practices, that burden lands harder than the averages suggest, and it’s exposing a coding problem that generic claims-automation tools weren’t built to solve.
What the MGMA Numbers Actually Measure
MGMA’s report, based on responses from practices where 60% are independent, found that 95% of respondents have seen regulatory burden increase over the past three years, and 77% named it a significant factor in physician burnout. Those figures describe medical practices broadly — primary care, specialty medicine, surgery. Behavioral health wasn’t broken out separately in the report. But the structural reasons denials and downcoding rank so high apply with particular force to psychiatry, psychology, and substance-use treatment billing, where coding depends on clinical judgment calls that are harder to standardize than a lab order or an imaging study.
Why Behavioral Health Carries a Heavier Coding Load
Time-Based and Add-On Codes Multiply Error Points
Most office-based medical visits code to a single E/M level supported by documented history, exam, and medical decision-making. Behavioral health visits frequently stack codes: a psychiatric evaluation or psychotherapy session (90791, 90834, 90837) alongside interactive complexity add-ons (90785), family or group therapy codes, and sometimes a separately billable E/M service on the same date for medication management. Each additional code is another place documentation can fail to support what was billed, and payers increasingly flag these combinations for review rather than paying them automatically.
Level-of-Care Documentation Rarely Maps Cleanly to a Code
Substance-use treatment in particular relies on ASAM criteria to justify a level of care — residential, partial hospitalization, intensive outpatient — and that justification lives in narrative clinical notes, not structured fields. A coder or an automated system has to infer medical necessity from free text that was written for clinical purposes, not billing ones. When that inference doesn’t match what the payer’s own review algorithm expects to see, the claim gets denied for lack of medical necessity even when the care was appropriate.
Where Agentic AI Is Actually Helping
The tools built specifically for this workload are narrower than general-purpose claims automation, and they’re earning their keep in a few concrete places:
- Pre-submission code-combination checks that flag add-on codes (like 90785) lacking a documented basis before the claim goes out, rather than waiting for a payer denial.
- Documentation-to-code matching that cross-references therapy notes against the billed CPT code and surfaces gaps — for example, a 90837 (60-minute session) billed against a note describing 35 minutes of contact.
- Prior authorization drafting that pulls ASAM-relevant details out of existing notes to pre-populate level-of-care justifications, cutting the manual assembly time MGMA’s report points to as a staffing drain.
- Denial-pattern tracking that groups rejected claims by payer and reason code, so practices can see whether a specific insurer is systematically downcoding a specific service rather than treating each denial as an isolated event.
Where It Still Falls Short
None of this replaces the clinical judgment MGMA’s report implicitly argues is being pushed onto administrative staff. An algorithm can confirm that a note’s documented session length matches the billed code; it can’t determine whether a clinician’s assessment of a patient’s level of care need is itself defensible on appeal. That gap is exactly why MGMA found 40% of practices need multiple dedicated staff per physician: someone still has to read the note, understand the clinical reasoning, and decide whether it holds up. Agentic AI shortens the time it takes to find the gap. It doesn’t close the gap on its own, and vendors who claim otherwise for behavioral health specifically are overstating what the current generation of tools does.
What This Means for Coding Teams
The practical takeaway from MGMA’s report isn’t that behavioral health coding is uniquely broken — it’s that the administrative load the report measures across all practice types concentrates disproportionately in specialties with high documentation-to-code complexity, and behavioral health sits near the top of that list. Coding teams that want to reduce denial rates should focus first on the narrow, well-defined checks — code-combination validation, documentation matching, denial-pattern analysis — where automation has a clear, auditable basis for its output, rather than treating level-of-care determinations as something software can fully own.
That sequencing matters for budget decisions too. A practice weighing where to spend a limited RCM technology budget gets more return from tools that catch a 90785 billed without supporting documentation before the claim goes out than from tools that promise to fully automate medical-necessity review. The former has a measurable, auditable error rate. The latter is the piece of the workflow MGMA’s own numbers suggest still needs a trained person reading the note.
Medikode’s automated medical coding platform applies that same principle: automation where the coding logic is well-defined, and clear flags for human review where clinical judgment still has to carry the claim.