ICD-10 coding for psychiatric encounters is one of the places where general-purpose ambient scribes struggle the most. The diagnoses are nuanced, the comorbidities are routine, and the codes that the model “remembers” from primary care training do not always map to what the chart actually says. Wrong code in, wrong claim out. Documentation gets bounced. Revenue cycle gets noisy. Nobody wins.
Nextvisit’s coding workflow was built specifically for the way psychiatric assessments read. Here is how it works in practice.
Codes come from the assessment, not from a popularity contest
The coding model does not suggest the most-common F-code for an apparent diagnosis. It reads the assessment you wrote and extracts the diagnostic claims, then matches them to ICD-10 codes that fit those claims specifically.
If your assessment says “moderate major depressive disorder, recurrent, without psychotic features, with a partial response to sertraline 100 mg,” the suggested code is F33.1, not F32.9. Specifier matters. Sequence (single vs recurrent) matters. Severity (mild, moderate, severe, with or without psychotic features) matters. The system reads them all.
If a specifier is missing from your assessment, the system flags it as a coding gap rather than guessing. You can fill in the specifier or accept the less-specific code.
Polypharmacy and comorbidity are first-class
Most psychiatric visits carry more than one active diagnosis. The coding workflow surfaces all active conditions documented in the chart, ranks them by relevance to the current encounter, and proposes the order most claims systems expect (primary, then comorbid, then chronic conditions monitored but not actively treated).
Polypharmacy gets specific handling. If you are managing a patient on two antipsychotics, the system surfaces F-codes that reflect the active conditions and prompts whether long-term medication monitoring (Z79.899 or its specifiers) belongs on the claim.
Add-on codes that show up correctly
The CPT add-on codes psychiatry uses, time-based prolonged services, interactive complexity (90785), and the smoking and alcohol counseling codes when they apply, get suggested when the documentation supports them and not when it does not. Time-based codes read the actual time you documented. The system does not invent time.
For the codes that require specific documentation, like 90785, the system checks for the elements (high-risk behavior, mandatory reporting, communication barrier, et cetera) and suggests the code when at least one element is documented.
What you still review
Every code is a suggestion. Nothing is final until you sign. The coder, biller, or supervising clinician on your team can override any suggestion. The system records the override, and over time the suggestions adapt to your group’s coding style.
This is the part general-purpose scribes get wrong. Coding is not a feature you bolt onto an ambient model. It is a workflow that has to read your assessment carefully, respect your specialty, and surface its work for clinical review. Done that way, it stops being a source of denials and starts being part of the documentation.