CPT 90837: the code payers read twice
90837 is the 60-minute individual psychotherapy code. It reimburses more than 90834, it is the most scrutinised code in outpatient behavioral health, and the thing that decides whether you keep the money is almost never the diagnosis — it is whether your note documents time and medical necessity in the same place.
Most billing guides tell you 90837 means "60 minutes of psychotherapy." That is true and almost useless, because nobody's session runs exactly sixty minutes and the code does not require that it does. What matters is the threshold, and the threshold is not sixty.
The time rule, stated precisely
CPT timed codes use a midpoint convention: you bill the code whose stated time you have passed the halfway point toward. For the individual psychotherapy family that produces three bands, and the bands — not the labels — are what you bill against.
| Code | Stated time | Billable range | In practice |
|---|---|---|---|
90832 | 30 minutes | 16–37 minutes | Brief contact, crisis follow-up, med-adjacent check-in |
90834 | 45 minutes | 38–52 minutes | The standard hour-that-isn't. Most common outpatient code. |
90837 | 60 minutes | 53 minutes and above | Trauma processing, EMDR, exposure work, intake-adjacent depth |
A 52-minute session is a 90834. A 53-minute session is a 90837.
That one minute is worth real money across a year, which is exactly why payers built audit
programs around it.
Count face-to-face time only. The clock covers time spent with the client. Writing the note afterward, coordinating with a psychiatrist, reviewing records before the session — none of that counts toward the psychotherapy minutes, even though all of it is real work. Some of it is separately billable under care-management codes; none of it converts a 90834 into a 90837.
Why this code specifically draws scrutiny
Two things make 90837 an outlier. First, the reimbursement gap: depending on region and payer, 90837 commonly pays meaningfully more than 90834 for the same clinician and the same client. Second, the shape of the data. A practice billing 90834 for most sessions and 90837 occasionally looks like normal clinical variation. A practice billing 90837 for 95% of sessions looks like a billing pattern, and payer analytics are built to find billing patterns.
This is worth being blunt about, because the advice you will read elsewhere splits into two bad camps. One says "avoid 90837, it triggers audits" — which is telling clinicians to under-bill for work they actually did. The other says "bill what you did, ignore the noise" — which is fine right up to the records request. The correct position is narrower: bill 90837 whenever the session genuinely ran 53 minutes or longer, and make sure the note can survive being read cold by someone who was not in the room.
What the note actually has to carry
An auditor reading a 90837 note is checking four things, in this order. Missing any one of them is how a clinically sound session becomes a recoupment.
1. Start and stop time, or total minutes
Not "60 minutes" as a template default on every note — an identical duration on every session for eighteen months is itself a red flag. Record the actual time. If your sessions genuinely run 53–55 minutes, that is what should appear.
2. A diagnosis that supports this intensity
The diagnosis has to justify not just treatment but treatment at this length. Adjustment disorder with a stable presentation and a 90837 every week invites the question of why. PTSD with active exposure work does not.
3. The treatment plan goal today's session advanced
This is the link most notes are missing, and it is the one that converts an audit from awkward into expensive. A note that describes a good session but never connects it to an active goal leaves the payer holding documentation of a conversation rather than documentation of treatment. See the golden thread for why this single link carries so much of the risk.
4. Clinical intervention, not just content
"Client discussed conflict with sister" is content. "Used cognitive restructuring to identify and test the belief that setting a boundary would end the relationship; client generated two counter-examples" is an intervention. Payers reimburse the second one. The first reads as supportive listening, which many contracts do not cover at 90837 rates.
The four denials you will actually see
- Time not documented. The most common and the most avoidable. The note describes an hour of work and never says how long it took.
- Frequency without justification. Twice-weekly 90837 without a note explaining acuity, a step-down plan, or a time-limited protocol. The fix is a sentence, written at the time, not reconstructed at appeal.
- Prior authorization lapsed. Some plans authorise a visit count rather than a date range, and the count runs out mid-course without a renewal notice.
- Code/telehealth mismatch. Right code, wrong place-of-service or missing modifier. Covered below.
90837 delivered by telehealth
Telehealth does not change the code. It changes the metadata around the code, and the
metadata is where claims break. You need the correct place-of-service — commonly 10
for the client's home and 02 for other locations — and, for many payers, modifier
95 to indicate a synchronous audio-video service. Requirements diverge sharply
between Medicare, Medicaid and commercial plans, and they have changed repeatedly since 2020.
Treat any list you read, including this one, as needing confirmation against the current
policy for the specific plan you are billing.
Add-on and adjacent codes
90785— interactive complexity. An add-on, not a substitute, for sessions complicated by a third party, mandated reporting, or communication barriers.90847— family psychotherapy with the client present. If a partner joins for the full session, that is usually not a 90837.90791— the diagnostic evaluation. A first session is typically 90791, not 90837.90833— the psychotherapy add-on billed alongside an E/M code by prescribers. Different documentation standard entirely.
What Weft does with this
Weft proposes a CPT code only when the documentation supports it, and shows you the sentence that justifies it. If the session ran 51 minutes, it will not offer 90837. If the note never names the treatment plan goal it advanced, you hear about it at signature rather than at appeal — because the note is written against the active plan rather than into a blank box. The audit log is the product, not a setting someone remembers to enable.
Frequency: the argument you win in advance
Medical necessity is not established once at intake. It is re-established, implicitly, on every claim. Weekly 90837 for a client in acute PTSD treatment is unremarkable. Weekly 90837 for eighteen months with flat outcome measures and no plan revision is a question waiting to be asked, and the answer has to already be in the chart when it is.
What makes the difference is a single sentence written at the time — a rationale for the current intensity and, ideally, the conditions under which it would step down. "Continuing weekly at 60 minutes while completing the exposure hierarchy; will reassess for biweekly at PHQ-9 below 10 sustained across two administrations" costs you eight seconds and settles the question permanently. Reconstructing that reasoning two years later, from memory, in response to a records request, is a different experience.
This is also where outcome measures stop being an administrative chore. A PHQ-9 trend in the chart is the cheapest possible evidence that treatment at this intensity is doing something. Without it, intensity is an assertion.
If a records request arrives
Records requests are routine and are not, by themselves, an accusation. How you respond determines whether they stay routine.
- Send exactly what was asked for. Not the full chart. Over-disclosure creates new surface area and, for substance-use records, can itself be a violation under 42 CFR Part 2.
- Do not amend signed notes. A late edit to a note under review is far worse than a thin note. If something genuinely needs correcting, use a dated, attributed addendum that is transparently an addendum.
- Check the sample period before you respond. If the request covers a window in which you know documentation was weak, that is worth knowing before you reply, and worth a conversation with a healthcare attorney rather than a billing forum.
- Expect extrapolation. Payers commonly project findings from a sample across the whole period. Four unsupported claims out of twenty is not a four-claim problem.
Where the numbers vary
Everything above concerns coding rules, which are national. Reimbursement is not. The gap between 90834 and 90837 differs by geography, by payer, by contract and by whether you are in-network. Medicare rates are published and predictable; commercial rates are negotiated and frequently confidential. Medicaid varies by state and sometimes by managed-care organisation within a state.
The practical implication is that no article — this one included — can tell you what 90837 pays in your market. It can only tell you the rule that determines whether you may bill it. Check your own fee schedule and your own contracts for the rest.
Verified 29 July 2026 against AMA CPT and CMS published guidance. Payer policy varies — confirm against your own contracts. Primary references: AMA CPT; CMS Physician Fee Schedule; APA Services reimbursement guidance. This page is billing reference, not legal or coding advice.