What an Auditor Reads First in Your Therapy Chart

When a payer pulls your charts, nobody starts by debating your clinical judgment. A reviewer works through a documentation checklist in a deliberate order — objective, pass–fail items first — because that is the fastest way to decide a claim. Is every note signed and dated? Is there a certified plan of care that covers each date of service? Do the documented minutes support the units billed? A chart can fail three times before anyone reads a sentence of your reasoning. That order is not a secret, and it is the single most useful thing to know about audit defense: if you can predict what gets read first, you can check it yourself, on your own schedule, while every problem is still free to fix.

The audit most outpatient practices actually meet is not a fraud investigation. It is routine medical review, and for Medicare Part B the main vehicle is Targeted Probe and Educate (TPE). Under TPE, your Medicare Administrative Contractor selects providers using claims data — unusual billing patterns relative to peers, high denial rates, or services on the contractor’s review list — and requests records for a probe of 20 to 40 claims. The contractor reviews the documentation against coverage and documentation requirements, sends claim-by-claim results, and offers one-on-one education. A provider with a high error rate goes through up to three rounds, with education between each; persistent failure after round three is referred to CMS for next steps that can include 100 percent prepayment review, extrapolation of the error rate across a larger universe of claims, or referral to a Recovery Auditor.

Alongside TPE, CMS runs the Comprehensive Error Rate Testing (CERT) program, which audits a random sample of paid claims every year to estimate how much Medicare paid improperly. The finding that should change how you think about audit defense: the leading driver of improper payments in therapy review is insufficient documentation — records that do not contain the required elements to support the claim as billed. Not fabricated visits, not upcoding schemes: paperwork that could not carry the claim it was attached to. That is a fixable problem, and the rest of this article is the fix.

Put yourself at the reviewer’s desk. There is a stack of records and a worksheet built from the Medicare Benefit Policy Manual’s documentation requirements. The efficient way through the stack is to ask the objective questions before the judgment questions — a missing signature can be spotted in seconds and decides the claim by itself, while weighing whether a service required a therapist’s skill takes a full read of the episode. So the review proceeds in tiers: first the administrative pass–fail items (signatures, dates, a certified plan of care covering every billed date), then the arithmetic (documented minutes against billed units, notes matching billed codes), and only then the clinical read (skilled care, medical necessity, progress that justifies continuing). Each tier is only reached if the chart survives the one before it.

This ordering is why audit outcomes feel unfair to clinicians. The therapy could have been textbook; the reasoning could have been sound. If the plan of care was certified late with no explanation, or the note documents 20 minutes while the claim bills two timed units, the claim can fail without the quality of care ever entering the conversation. The flip side is genuinely good news: the items that decide claims earliest are the most mechanical to check, which means the bulk of audit risk yields to a checklist run by anyone in the practice — no clinical judgment required.

This is the review order turned into a tool. Pull one complete episode — evaluation through the most recent visit — and answer each item yes or no, in order, exactly as a reviewer would. Every item is binary on purpose: if you find yourself writing an explanation instead of checking a box, you have found the thing a reviewer would deny. The first eight items need no clinical training, so a well-briefed front-desk lead or biller can run them; the last four belong to a clinician.

Signature problems are the least interesting failures and among the most common, because they are invisible in daily work — the note reads fine on screen whether or not it was ever signed. Medicare requires services to be authenticated by the author with a handwritten or electronic signature; a stamped signature does not qualify, and an illegible scrawl with no printed name or credentials fails unless a signature log establishes who it belongs to. Two habits close most of the gap: a weekly report of unsigned notes older than a few days, and a signature log listing every clinician’s name, credentials, and signature style. If a review request arrives and you discover a missing signature, do not backdate — CMS’s signature guidance provides for an attestation statement signed by the author; a late entry is permitted when it is clearly labeled as late and dated the day it was actually written.

Certification is the other tier-one killer, and it is a calendar problem more than a clinical one. The plan of care needs a physician or non-physician practitioner’s dated signature within 30 calendar days of the first treatment, and again at least every 90 calendar days for as long as treatment continues — and the clock runs on calendar days, indifferent to cancellations and holiday closures. Delayed certification is permitted with a documented reason, but a chart that leans on the exception every episode is telling the reviewer something about the practice’s systems. Because the failure mode is almost always tracking rather than refusal, the fix is workflow: our recertification tracking guide covers the two dates to log and the chase cadence that keeps signatures arriving before windows close.

Tier two is where the reviewer puts the note and the claim side by side. For timed codes, the treatment note must document total timed-code minutes and total treatment time, and the timed minutes must support the units billed under Medicare’s counting rules — the 8-minute rule guide works through that arithmetic, and the timed-versus-untimed codes article covers which codes the minute-counting applies to at all. The mismatches a reviewer finds are rarely dramatic: a note that says “45-minute session” with no split between timed and untimed services, or minutes that sit one unit short of what was billed. Each is a denial that took the reviewer under a minute to decide.

Only after the mechanical tiers hold does the reviewer read for substance, asking two questions. First: did this service require the skills of a therapist? A note that records only performance — “completed 3×10 with minimal assist, tolerated well” — describes something an exercise log could hold, and skilled therapy has to read differently: the assessment made mid-session, the cue that changed the outcome, the parameter adjusted and why. Our medical necessity guide is a full treatment of writing that reasoning into notes without doubling documentation time. Second: is the patient progressing in a way that justifies continued treatment — or, where progress has slowed, does the record show the clinical reasoning for continuing, changing, or discharging? That evidence is supposed to accumulate in the progress report, which is why the 10-treatment-day cadence is worth defending in your scheduling system rather than your memory; the daily note versus progress note guide maps which document carries which burden.

The last read is across documents, and it is where charts that survived every tier can still unravel. A reviewer with the note, the claim, the schedule, and the plan of care in front of them will notice when they disagree: a note for a date the schedule shows as a no-show, goals in the notes that stopped matching the certified plan two months ago, an evaluation that names deficits the treatment notes never address again. The most damaging inconsistency is the one that questions authorship — identical paragraphs recurring note after note, or the same sentence appearing in three different patients’ charts the same afternoon. Cloned text does not just risk a denial for the visit in question; it invites the reviewer to doubt every note that resembles it. Our guide to templates that do not read as cloned covers how to keep structure without inheriting that risk.

A checklist that only exists in an article defends nothing. The practices that come through probes cleanly are the ones that made the self-audit a small, boring, recurring event with a named owner — the same move that works for assistant supervision documentation and every other compliance obligation that fails silently. The cadence below assumes nothing about practice size; a solo clinician and an office manager at a six-therapist clinic run the same loop at different scale.

1. 01

   Pull a small sample nobody hand-picked

   Each quarter, select a handful of episodes across clinicians, payers, and service types — recent discharges plus one or two long-running episodes, since certification and progress-report failures concentrate where treatment runs longest. Choose by rule (for example, every nth chart from the quarter’s discharge list), not by memory of which charts are probably fine.

2. 02

   Run the checklist in reviewer order and log failures by item

   Work each chart top to bottom and record which item failed, not just that something did. Three missing progress reports across three clinicians is a scheduling-system gap; three unsigned notes from one clinician is a conversation. The pattern in the log is the finding — individual misses are just its symptoms.

3. 03

   Fix the workflow, not the chart

   For documentation already submitted with a claim, correct the record only through proper channels — clearly labeled late entries or addenda, never edits that obscure what the record said before. Then spend the real effort upstream on the system that produced the miss: the unsigned-note report that runs weekly, the certification-date field the biller can see, the progress-report counter in the EHR.

4. 04

   Re-check the failed item next quarter

   The loop closes when the item that failed last quarter passes on a fresh sample, not when the fix is announced. Keep the log; a practice that can show quarters of self-audit findings and corrections is also, incidentally, showing a reviewer exactly the compliance posture that makes education rounds end early.

None of this asks you to write longer notes. Almost every item in the reviewer’s first two tiers is about completeness and agreement — signed, dated, certified, counted, consistent — and the clinical tier is about making visible the reasoning you are already doing. The chart that survives review is rarely the longest one. It is the one where every document answers the question the next document raises, in the order someone else will read them.