snf-appeal-critique

name

snf-appeal-critique

description

Independent adversarial reviewer critique of a drafted SNF / Medicare Advantage appeal supplement BEFORE filing. Stress-tests the record as a hostile payer medical director would — verifying each cited exhibit against the underlying source (OCR / primary PDFs), ranking strongest and weakest arguments, anticipating counter-arguments, and returning a fixed 8-deliverable critique document with a recommendation, minimum-fix list, and final integrity check. Use when the user says "/snf-appeal-critique", "stress-test the appeal", "red-team the supplement", "independent reviewer pass on the filing", "critique the appeal before I file", or "what will a hostile reviewer find in this draft?".

SNF Appeal Critique

Produce an independent clinical-utilization-reviewer critique of a drafted appeal supplement, in the posture of a hostile payer medical director, with every factual claim in the draft independently verified against source exhibits. This runs AFTER /snf-appeal-supplement (or a hand-drafted supplement) is ready, and BEFORE the doc-hold / filing deadline. Companion files:

• REVIEWER_PROMPT.md — the hostile-reviewer persona calibration.
• CRITIQUE_TEMPLATE.md — the exact 8-section output structure.

When to run

• Doc-hold window is open — a first-level supplement has been filed, a reviewer has requested or permitted further documents, and there is a short window to strengthen the record before final determination.
• A reconsideration / second-level QIO brief is drafted and needs an adversarial pass before submission.
• Counsel or a patient advocate is about to review the supplement and the user wants to pre-empt the obvious criticisms.
• Anytime the user says the words "independent review," "stress-test," "red-team," or "what will they attack first?".

Not for a first-draft synthesis pass — this skill assumes a drafted supplement exists. If no draft exists yet, run /snf-appeal-supplement first.

Inputs to collect

1. Appeal supplement(s) to critique — markdown or docx path. Multiple files acceptable (e.g., a primary doc-hold supplement + earlier date-stamped supplements incorporated by reference).
2. Payer denial letters and QIO determinations — DENC, NOMNC, any prior reconsideration letters (favorable or adverse) on the same stay. These anchor both the "what must be defeated" side and the "internal inconsistency" side of the critique.
3. Source-exhibit directory — the HIPAA production or primary-record folder the supplement cites against. Used for independent verification.
4. Findings roll-ups (if available) — <prefix>_Findings.md, <prefix>_HotItems.md, <prefix>_Timeline.md, <prefix>_Gaps.md from /snf-review. Used to spot findings that SHOULD be in the supplement but aren't, and findings that ARE in the supplement but may be overstated.
5. Filing deadline — the doc-hold / reconsideration deadline. Drives the "minimum fix" cutoff in the final chat summary.
6. Output path — where to write the critique document. Default: <same-directory-as-supplement>/Independent_Reviewer_Critique.md.
7. Chat-summary word cap — default 400 words. The final chat reply must fit under the cap.

Preparation (in order)

1. Read ${CLAUDE_PLUGIN_ROOT}/skills/snf-records-review/RUBRIC.md to re-load the five-tag semantics (APPEAL / LTC-DISPUTE / MEDICAID / QUALITY / TIMELINE). The critique references these throughout.
2. Read ${CLAUDE_PLUGIN_ROOT}/skills/snf-appeal-critique/REVIEWER_PROMPT.md to load the hostile-reviewer persona and the four-part review framework (coverage standard, procedural compliance, evidentiary weight, clinical plausibility).
3. Read ${CLAUDE_PLUGIN_ROOT}/skills/snf-appeal-critique/CRITIQUE_TEMPLATE.md to load the exact 8-section output structure.
4. Read each drafted supplement end-to-end. For long files, use offset/limit passes rather than truncation.
5. Read the payer denials / QIO letters end-to-end. Note internal inconsistencies between determinations (same signer, same patient, verbatim language differences).
6. List every cited exhibit (filename + page) in the supplement. Build a verification backlog.

Verification pass (the core of the critique)

For each cited exhibit:

1. If the source PDF has a text layer, use pdftotext (or the pdf skill) to extract and grep the verbatim quote.
2. If the source is scanned without a text layer (common for SNF chart productions), use pre-extracted OCR output first — typical locations: <records-dir>/*/chart-part-*.txt, <records-dir>/findings/, <records-dir>/optimized-*/, or any pre-run OCR sidecars. Fall back to a targeted pdftotext / tesseract pass on the specific page range only if needed.
3. Record status per exhibit:
   • VERIFIED — full quoted text confirmed at cited source with the same author/date attribution.
   • PARTIAL — citation correct but a portion of quoted text (author name, drug name, specific numeric value, code value) not cleanly extractable from OCR and needs human visual confirmation on the scanned page. This is common and not disqualifying — flag and proceed.
   • NOT-FOUND — citation not corroborated at source. Treat as a hard finding; a hostile MD will attack any NOT-FOUND citation.
4. For declarations / witness statements cited as evidence, separately classify chain of knowledge:
   • Direct observation — declarant personally observed the event.
   • Contemporaneous receipt of report — declarant received a phone call / message from a witness during or immediately after the event. This is patient hearsay transmitted through a family declarant and is bottom-of-weight-tier evidence. Flag explicitly.
   • Non-contemporaneous report — declarant heard about the event later. Weaker still.
5. Track exhibits NOT cited in the supplement but present in the findings roll-up — these are strengthening candidates a hostile reviewer may argue the appellant avoided.

The four-part review framework (applied simultaneously)

1. Medicare / Medicaid coverage standard — MBPM Chapter 8 §§30.2 (skilled observation and assessment), 30.3 (care plan management), 30.4 (teaching and training); 42 CFR 409.31–409.35 (daily skilled services requirement); Jimmo v. Sebelius (2013) and the subsequent CMS policy revisions (maintenance-and-prevention-of-deterioration standard). Each coverage claim in the supplement must map to a specific provision.
2. Procedural compliance — 42 CFR 483.21(c) — (c)(2)(ii) final discharge summary referencing comprehensive assessment under §483.20(b)(1); (c)(2)(iii) medication reconciliation; (c)(2)(iv) post-discharge plan of care with resident representative participation. Procedural gaps are independently dispositive of any "safe plan" finding and must be explicitly tethered to the payer's own "safe plan" language.
3. Evidentiary weight ranking — facility-authored contemporaneous note (highest) > facility-authored retrospective note > payer-acknowledged statement > recorded multi-speaker IDT meeting > direct family observation > family declaration of patient phone report (lowest). Rank every piece of evidence in the supplement.
4. Clinical plausibility — read each clinical claim against the diagnosis, medication list, and event timeline. Flag claims that over-read the chart, smooth over artifact/edge-case readings, or elide contrary evidence elsewhere in the record.

Output document — 8 fixed deliverables

Follow CRITIQUE_TEMPLATE.md. The eight sections in order:

1. Reviewer Recommendation — OVERTURN / UPHOLD / REMAND, with 2–4 paragraphs of reasoning grounded in the four-part framework. State the recommended effective date (match the NOMNC date, not any later determination-letter date, to avoid one-day coverage gaps).
2. Top 5 Strongest Appellant Arguments — ranked by what would actually move a reviewer, not by appellant-preferred framing. Each with evidentiary-weight note.
3. Top 5 Weakest Arguments / Evidentiary Vulnerabilities — each entry strictly formatted: Vulnerability → Hostile use → Minimum fix before filing.
4. Top 5 Counter-Arguments a Hostile Payer MD Would Raise — written in first person from the MD perspective; each with the strongest response already in the record and the rebuttal-to-the-response the MD would attempt.
5. Exhibit Cross-Walk — table of every cited exhibit, VERIFIED / PARTIAL / NOT-FOUND status, and note. Target minimum 10 exhibits; more if the supplement cites more.
6. Missing Evidence — two columns: "would strengthen" (pursue via ROI) and "would weaken" (anticipate hostile production and pre-frame).
7. Devil's-Advocate Write-Up (under 500 words) — a first-person uphold-the-termination argument as a hostile MD would write it. Include the word count in parentheses at the end.
8. Final Integrity Check — numbered list of every factually incorrect, overstated, internally inconsistent, or OCR-unconfirmed claim in the supplement. Each with a minimum fix. Conclude with a one-line summary of whether the issues undermine the core recommendation.

Drafting guardrails

• The critique is adversarial on evidentiary rigor and neutral on the underlying coverage determination. The recommendation follows the record, not the appellant's interest.
• Every factual claim MUST be cited to source (exhibit file + page, or supplement section reference). No uncited assertions.
• Preserve verbatim quotes. Paraphrase only when necessary and flag paraphrase.
• Flag OCR-dependent claims for human visual confirmation before filing rather than guessing.
• Do NOT provide medical or legal advice — frame as "the record shows," "a hostile reviewer would argue," "the appellant's minimum fix is."
• Do NOT soften a weakness finding to spare the appellant. The whole point is to surface seams before the filing deadline.
• If a prior critique exists at the output path, read it first. If it contains a factual error you can correct, state the retraction explicitly in the Integrity Check — do not silently overwrite.
• Federal authorities apply in every state and may be cited directly: 42 CFR 409.31–409.35, 42 CFR 483.21(c), 42 CFR 483.25, the Medicare Benefit Policy Manual Ch. 8, and Jimmo v. Sebelius. Do not label these as state-specific.
• State-specific citations — state statutes, administrative codes, agency names, and state waiver-program names (e.g. a state HCBS waiver such as Ohio's PASSPORT) must carry the [user: verify before submission] marker, the same pattern as snf-appeal-supplement. If a cited state authority cannot be verified, treat it as a seam a hostile reviewer will find and flag it in the critique.
• PHI stays in the critique file and chat — do not leak to unrelated outputs.

Chat reply conventions

After writing the critique file, the final chat reply must:

1. State the recommendation (OVERTURN / UPHOLD / REMAND) in the first sentence.
2. Identify the three most important fixes the appellant should make before the filing deadline — numbered, concrete, action-ready.
3. Give the absolute file path to the critique document (both the workspace path and the user-facing /Users/... path if known).
4. Stay under the word cap (default 400 words).

Do NOT recap the full 8-deliverable document in chat — the file is the deliverable; the chat is the cover sheet.

Typical workflow

1. /snf-review            → <prefix>_Findings.md + _HotItems.md + _Timeline.md + _Gaps.md
2. /snf-appeal-supplement → draft appeal supplement .docx / .md
3. /snf-appeal-critique   → Independent_Reviewer_Critique.md  (this skill)
4. (user fixes items from §8 Integrity Check before deadline)
5. Final supplement filed.

The critique skill can be re-run after fixes to confirm the seams are closed — call it "second pass" in the header and reference the prior critique by filename.

Guardrails (repeat for emphasis)

• This is a draft critique for the user's review and a lawyer / patient advocate's review before the filing. Not filed work product.
• Not medical advice. Not legal advice. The reviewer posture is independent clinical utilization review.
• Where the record is ambiguous, the critique flags ambiguity rather than resolving it — the user (and counsel) decide how to frame ambiguity in the filing.