hh-health-ai/healthcare-equity/coverage-check

Coverage Check

Determine the Medicare coverage position for a named drug, device, diagnostic, or procedure and translate it into model variables.

Workflow

1. Frame the ask. Identify the product, the clinical use, the likely site of service (inpatient, HOPD, ASC, office, lab, home), and the ticker(s) exposed. If ambiguous, state the assumption and proceed.
2. Search the Medicare Coverage Database via the CMS Coverage connector: national coverage first (NCDs), then local (LCDs and coverage articles by MAC). Capture: policy ID and title, national vs local, effective/retirement dates, covered indications and restrictions (patient criteria, physician/site requirements, frequency limits, documentation), and Coverage with Evidence Development (CED) conditions if any.
3. Classify the position — one of: covered nationally · covered locally (name the MACs and note geographic gaps) · covered with restrictions/CED · explicitly non-covered · silent (no policy — contractor discretion; for new technology, silence plus a new code often means case-by-case review).
4. Check for motion. Look for proposed NCDs/reconsiderations, draft LCDs, or recent coverage articles that signal a change window; note the comment/decision dates as catalysts for the catalyst calendar.
5. Label the universe. State clearly this is Medicare FFS coverage; note that MA plans must cover what NCDs/LCDs cover (but can add utilization management), and commercial coverage is separate — if the thesis needs it, do targeted web research on the top commercial payors' medical policies and mark that as a separate evidence line.
6. Translate. Per the coverage row of ${CLAUDE_PLUGIN_ROOT}/references/metrics-payor.md: coverage evidence moves authorization, abandonment, paid conversion, and net price/accessible population. It does not automatically move clinical adoption or prescribing — say so.
7. End with the EVIDENCE BRIEF block (contract in the ${CLAUDE_PLUGIN_ROOT}/CLAUDE.md), with retrieval date and policy vintages. For payment rates, hand off to the reimbursement-impact skill — coverage and payment are different questions.

Quality bar

Quote restriction language exactly when it drives the model (e.g., patient-criteria clauses that halve the eligible population). Never infer coverage from the absence of a policy without labeling it "silent/contractor discretion". If results are thin or ambiguous, say "missing" and name the data point that would settle it (e.g., MAC call, payor policy portal).

REMS/ETASU friction check (v0.2)

For drug products, check whether a REMS program — especially ETASU elements (prescriber/pharmacy certification, patient registries, required monitoring, restricted distribution) — sits between coverage and paid demand. REMS friction is quantifiable access friction: per the coverage translation row, it moves authorization, abandonment, and paid conversion, and functions as an accessible-population/uptake-pace haircut in the funnel (not a qualitative caveat). Note REMS changes (additions, modifications, removals) as dated demand-side events for the catalyst feed, and cross-reference the clinical-catalysts quality-signals skill when a safety development could trigger one.