Sources and review notes
These links support workflow-level research and do not establish the regulatory status, clinical safety, diagnostic performance, or suitability of any product.
CMS describes Medicare medical review as examination of medical records and related information to determine whether a claim meets applicable coverage, coding, billing and medical-necessity requirements, and it may request documentation supporting an item or service reported on a claim. The Medicare Program Integrity Manual addresses review, data analysis, potential errors, corrective action, overpayment estimates and related program-specific processes. CMS's NCCI program promotes correct coding and reduces improper payment for defined Medicare Part B and Medicaid claim contexts through versioned policies and edits, but its Medicare and Medicaid materials have different scope and no edit set contains every possible coding or coverage rule. CMS remittance guidance provides adjudication and adjustment evidence after processing; payment or denial does not by itself establish that documentation, charge, code, contract interpretation, patient responsibility or net revenue is correct. HHS-OIG's General Compliance Program Guidance discusses voluntary, nonbinding compliance infrastructure and federal risk areas; it does not certify an AI product or supply coding, billing or legal advice. These sources do not establish one universal revenue-integrity standard, apply every Medicare rule to commercial payers, or make an AI-detected opportunity billable, supported, compliant, collectable or clinically accurate. Buyers should define authoritative requirements by legal entity, payer and plan, contract and network, enrollment, provider and specialty, facility and department, setting, service date, encounter and charge type, code system and release, coverage and coding policy version, fee schedule, documentation standard, order and authorization, and responsible clinical, coding, billing, compliance and finance owner. Preserve the original clinical record and orders, service and supply evidence, device or system event where relevant, charge-master and pricing version, code and modifier, units, diagnosis and procedure relationship, claim and correction history, payer acknowledgements, policy and contract citations, remittance and deposit, patient adjustments, refunds, recoupments, audit requests and findings, and final disposition. Every AI finding should link to exact source evidence and distinguish missing documentation, missing or duplicate charge, coding question, payer edit, contract variance, denial risk, overpayment risk, compliance concern and data-quality issue; it should retain model or rule version, confidence, assumptions, reviewer role, edits, approval or rejection reason, downstream action, submitted record and final financial and audit outcome. AI may surface discrepancies and assemble evidence, but qualified clinicians must own clinical facts and attestations, qualified coders and billing staff must determine codes and submissions, and contracting, compliance, legal, accounting and audit owners must approve their respective interpretations and actions. Acceptance testing should use independently reviewed known-answer cases across specialties and sites, professional and institutional claims, inpatient and outpatient care, bundled and separately payable services, supplies and drugs, timed and unit-based services, modifiers, add-on codes, status indicators, orders and signatures, incomplete or contradictory records, late and amended documentation, canceled and duplicate charges, missing interfaces, code and policy changes, multiple payer edits, secondary insurance, retrospective authorization, corrected and voided claims, denials and appeals, partial payments, overpayments and takebacks, refunds, audits, source downtime, and encounters that appropriately produce no opportunity. Test both underbilling and overbilling signals and block action when source evidence is missing, identity or encounter matching is uncertain, versions conflict, the recommendation changes clinical meaning, or required review is absent. Measure finding precision and recall by type, unsupported-suggestion rate, reviewer agreement and edits, duplicate and false-opportunity rate, missed undercharge and overcharge cases, documentation queries and response, charge and coding lag, submitted and accepted corrections, initial and final denials, confirmed recoveries net of refunds, fees and recoupments, avoided overpayments, audit findings, patient-balance corrections, staff time, complaints and compliance or safety events. Report gross identified, qualified, submitted, adjudicated, collected and retained amounts separately with stable denominators and sufficient payment and recoupment runout. More charges, diagnoses, relative weight, accepted suggestions, gross opportunity or short-term payment does not by itself establish accurate documentation, correct coding, compliant revenue, sustainable net recovery, improved care or causation. Governance should include bidirectional auditing, independent sampling of accepted and rejected findings, subgroup and specialty review, version and change control, conflict-of-interest and incentive review, role permissions and separation of duties, immutable source and decision logs, PHI safeguards, retention and legal hold, incident escalation, rollback, export and vendor exit. Systems must not invent or alter clinical facts, prompt clinicians toward unsupported documentation, code from inference without source support, suppress overpayment or unfavorable findings, silently change charge or claim records, submit corrected claims or appeals autonomously, or initiate write-offs, refunds, transfers or patient bills without accountable controls.
