Clearing the backlog: How AI can transform CMS medical review

Medical review is the operational engine for CMS’ program integrity activities. It’s also where the biggest bottlenecks in the fraud, waste, and abuse investigation process happen. The legacy manual review process is labor-intensive, with clinicians manually sifting through reams of documentation to substantiate medical claims. On the other side, providers wait for months or longer to confirm the outcome of those reviews.

It is essential for CMS and other federal and State agencies to improve these processes, employing artificial intelligence and other technologies to meet these challenges while maintaining human oversight. In the context of CMS, improving medical review will allow investigators to better triage potential cases, resulting in more recoveries and reductions in health care fraud, waste, and abuse.

Augmenting human medical reviewers—not replacing them

The most successful approaches to implementing advanced technology into medical review augment human coders’ capabilities. AI can address high-volume, repetitive tasks such as document ingestion, classification, data extraction, and pre-population of coding templates. Automating this work allows human coders to spend their time on activities requiring professional expertise, such as applying clinical judgment, resolving ambiguities, and ensuring compliance.

In this technology-assisted approach to medical review, AI-generated recommendations are accompanied by clear explanations, source references, and confidence indicators, enabling human coders to make informed decisions quickly. This reduces coders’ cognitive burden, increases throughput, and improves consistency across large review populations.

Streamlining data ingestion, classification, and analysis

Technology-assisted medical review solutions must address data ingestion, document classification, and data analysis as a single, connected workflow. Pairing interoperable data access with AI-enabled tools creates a more efficient end-to-end process.

Data ingestion is foundational because medical review depends on the completeness and accuracy of its records and how quickly they can be used. In practice, that means reliably pulling clinical data from fragmented EHR environments in batches. Standards-based approaches (such as FHIR APIs) and secure, FedRAMP-authorized cloud environments are demonstrating they can support high-volume retrieval and processing while meeting CMS security and compliance requirements. The advantage is faster access to complete records, quicker review cycles, and stronger audit readiness.

The next step is document classification: separating a complex record package into its constituent parts (e.g., physician orders, lab results, clinical notes, discharge summaries, imaging reports, and other supporting documentation). This classification layer turns an unstructured or inconsistently organized record into a navigable review file before any AI-driven recommendation is made, ensuring those decisions are explainable and defensible. Combining natural language processing, machine learning, and rules-based validation allows teams to extract relevant clinical information while maintaining visibility into how conclusions are reached.

Finally, AI tools can bring efficiency and transparency to record analysis. Pre-populated coding templates, linked directly to supporting documentation, allow reviewers to validate or adjust recommendations efficiently. Instead of building cases from scratch, reviewers can focus on accuracy and edge cases. Side‑by‑side views of AI‑generated outputs and reviewer determination, supported by confidence indicators and audit trails, can make discrepancies visible early. This allows teams to prioritize high‑risk cases, apply resources where they matter most, and strengthen the defensibility of findings.

At each step, clear separation between AI-generated suggestions and human decisions preserves accountability. Expert review becomes faster, more consistent, and easier to defend.

Where this technology is already at work

This kind of AI‑supported review workflow is already being applied in adjacent federal contexts where accuracy, transparency, and speed are equally critical. Under congressional mandate, agencies responsible for federal rulemaking must process and analyze millions of public comments before issuing final rules, often under tight deadlines following the close of public comment periods.

In this environment, AI‑enabled review tools have been used successfully to accelerate analysis while preserving human oversight, auditability, and trust in the results. These implementations demonstrate that large‑scale, high‑stakes review processes can be modernized without sacrificing explainability, a precedent with clear relevance for CMS medical review.

This integrated approach is more adaptable to evolving requirements from federal agencies (including CMS), states, and commercial plan sponsors, while also better aligning with real-world coding operations. Notably, the underlying technology is already being used by agencies responsible for the federal rulemaking process.

ICF has recently worked with a federal client to accelerate the comment review process by incorporating AI into the workflow. Using a fusion team of data science experts, AI practitioners, and user experience specialists, we delivered a FedRAMP-compliant, cloud-based, AI-powered solution. We also pressure-tested the solution with federal policy experts to ensure its results were accurate, ethical, and aligned to how human regulatory experts would classify the comments. The result: A solution that processes, analyzes, and summarizes comments at scale, producing outputs that agency leaders can trust.

Faster reviews, stronger cases, better outcomes

By reducing the friction that slows medical review today, AI‑supported approaches can help CMS move cases faster without compromising clinical judgment or oversight. Reviewers spend less time searching for information and more time making defensible decisions. Investigators receive stronger, better‑documented cases, while providers avoid prolonged periods of uncertainty.

The decision for CMS leaders is not whether to automate medical judgment, but whether to continue relying on processes that constrain capacity and delay outcomes. As review volumes grow and enforcement expectations increase, modernizing medical review is becoming less of a technology upgrade than a program integrity imperative.