Artificial intelligence (AI) has tools such as natural language processing (NLP) that have been integrated into a wide range of applications, including risk-adjusted coding tools to improve efficiency and accuracy in the healthcare industry. For Medicare Advantage (MA) programs, these tools can significantly improve the accuracy of diagnostic and hierarchical status category (HCC) data required to support a risk adjustment program and help ensure appropriate reimbursement.
Prepare for new RADV changes
With NLP-driven tools, MA plans to discover errors during a retrospective chart review and then perform a Risk Adjustment Data Verification (RADV) audit. Once only 10% of the MA plan is required per year, RADV audits affect all MA plans as the centers of Medicare & Medicaid Services (CMS) are gradually working to reduce overpayments.
As part of its active strategy, CMS will also review more records – up to 200 records per program. Policy changes highlight the need for accuracy and efficiency of MA programs.
The RADV audit extension follows other significant policy changes and now allows CMS to infer its audit results from a sample of medical records reviewed to the entire planned contract – if the agency decides that records do not adequately support participants’ diagnosis, it could put millions of single contracts at millions of risk. The cancellation service (FFS) regulator also adds the burden on the program to ensure accurate, complete HCC reports or risk inference fines.
How AI helps MA program
For MA programs that have not previously been audited for RADV, these changes provide a timely opportunity to incorporate AI into its coding practices and establish appropriate policies and procedures with technology.
By bringing AI-enabled tools into their workflows, MA programs can prioritize critical documents and ensure their coding teams focus on the most relevant lengthy, complex areas of medical records. For example, these tools can easily identify common errors, such as HCC reported in a wrong environment (Inpatient vs. Outpatient) or wrong specialties. NLP enabled tools can also help coders quickly find instances where medical records from two different members are not accidentally merged, causing inaccuracy for retrospective chart review or RADV chart submission process.
Strategies for launching AI support tools
Here are the best practices for planning as they implement AI-enabled tools to improve the accuracy of their coding and risk-adjusting plans.
- Establish an AI Governance Committee for Human Supervision. Plans should establish a framework to review and monitor new uses of AI or NLP in their organizations. By establishing a governance committee of clinical, technical and coding experts, programs can review different use cases of AI and have a forum to raise concerns about potentially inappropriate uses. To guide organizations in healthcare and other industries, the responsible AI Institute provides best practices for AI governance structures, as well as principles for reviewing AI projects. Following the guidance of industry advocacy groups can help leaders ensure that AI is ethical in coding and other areas.
- Create a “sandbox” environment for encoder testing tools. Provide coders with test documentation so that they can try the tool that can help them practice the workflow they will encounter in real life. The plan can also provide a user manifest to help the encoder simulate various scenarios and record any issues related to performance or availability.
- Release scorecards with metrics to measure performance overall. Leaders should maintain an ongoing commitment to assess the performance of AI-enabled tools. Plans should fully view their performance and track overall and individual productivity and accuracy metrics. Potential red flags are encoders, which are very slow or fast when using AI tools compared to peers. Plans should also look for signs of over-reliance on AI, such as coders who accept recommendations for AI generation almost 100% of the time. The specific benchmarks set by the plan should depend on its business family, the type of software used, and whether data is retrieved from an electronic medical record (EMRS) or scanned PDF records. Plans should review their metrics at least monthly to identify opportunities for improvement and share results with key stakeholders.
- Leverage end-user feedback for continuous improvement. Soliciting feedback from the encoder is essential to ensure a positive user experience. Sometimes, creating too many suggestions for coders can slow them down, hinder productivity and create frustration. Let coding “superusers” make suggestions to managers, and leaders can continuously improve technology and procedures.
- Maintain performance expectations with suppliers. If you plan to leverage AI-enabled software through a coding partner, you should have performance guarantees related to system performance, uptime/downtime metrics, and NLP accuracy, with deadlines and potential delay fines. This can help maintain plans to resist system disruptions and other issues that may derail their project deliverables and reporting deadlines.
Prepare for new CMS audit work
As CMS strengthens its RADV program in the coming months, plans should ensure that their risk adjustment programs meet the highest standards of accuracy and compliance. AI-enhanced anticipation and retrospective analytics can help programs work with providers to optimize documents at the point of care and identify coding errors during audit preparation. Plans may also consider secondary reviews of coding results, which enables them to correct unsupported HCCs that can be easily overlooked in first-level comments. By combining AI-enabled tools with expert oversight, programs can improve the success of these efforts as they encounter greater regulatory oversight in the future
Photo: Thanakorn Lappattaranan, Getty Images
Katie Sender, MSN, RN, PHN, CRC is Vice President of Clinical and Coding Services, Cotiviti. Katie has over 25 years of healthcare experience and is responsible for ensuring leadership and management oversight of teams across the globe by managing key performance indicators related to clinical and coding solutions.
This article passed Mixed Influencer program. Anyone can post a view on MedCity News' healthcare business and innovation through MedCity Remacence. Click here to learn how.
