Category: Uncategorized

Key Takeaways

• HealthEdge® partnered with Ellipsis Health to integrate Sage, an AI-powered voice agent, directly into the GuidingCare® platform—giving health plans a scalable way to automate member outreach.
• Sage uses clinically validated conversational AI and an Empathy Engine to conduct emotionally intelligent phone conversations with members across a range of care management programs.
• Key use cases include health risk assessments (HRAs), post-discharge follow-up, risk profiling, care gap closure, and member enrollment outreach.
• The partnership directly addresses clinical workforce shortages by automating high-volume, routine outreach and freeing nurses to focus on complex, high-acuity cases.
• GuidingCare customers can deploy Sage quickly through a referral partnership, with conversation flows configurable to their specific clinical goals and workflows.

Scaling Smarter with Sage and GuidingCare

Health plans are under pressure to meet increasing demands for member engagement, care coordination, and improved clinical outcomes. But the ongoing healthcare workforce shortage makes it difficult for payers to scale resources.

The result is familiar: care managers are stretched thin by outreach and administrative requirements, reducing the time they can spend proactively engaging with patients and assessing potential gaps in care.

Payers need a scalable way to automate member outreach. That’s why HealthEdge® partnered with Ellipsis Health to integrate Sage, an AI-powered voice agent for care management, directly into the GuidingCare platform. Sage enables health plans to

automate outbound calls, broaden member reach, drive proactive health management, and lower administrative costs.

What Is AI-Powered Virtual Nursing?

AI-powered virtual nursing tools use conversational AI to conduct phone-based outreach with health plan members. Unlike traditional automated calls or Interactive Voice Response (IVR) systems, AI-powered voice agents like Sage are designed to hold natural, emotionally intelligent conversations—and adapt in real time based on how a member responds.

How Sage Works—and How it Stands Out

The new partnership between HealthEdge and Ellipsis Health makes AI-powered virtual nursing available to GuidingCare users. Sage is integrated within the GuidingCare ecosystem, allowing it to draw from existing member information to personalize each conversation and deliver updated information from these interactions back into the platform.

For health plans already using GuidingCare or Wellframe, it’s a seamless extension of their existing care management infrastructure.

Sage is not a scripted phone bot. Instead, it mirrors human conversation—including intonations and natural pauses—using Ellipsis Health’s patented sentiment matching technology, which was trained using millions of live clinical patient calls. This allows Sage to engage in active listening and interviewing, and adjust its responses according to member cues.

What does this look like for health plan members? During conversations, Sage can:

• Deliver affirmations and acknowledgements that reflect active listening
• Recognize emotional cues, like irritation, sadness, or confusion
• Calmly redirect conversations to stay focused on the goal of the call

Why is this important for health plans? Because member engagement is more than a checklist of questions asked over the phone. It’s about building a trusting relationship with members so they feel supported and actively involved in their own health.

Where Health Plans Can Put Sage to Work

Sage helps payers address the highest-volume, most time-intensive outreach tasks care management teams face. Configurable to each health plan’s clinical goals and workflows, Sage supports:

• Clinical workforce optimization: Extend the reach of care managers with automated routine outreach and empower staff to focus on high-acuity cases and help support more members.
• Predictable cost savings: The technology behind Sage enables cost-efficient, scalable outreach and delivers rapid ROI with improved engagement outcomes.
• Seamless ecosystem integration: GuidingCare customers can quickly deploy this AI solution via referral partnership, enabling customized engagements based on health plan’s individual goals, as well as Medicare and Medicaid requirements.
• Health risk assessment (HRAs) completion: Scale outreach to meet regulatory, quality, and risk adjustment requirements. Sage conducts the full HRA in a single outreach call and adds updates to GuidingCare automatically.

Sage gives care management teams back hours of time per week—time that can be redirected toward the members who need the highest levels of clinical expertise and intervention.

“As a nurse, I’ve seen and experienced what happens when care managers are stretched thin,” says Marla Kraak, Product Director of Care Management at HealthEdge®. “Members fall through the cracks—not because anyone stopped caring, but because there simply aren’t enough hours. An AI voice agent that can handle routine HRA outreach and flag members who need a real conversation isn’t replacing the nurse. It’s making sure the nurse shows up where it counts.”

Scaling Engagement Without Sacrificing Quality

A core challenge in care management is finding the right balance between scale and quality. Reaching more members matters, but so does ensuring those interactions remain meaningful, effective, and capable of driving better outcomes at a lower total cost of care. Sage is built to engage entire member populations simultaneously, enabling health plans to scale high-volume outreach far beyond what manual efforts alone can support. With conversations that are personalized, clinically grounded, and emotionally aware, Sage helps plans identify needs earlier, close gaps in engagement, and guide members toward the right interventions before issues escalate into more costly utilization.

For health plan decision makers focused on lowering total cost of care, that combination matters. Timelier outreach, better member navigation, and stronger follow-through can help reduce avoidable emergency department visits, unnecessary utilization, and delays in care—while supporting more efficient use of care management resources.

For health plan IT leaders, integration details are equally important. Sage connects directly to the GuidingCare platform, allowing member data and outcomes to flow seamlessly between systems without manual reconciliation or another disconnected point solution to manage. The result is a more unified platform that supports both operational efficiency and more cost-effective care management at scale.

A Practical Path Forward for Care Management Leaders

With Sage, the combination of clinical intelligence, emotional design, and seamless platform integration empowers health plans to extend their existing resources and better support members across the risk spectrum.

GuidingCare customers: Want to learn more about elevating your care management strategy with Sage? Contact your Customer Success Executive for a demo.

Not yet a HealthEdge customer? Download the Care Solutions data sheet to learn how your health plan can leverage AI-driven tools within your care management operations.

Frequently Asked Questions

What is Sage, and how does it differ from a standard automated phone system?

Sage is an AI care management voice agent developed by Ellipsis Health. Unlike traditional IVR or scripted call systems, Sage uses a patented Empathy Engine to analyze vocal cues in real time and adapt its responses based on emotional signals. It’s trained on millions of live clinical patient calls, making it capable of holding natural, clinically informed conversations with members.

How does the HealthEdge and Ellipsis Health integration work?

Sage integrates directly with HealthEdge’s GuidingCare platform. It draws on member data stored in GuidingCare to personalize outreach conversations, and it returns outcomes back into the platform once calls are completed. This creates a closed-loop workflow with no manual data reconciliation required.

What types of care management programs can Sage support?

Sage supports a range of high-volume outreach programs, including health risk assessments, post-discharge follow-up calls, risk profiling, care gap closure, and member enrollment outreach. Conversation flows are configurable to each health plan’s specific clinical goals and compliance requirements.

How does this partnership address clinical workforce shortages?

By automating routine, high-volume outreach tasks, Sage frees nursing staff to focus on complex, high-acuity cases that genuinely require clinical expertise. Health plans can engage entire member populations simultaneously without adding headcount, making care management programs more scalable and cost-efficient.

Is this solution compatible with Medicare and Medicaid programs?

Yes. Sage is designed to be highly configurable, allowing health plans to align conversation flows to the specific requirements of Medicare, Medicaid, and other program types. GuidingCare customers can work with their Customer Success Executive to configure deployments accordingly.

What is a health risk assessment, and why does it matter for health plans?

A health risk assessment (HRA) is a structured set of questions designed to evaluate a member’s general health status, lifestyle factors, and potential care gaps. For payers, HRAs are critical for risk adjustment, quality reporting, and regulatory compliance. Completing them at scale—and accurately—directly affects a health plan’s risk profile and program performance.

How quickly can GuidingCare customers deploy Sage?

Sage is available to GuidingCare customers through a referral partnership, designed for efficient deployment. Health plans can contact their HealthEdge Customer Success Executive to begin the process and explore configuration options aligned to their care management strategy.

For many health plans, HEDIS® final submission feels like the finish line. In reality, it’s the moment when months of quality improvement efforts, chart retrieval activities, and data validation processes are put to the test. A single discrepancy can affect reported performance, compliance outcomes, and ultimately the quality ratings that influence revenue and member growth.

To help health plans navigate this important stage, we’ve outlined best practices for submission readiness, key quality assurance (QA) activities, and lessons learned to strengthen future HEDIS cycles.

Laying the Groundwork for Success in 3 Steps

The most successful HEDIS submissions don’t begin in the weeks leading up to the deadline — they’re built on a foundation of preparation, organization, and cross-functional collaboration throughout the measurement year.

As the final submission approaches, health plans should focus on the following three critical areas:

1.     Establish a Clear Submission Timeline

Submission deadlines often create a flurry of activity across quality, operations, analytics, compliance, and vendor teams. Without a structured timeline, critical validation and review activities can become compressed, increasing the risk of errors.

Leading organizations establish milestone-based submission plans that include:

• Data validation checkpoints
• Medical record review completion targets
• Measure-level review sessions
• Compliance reviews
• Executive sign-off milestones
• Submission readiness assessments

Organizations that implement structured submission timelines often see a marked reduction in submission errors, highlighting the value of proactive planning and accountability.

2.     Reconcile and Validate All Data Sources

As HEDIS measures increasingly rely on supplemental data, electronic clinical data systems (ECDS), and multiple reporting sources, ensuring consistency across datasets has become more challenging and more important than ever.

One of the most common challenges during final submission is ensuring consistency across multiple data sources. Claims data, pharmacy records, encounter information, laboratory results, supplemental data, and medical record review findings must all align before submission. Even minor discrepancies can create downstream reporting issues and require costly last-minute remediation efforts.

Health plans should perform comprehensive reconciliation activities to confirm:

• Measure calculations are accurate
• Supplemental data is properly integrated
• Member eligibility files are current
• Medical record review findings are reflected appropriately
• Vendor-delivered data aligns with internal reporting

Strong retrieval performance also plays a significant role in submission success. Across its client base, HealthEdge Quality360™ has achieved retrieval rates exceeding 95%, helping plans maximize data completeness and reduce reporting gaps.

3.     Align Stakeholders Early

Success requires coordination across multiple departments, including quality improvement, operations, compliance, analytics, IT, provider engagement, and executive leadership.

Establishing clear ownership and accountability before submission deadlines helps ensure everyone is aligned on reporting requirements, measure interpretations, and validation responsibilities.

The Power of Precision: QA in Action

As submission deadlines approach, quality assurance becomes the final safeguard against reporting errors and compliance risks. To ensure submission readiness, health plans should go beyond basic validation and incorporate measure-level reviews, documentation verification, and regulatory compliance checks into their final QA process. Let’s dig into each of the comprehensive QA best practices:

Conduct Comprehensive Measure Reviews

To help identify anomalies before they impact final results, every measure should undergo a conclusive review to validate:

• Numerator and denominator accuracy
• Exclusions and exceptions
• Supplemental data inclusion
• Medical record review outcomes
• Measure-specific logic and calculations

Leverage Automation to Improve Accuracy

Manual review processes remain important, but automation can significantly improve efficiency and consistency by helping health plans:

• Identify data discrepancies
• Detect incomplete records
• Validate measure calculations
• Highlight compliance concerns
• Reduce manual review effort

HealthEdge’s automated QA capabilities have helped improve data accuracy by 30% while reducing submission preparation time by 15%.

Verify Regulatory Compliance

Compliance validation is often one of the last and most important steps before final submission. Health plans should verify alignment with current NCQA requirements, audit standards, and submission guidelines to minimize risk and ensure confidence in reported results.

HealthEdge clients have achieved a 100% medical record review validation pass rate, underscoring the importance of comprehensive QA and audit-readiness processes.

Building a Smarter Future: Lessons from the Field

Organizations that consistently improve quality outcomes often treat each submission cycle as a learning opportunity to identify what worked, what didn’t, and how to improve future performance.

They view their final submission not as the end of the measurement year, but as the beginning of preparation for the next one. Here’s how they do it:

Capture Lessons Learned

To help teams refine workflows and avoid repeating challenges in future years, conduct structured reviews after submission to document:

• Process bottlenecks
• Data quality challenges
• Resource constraints
• Vendor performance
• Successful strategies and interventions

Gather Feedback Across Teams

Soliciting feedback from quality teams, analysts, auditors, provider engagement teams, and operational stakeholders can uncover opportunities for improvement that may otherwise be overlooked.

Turn Insights into Action

Continuous improvement requires more than documenting lessons learned. It requires acting on them.

Health plans that systematically incorporate lessons learned into future quality programs have achieved a 20% improvement in Star Ratings within two years. The impact can be even more dramatic when organizations combine process improvements with targeted quality strategies.

Through HealthEdge Stars Consulting, one health plan achieved a full 1-Star gain in a single year, a milestone accomplished by only 1.5% of plans.

Turning Final Submission into Future Success

Final submissions are an opportunity to validate the quality initiatives, operational processes, and member engagement efforts that took place throughout the year.

By establishing structured submission processes, implementing rigorous quality assurance practices, and capturing lessons learned for future improvement, health plans can reduce risk, strengthen compliance, and improve the accuracy of their reported results.

As HEDIS requirements continue to evolve and quality performance becomes increasingly tied to financial outcomes, organizations that approach final submission with discipline and precision will be better positioned to achieve stronger quality results, higher Star Ratings, and better member outcomes.

Discover how HealthEdge Quality360® and Stars Consulting can help your organization improve data accuracy, strengthen compliance, and maximize quality performance.

Download the brochure, HealthEdge Quality360: The Next-Generation Integrated HEDIS® Engine & Analytics Platform.

Key Takeaways

• For health plans, the implementation period is one of the highest-risk, highest-stakes phases of any technology investment—and AI can significantly reduce that risk.
• HealthEdge Delivery Services integrates AI-powered tools across the 4 key stages of solution implementation: Discovery, Configuration, Testing, and Deployment and Stabilization.
• The combination of deep domain expertise and AI-enhanced tooling means health plans get faster time to value without sacrificing quality or compliance.

How HealthEdge Applies AI Across 4 Key Stages of Implementation

For health plans, selecting the best-fitting technology solution is only half the battle. How payers approach implementation can drastically affect adoption timelines and time to value. Stakes are high, and new implementations impact critical areas like claims processing, benefit configuration, provider relationships, and regulatory compliance all at once.

Historically, common challenges include long implementation timelines, troubleshooting and rework, varying levels of implementation consultant expertise, and late discovery of requirement gaps. But at HealthEdge, our Delivery Services team applies AI where implementation work has traditionally been highest in effort, variation, and risk — across four key stages: Discovery & Requirements Analysis, Configuration & Build, Testing, and Deployment & Stabilization.

In addition, Delivery Services maintains a human in the loop model when using AI. AI outputs are treated as intelligent drafts that experienced implementation consultants validate and refine — preserving quality while enabling speed.

Stage One: Discovery and Requirements Analysis

Discovery and Requirements Analysis is the stage that sets the foundation. This is where HealthEdge teams and health plan leaders define project scope, align stakeholders, and identify initial requirements.

The Challenge

Health plans must provide a significant amount of information up front, often under time pressure. Workflows may not be fully documented, and critical details can be spread across multiple teams and systems. For HealthEdge, implementation consultants must quickly absorb large volumes of legacy documentation, a time-consuming process where gaps or missed information can result in costly rework later.

How HealthEdge uses AI During Discovery and Requirements Analysis

HealthEdge implementation consultants use AI tools to analyze the intake documentation and automatically pull out the relevant information, cross-reference requirements against platform capabilities to identify any gaps, and surface any potential compliance issues early.

In addition, rather than arriving to a session with a blank questionnaire, our implementation consultants bring a pre-populated toolkit of information based on an AI-assisted analysis of the health plan’s existing documentation, public regulatory information from each state or the Centers for Medicare and Medicaid Services (CMS), and comparable prior implementations. The session is spent validating and refining, not capturing potentially repetitive information.

What Health Plans Can Expect

• Initial discovery timelines reduced by 50% (from 8 weeks to 4)
• Less administrative burden on the health plan team
• Earlier risk identification
• A stronger foundation for the stages that follow

Stage Two: Configuration and Build

During the configuration stage, the HealthEdge implementation team translates Discovery requirements into a functioning system — configuring benefit structures, building integrations, establishing workflows, and creating business rules that govern how the health plan operates.

The Challenge

The configuration and build stage is complex and requires significant manual effort. Teams spend considerable time drafting configuration artifacts, creating business rules, and translating documentation into system setup.

How HealthEdge uses AI During Configuration and Build

Teams start with intelligent drafts that experienced implementation consultants review, refine and validate – eliminating the need to build from scratch. For GuidingCare, this includes generating draft business rules and translating specifications into ready-to-review forms and workflows. For HealthRules® Payer, the Delivery team uses AI to accelerate the conversion of plan artifacts into configuration-ready inputs – extracting benefits and plan provisions from source documents and generating draft configuration logic.

What Health Plans Can Expect

• Faster time to value
• Reduced manual work
• Greater consistency across modules
• Lower risk of errors reaching production

Stage Three: Testing

Testing is where the solution comes to life. The HealthEdge implementation team ensures the configured system meets the health plan’s operational requirements and performs as expected across real-world scenarios—before it impacts members, providers, and operations.

The Challenge

Testing is time-consuming and resource-intensive, and many health plans have limited resources to dedicate. Both the HealthEdge implementation team and the health plan users must validate numerous workflows, integrations, and benefit configurations under strict timelines. Traditional manual testing approaches often leave coverage gaps and allow defects to surface too late in the process.

How HealthEdge uses AI During Testing

Testing is where AI delivers some of its most measurable impact. Delivery Services uses AI to automatically generate test cases, produce regression tests when rules or workflows change, and identify recurring defect patterns early. For upgrades, AI maps what has changed to a health plan’s specific workflows, so teams know exactly where to focus testing.

What Health Plans Can Expect

• Faster testing cycles
• More comprehensive coverage
• Reduced workload for health plan teams
• Lower go-live risk.

Stage Four: Deployment and Stabilization

During Deployment and Stabilization, the HealthEdge implementation team transitions system management to the health plan, monitors performance, and maintains high responsiveness to ensure a stable operational rollout.

The Challenge

Health plan customers and the HealthEdge Implementation team must triage and prioritize issues rapidly while keeping leadership informed, which can cause significant pressure.

How HealthEdge uses AI During Deployment and Stabilization

HealthEdge uses AI to keep implementations on track and issues from escalating. The team leverages AI to help monitor project risks, triage defects by severity and business impact, and analyze patterns across issues to shift teams from reactive troubleshooting to proactive quality management.

What Health Plans Can Expect

• Faster issue resolution
• Shorter stabilization periods
• Lower operational disruption
• Faster realization of business value

How HealthEdge Implementations Position Payers to Achieve Real Results

AI-enhanced delivery directly impacts outcomes health plan leaders care about across the implementation lifecycle:

• Faster time to value: Leveraging AI helps accelerate every stage of implementation, so health plans can realize the benefits of their investment sooner.
• Higher quality outcomes: Standardized patterns, automated validation, and broader test coverage help reduce rework and lower the risk of downstream adjustments.
• Lower-risk, scalable delivery: AI tools help Implementation teams and users combine platform knowledge and institutional experience to streamline adoption.
• A delivery model built for the future: For HealthEdge, the human-in-the-loop approach makes sure AI enhances expert judgment to give health plans a partner equipped to handle complexity without sacrificing rigor.

Ready to See AI-Enhanced Delivery in Action?

Interested in learning how AI-enhanced delivery can accelerate your next implementation? Existing customers, reach out to your HealthEdge Customer Success Executive to get started.

Want to learn more about how HealthEdge is leveraging AI across internal functions and within our award-winning solutions? Watch the recent webinar with HealthEdge Chief Technology Officer Rob Duffy, “AI Capabilities: Transforming Payer Strategies with HealthEdge.”

 

Frequently Asked Questions about HealthEdge Implementation and AI

What is HealthEdge Delivery Services?

HealthEdge Delivery Services is responsible for deploying, upgrading, and optimizing HealthEdge solutions for health plan customers. The team brings deep expertise in health plan operations, benefit configurations, testing, data migration, and integration design.

What are the four stages of implementation?

The four stages are Discovery, Configuration, Testing, and Deployment and Stabilization. AI-powered tools are embedded across all four to reduce manual effort, improve accuracy, and accelerate timelines.

How does AI reduce implementation risk for health plans?

AI reduces implementation risk by catching problems early and preventing them from compounding. During Discovery, AI flags requirement gaps and compliance issues before configuration begins. In Testing, it generates test cases and detects defect patterns before go-live. At Deployment, AI shifts teams from reactive troubleshooting to proactive issue management. Throughout, experienced implementation consultants validate every AI output – preserving quality while enabling speed.

Does AI replace the expertise of HealthEdge implementation consultants?

No. AI amplifies implementation consultants’ expertise by handling high-volume, repetitive tasks—such as documentation generation, test scenario creation, and requirements analysis—so implementation consultants can focus on complex, judgment-intensive work where their domain knowledge matters most.

How does AI-enhanced implementation affect time to value for health plans?

By compressing the time required for discovery, requirements gathering, documentation, testing, and validation tasks, AI-enhanced delivery significantly shortens implementation timelines. That means health plans can begin realizing the operational and financial benefits of their HealthEdge solution sooner.

Which HealthEdge solutions does Delivery Services support?

HealthEdge Delivery Services supports implementation across the HealthEdge product portfolio, including HealthRules® Payer, HealthEdge GuidingCare®, and other integrated platform components. Customers should connect with their Customer Success Executive to discuss scope and available services.

This is part 3 of a blog series on Ethical AI. For context on this series and why we’re writing it, see our introduction in part 1, Ethical AI: Privacy and Security.

How Can Payers Address Bias in AI Systems?

The previous article in this series covered what AI bias is, how it surfaces in outputs, and how it can enter AI systems. The key points: fairness is a key component of ethical AI, and its effects can be subtle and difficult to detect. Bias exists along a variety of demographic, medical, and socioeconomic lines, and can be unwittingly introduced even with good intentions.

This post addresses the practical question that follows: how can an organization address bias in AI systems?

The answer varies by role. Every person at HealthEdge® engages with AI in some capacity—as a user, as a tool selector, or as a builder—and the interventions available differ accordingly.

Guidelines for AI Users

The most immediate leverage most people have over AI bias is how they interact with AI systems day-to-day—and that leverage is more significant than it might appear.

Before writing a prompt, consider the framing. AI output reflects the inputs provided, and embedded assumptions shape the result. For example, consider querying an AI for a healthcare use case. Would this input be described the same way if the person had a different name, different insurance coverage, or a different background? For example, perceptions of a patient being “drug-seeking” versus exhibiting “undertreated pain” represent the same clinical presentation framed differently, and the AI model will respond to each in meaningfully different ways. In addition, stay aware of the degree and quality of context you provide—if that differs across cases, the model’s output quality may differ as well.

Be aware of common bias patterns when reviewing AI output. For example, recommendations that vary by demographic attribute, summaries that are shorter or less detailed for certain groups, tone differences across groups, or assumptions that fill ambiguous information with stereotypes.

When an output seems problematic, reporting the observation and adapting in the short term are both important steps. In the interim, adjusting prompts to counteract observed patterns is a practical response.

Guidelines for AI Buyers

Beyond individual interactions, many people at HealthEdge influence which AI tools and vendors the organization adopts.

For any decision involving the selection of AI tools—for enterprise software, vendor collaborations, or personal day-to-day use—bias considerations should be explicitly included in the assessment.

Evaluation should include:

• What bias metrics does the system use, and what justifies that choice for this use case?
• What training populations does the model reflect?
• Are disaggregated performance metrics available?
• Are there published model cards or transparency reports that acknowledge known limitations?
• What monitoring occurs after deployment?

Responses like these should serve as warning signs:

• Dismissing bias concerns with “we don’t use race or gender” and ignoring proxy variables
• Claiming a tool is “objective” or “unbiased” without supporting evidence
• An inability to name a specific bias metric
• No acknowledgment of model limitations
• Testing that is limited to pre-deployment with no ongoing monitoring

Guidelines for AI Creators

For those involved in building AI features—whether as designers, engineers, product managers, or testers—the responsibility to address bias is essential. Building fair AI systems should be part of each step of the software development lifecycle.

1. During design: The definition of “fair” should be established before development begins. Subject matter experts (SMEs) with a deep understanding of the use case should be included early in the process to identify the areas where bias is most likely or would cause the most harm.
2. During development: Teams should audit source data for representation gaps and prompts for embedded assumptions. For example, what does an instruction to “be concise” or “extract the most relevant information” implicitly prioritize? Try testing alternative phrasings and reviewing few-shot examples for demographic diversity.
3. During evaluation: Performance should be measured on subgroups rather than relying solely on aggregate metrics. Counterfactual testing should be built into the evaluation pipeline by systematically varying one demographic dimension at a time and measuring output differences. Edge cases and ambiguous inputs, where bias is likely to surface, should be included in the test set. Qualitative review is also necessary, as differences in tone, framing, and agency attribution may not appear in automated accuracy metrics and could require direct human comparison of outputs.

After deployment, monitoring for drift is important, since patterns can emerge or intensify over time. Known disparities should be documented even when they cannot be fully resolved, as transparency about the limitations of AI tools enables informed use and creates a foundation for future improvement.

Building Trust Through Shared Accountability

The steps outlined above span different roles and different stages of the AI lifecycle, but they share a common thread. The central theme is that fairness in AI is not a task to be delegated or a box to be checked at the end of a project. Instead, it requires deliberate attention across every stage of the process—from everyday use to vendor selection, and prompt construction through deployment and post-release monitoring.

The organizations that sustain that attention will not simply avoid causing harm but build systems that healthcare organizations and the members they serve can genuinely trust.

Introduction

This is part 2 of a series on Ethical AI.

The content was adapted from an internal learning and development session developed by the Machine Learning Engineers at HealthEdge®, focused on educating our organization on ethical use of artificial intelligence (AI). At HealthEdge, our belief in safe and responsible AI use shapes how we use these tools internally and how we build AI-powered solutions for our customers.

Addressing Bias and Fairness in AI

Bias and fairness are foundational components of ethical AI, and among the most difficult to address in practice. Most people using an AI agent wouldn’t know that factors like someone’s name, insurance provider, or demographic information can drastically alter a response—even with an otherwise identical prompt.

Bias in AI systems is rarely visible in a single output, and without deliberate measurement it often goes undetected.

Understanding AI Bias

When it comes to AI, bias refers to when AI systems produce outputs that unfairly favor or disadvantage certain groups.

Bias is not a hypothetical scenario. Production AI systems have demonstrated measurable bias across a range of industries and use cases, often without intention on the part of the developers. In each case, these systems appeared to function as expected until the outputs were measured across demographic groups.

Dimensions of Demographic Risk

Bias does not affect all people equally, and understanding its scope requires looking at the range of characteristics that can impact the output from an AI tool. Categories of bias can include:

• Legally protected categories (e.g. age, disability, religion, nationality)
• Health-related attributes (e.g. mental health diagnoses or chronic conditions)
• Socioeconomic factors (e.g. literacy, immigration status, living conditions)

Many people belong to more than one of these categories, and bias can compound across each. For AI tools, this means a single output can carry the weight of multiple intersecting inequities.

4 Ways to Recognize Bias in AI Outputs

Knowing that bias exists is not enough—addressing it requires understanding the specific forms it takes in real system outputs.

Bias rarely takes the form of a clearly wrong answer. In practice, it tends to surface in more subtle ways. These are four common manifestations of bias in AI outputs:

1. Quality differences

Appears as inconsistent accuracy or reliability across demographic groups. For example:

• Higher hallucination rates or lower accuracy for certain groups of users
• Greater uncertainty in answers about member groups where training data was underrepresented

2. Tone and framing

Appears as different verbiage, tone, or framing across groups.  For example:

• Characterizing identical behavior as “assertive” for one group and “aggressive” for another
• Cold language for one group and warmer, familiar language for another
• Less detailed output for or about certain demographic groups

3. Stereotype-driven gap-filling

Appears as AI filling information gaps with learned assumptions. For example:

• Different assumptions about a member’s likely needs based on demographics
• Assumptions about a provider’s specialty based on demographics

4. Outcome differences

Appears as different recommendations, actions, or end outcomes for different groups. For example, in agentic systems, different thresholds for autonomous action versus requesting human confirmation.

Distinguishing Bias from Appropriate Differential Treatment

The relationship between differential treatment and bias is not always straightforward and conflating the two can lead to over-correction and under-detection.

Not every instance of differential treatment represents bias. For example, women experiencing heart attacks frequently present atypically, and an AI that adjusts its diagnostic approach accordingly is addressing a historical gap in clinical practice, not perpetuating existing disparities. A conversational AI that simplifies its language for a user who identifies English as their second language is tailoring its response to serve that user better.

The central question is whether differential treatment reinforces historical patterns of disadvantage or not.

Sources of Bias

Bias is not introduced at a single point in an AI system’s development. Rather, it can accumulate across every stage, compounding the impact. Some of these steps may include:

• Training data reflects historical human behavior, including historical discrimination.
• Source data is shaped by those who produced it. For example, an AI system handling patient records relies on providers’ recorded interpretation of those patients. If provider notes document pain differently by patient race, or apply more skepticism to patients with mental health conditions, those biases get passed to the AI system.
• Developer prompts can un-intentionally embed assumptions. For example, a system prompt to “synthesize medical information” may lead to lower emphasis on mental health conditions if the model has absorbed the historical conflation of “medical” with physical health.
• User phrasing can cause framing effects in their inputs. For example, “why is this patient non-compliant?” could produce a materially different result than “what barriers might be affecting this patient’s care?”
• Evaluation data may overrepresent certain populations, causing the model to be optimized primarily for those groups.

The Limits of Demographic Omission

A common instinct when trying to reduce bias is to remove demographic fields from model inputs. However, this approach misunderstands how AI systems operate in practice. AI systems infer group membership indirectly through proxies. For example, names can signal gender and ethnicity, zip codes correlate with race and income, and historical cost of care can encode prior access disparities. These proxies can influence AI system behavior in the same way raw demographic fields can.

The claim that a system is unbiased because it does not explicitly use race or gender overlooks how demographic information enters the model through these correlated features.

Continuing the Series

This post has explored the nature of AI bias, the forms it takes in practice, and the layers of the AI pipeline through which it is introduced. The following post in this series turns to the practical question of what can be done to detect, measure, and mitigate it.

Key Takeaways

• Post-hybrid review is a critical audit readiness phase: Success depends on strong documentation workflows, accurate validation, and timely support for auditor requests.
• Centralized HEDIS coordination improves efficiency: Clear ownership, organized tracking, and streamlined auditor communication help reduce bottlenecks during validation.
• Documentation quality drives compliance confidence: Complete, legible, and defensible records are essential to support NCQA-aligned audit requirements.
• Audit feedback should fuel performance improvement: Submission issues and internal uncertainty can uncover training gaps, workflow breakdowns, and other opportunities to strengthen future HEDIS performance.

As the hybrid review phase concludes, another critical stage of the HEDIS® season begins: supporting the audit process with complete, traceable, and compliant documentation workflows.

At this point in the cycle, operational precision becomes just as important as abstraction accuracy. Health plans are no longer focused on identifying compliant members for the reportable HEDIS cycle. Instead, the priority shifts toward validating numerator counts, supporting auditor sampling requests, delivering accurate medical records quickly, and maintaining confidence in every submitted record.

Health plans that navigate this phase successfully typically have disciplined processes already in place, such as:

• Clear documentation workflows
• Centralized tracking
• Strong communication between abstraction, quality, and audit teams

Streamlining these areas can significantly reduce delays, minimize discrepancies, and improve overall Medical Record Review Validation (MRRV) audit readiness.

This phase of the HEDIS cycle also creates an opportunity for continuous improvement. Records that create uncertainty during submission, whether due to incomplete documentation or inconsistent abstraction interpretation, can help identify operational gaps and inform future training strategies for quality and abstraction teams.

1. Producing Compliant Hit Lists by Measure

Once abstraction activities are finalized, health plans must generate comprehensive hybrid input files for each hybrid HEDIS measure—along with all applicable exclusion counts signaling to the auditor that medical record review is completed for the measurement year. Accuracy at this stage is essential, as these lists serve as the foundation for auditor sampling and validation activities.

Plans should reconcile hit lists against the final current abstraction outputs and reported numerator data before submission. Even minor discrepancies between abstraction systems, reporting files, or audit documentation can create unnecessary delays during validation.

Strong organizations also prioritize traceability and audit-ready workflows throughout this process. Compliance criteria, abstraction methodologies, supporting documentation standards, and submission logic should be clearly documented so teams can quickly explain how numerator compliance or exclusions were determined if questions arise during an audit review.

Operational maturity becomes especially important when deadlines approach. Final numerator-compliant counts and exclusion counts must be submitted within established timelines, with little room for delays or rework.

Many plans are increasingly leveraging automation and reconciliation workflows to strengthen accuracy during this stage. According to industry reporting, organizations using automated abstraction reconciliation processes have reduced hit list discrepancies by up to 25%, improving downstream audit efficiency and reducing manual validation work.

As outlined in NCQA’s HEDIS Measures and Technical Resources, maintaining consistent reporting methodologies and traceable compliance documentation is essential to supporting audit transparency.

2. Supporting Auditor Sample Selection

After the hybrid input file is finalized, plans move into the auditor sampling phase, and the Random Selection File is used to submit full member lists and for returned sample. The Random Selection File contains member level data on numerator events for all members in the product line. During this process, auditors select records from measure groups and exclusion categories for MRRV review, typically requesting a defined sample set for validation.

Preparation and organization directly affect how efficiently this phase progresses.

How should health plans structure member lists?

Health plans should structure member lists according to the specifications provided by their auditor. Teams should maintain comprehensive change logs and clear audit trails for every submission to ensure transparency throughout the review process.

Many organizations also benefit from establishing a centralized audit coordination process. Assigning a dedicated point of contact to manage auditor communication can help streamline clarification requests, reduce duplicate outreach, and accelerate turnaround times during active audit periods.

Recent operational case studies have shown that proactive auditor engagement and centralized coordination reduced clarification and selection timelines by nearly 30% during HEDIS validation activities.

Many of these workflows align closely with established NCQA HEDIS audit methodologies and commonly used “8 and 30” file sampling procedures designed to support audit validation and traceability.

How should health plans respond to organizational feedback?

Equally important to member list structure is how organizations respond to auditor feedback. Questions, corrections, or flagged inconsistencies should not simply be resolved in isolation. They should feed back into operational process improvement efforts, helping teams refine documentation standards, abstraction workflows, and training programs for future measurement years.

3. Pulling and Delivering Requested Documentation

Once auditors identify selected samples, speed and accuracy become operational priorities.

Teams must pull medical records from their internal systems and supporting documentation quickly while validating that each submission is complete, legible, and compliant with NCQA requirements before delivery. Incomplete or inconsistent documentation often creates additional follow-up requests, extending timelines and increasing administrative burden during already compressed audit windows.

Many organizations now rely on centralized tracking dashboards and real-time operational visibility to monitor fulfillment progress across retrieval teams, identify aging requests earlier, and maintain tighter coordination across abstraction, retrieval, and audit support functions. These workflows help reduce bottlenecks, identify aging requests earlier, and improve visibility into outstanding documentation needs.

This phase is where operational coordination matters most. Retrieval, quality, abstraction, and audit support teams all need aligned visibility into document status, escalation paths, and submission deadlines to maintain audit readiness throughout the process.

NCQA also continues to evolve guidance around audit sample frame and submission requirements, including recent Measurement Year 2025 HEDIS CAHPS sample frame updates that may affect documentation and validation workflows for participating organizations.

4. Addressing Common Documentation and Workflow Challenges

Even highly organized teams encounter documentation challenges during HEDIS audit activities.

Unavailable records, conflicting documentation, incomplete chart data, and ambiguous abstraction findings can all compromise the accuracy and reliability of final validation.  The most common issues include misinterpretation of NCQA specifications, acceptance of an invalid record type such as a continuity of care document (CCD) for MRRV, and inconsistent interpretation of specification requirements during abstraction. The key difference between reactive and high-performing organizations is how quickly those issues are identified, escalated, and resolved.

Standardized internal quality assurance (QA) reviews also help ensure that submitted batches meet documentation expectations before they reach auditors.

Just as important, organizations should pay close attention to records that create hesitation internally. If abstraction or quality teams lack confidence in the completeness or defensibility of a submitted record, those cases should be flagged and reviewed after the audit cycle concludes.

Those scenarios often reveal larger process opportunities, such as:

• Inconsistent abstraction interpretation
• Retrieval delays
• Unclear provider documentation
• Training gaps
• Workflow breakdowns between teams

Using audit feedback as an operational learning mechanism can significantly improve future-cycle preparedness and strengthen overall documentation quality across the organization.

This is also where integrated quality operations can help reduce friction. HealthEdge® Risk Adjustment and Quality solutions are designed to help plans strengthen audit-ready workflows, improve documentation accuracy, increase operational visibility, and create more connected quality and abstraction processes across HEDIS operations.

5. Strengthening Audit Readiness Beyond the Review Cycle

Coordinated, well-documented post-hybrid review processes are essential to maintaining HEDIS compliance and audit integrity.

Health plans that invest in structured hit list production, organized auditor engagement, centralized documentation tracking, and proactive QA workflows position themselves for stronger submission accuracy and more efficient audit cycles. They also create opportunities for continuous operational improvement by identifying gaps, refining workflows, and strengthening training programs year over year.

Increasingly, leading organizations are moving toward more integrated quality operations models that connect abstraction, audit support, retrieval management, analytics, and quality oversight into a more unified workflow.

As HEDIS requirements continue to evolve, audit readiness increasingly depends on the ability to combine operational discipline with scalable quality processes. By improving how compliant records are organized, validated, retrieved, and monitored, plans can reduce audit risk while supporting more accurate quality measurement outcomes across the enterprise.

Achieve Quality Excellence with HealthEdge

Discover how a regional health plan went from deficient analytics and reporting inefficiencies into a 4+ Star rating and more than 90% HEDIS Data Retrieval. Download the case study: From Underperformance to a 4+ Star Rating – How HealthEdge Propelled a Health Plan to Quality Excellence and Revenue Growth.