Overcoming Denied Claims: The Role of AI in Denials Scoring Models

Today’s hospital and health system revenue cycle management (RCM) leaders face a dual challenge: rising claims denials and labor shortages. According to a recent survey, approximately 15 percent of all claims submitted to private payers are initially denied, while a staggering 100 percent of revenue cycle leaders acknowledge the pervasive impact of the healthcare workforce shortage on their facilities’ ability to get claims paid. Adding to provider woes is the need to manage the monumental aftermath of the Change Healthcare cyber attack, which has caused some hospitals to lose roughly $2 billion a week.

To address these challenges head-on, RCM leaders should explore the use of AI to bolster their denials management strategies. One effective approach involves deploying AI to automate and streamline denials scoring models. AI-boosted denials scoring models offer a multifaceted approach to improving outcomes throughout the denials management process—from use of historical data and account scoring, prioritization, and routing, to predictive measures and workflow acceleration. Moreover, less reliance on manual processing assists in addressing staffing shortages by reallocating limited resources to higher value tasks.

What Exactly Is a Denials Scoring Model?

Denials scoring models utilize machine learning algorithms trained on historical data to predict the likelihood of appeal success. Key components include:

• Regression Models: Estimating a discrete number (probability score), ranging from basic to complex algorithms
• Historical Denials Data: Roughly 10,000 examples are needed for an accurate model, using input such as diagnostic codes, payer information, and prior appeals
• Likelihood of Appeal Success: Providing a probability score indicating the likelihood of payment success, with more complicated models estimating value and timing of payment

Because these models are trained on historical denials data, they take into account various factors such as CARC & RARC codes, diagnostic and procedural codes, payer information, account age, and balance. By estimating the probability of appeal success, scoring models enable hospitals to prioritize accounts effectively, optimize routing, and ultimately maximize revenue recovery.

How do scoring models fit into the AI landscape?

Underneath the umbrella of AI is machine learning (ML), and within that subset lie scoring models. Here is the relationship among these tools:

• AI is about creating smart devices and systems that can tackle problems creatively, mirroring human behavior and intelligence.
  • ML is a subset of AI. It enables computers to recognize patterns in data and make predictions without explicit programming, using algorithms like neural networks for problem-solving.
    • Scoring models, in the context of denials management, fall within ML and predict the likelihood of an appeal to succeed. This probability score can be used to calculate the estimated value and help prioritize accounts or optimize account routing.

“When we think about AI as a broad category, it’s really about how do we mimic human problem solving,” says Aspirion Chief AI Officer Spencer Allee. “Machine learning is also a broad category but it’s specifically about finding patterns in data, learning from those patterns, and then using those patterns to make better decisions.”

“And finally, a subset within machine learning is scoring models. In denials management, what we’re talking about is if you look at a denied account, what information can you use and how can you make predictions on the likelihood of that denial getting overturned? And what is the expected value of that account if we do overturn it? Or what’s the right place in your workflow to route that account to maximize yield on it?”

The Value and Complexity of Scoring Models

The sophistication of scoring-model solutions can vary, ranging from basic prioritization based on standard business rules to more advanced models that calculate an account’s expected value and optimize workflow steps. The more complex the sophistication level is, the more value is realized. Levels of scoring-model sophistication include:

1. Naïve Baseline: Accounts worked in the order received
2. Business Rules: Prioritize accounts by account balance and age with no ML
3. Probability of Success: Predict probability of success and prioritize high-probability accounts
4. Account Expected Value: Multiply probability of success by value of account to sort by expected value
5. Right Account to Right Desk: Factor in data on team performance by denial type, payer, etc., to optimize account routing within the team
6. Next Best Action: Re-score account after each touch and automatically route to optimal workflow step for resolution

Aspirion conducted a scoring-model pilot program with a subset of clients, which yielded a significant improvement in overall success rates. By implementing scoring models, Aspirion enhanced success rates by 3-5 percentage points with their initial model, showcasing the tangible benefits of integrating scoring models into denials management.

It’s important to remember that when implementing scoring models, success relies on having four key pillars in place: use cases, data, platform, and talent.

Measuring the success and performance improvements facilitated by scoring models is crucial. By adopting a data-driven mindset and implementing an AI measurement framework, healthcare providers can effectively track progress, pinpoint areas for improvement, and continuously refine processes to increase revenue recovery.

When it comes to effectively handling claim denials, AI stands out as a transformative force. It has the potential to boost RCM efficiencies and ease the financial burdens on healthcare providers. The use of AI by payers has already impacted denial rates, underscoring the imperative for healthcare leaders to swiftly integrate AI into their revenue cycle management strategies.

To dig deeper and learn more about using AI to overcome denials, join us for our next Level Up Revenue Cycle webinar, “Optimizing Efficiency by AI-Accelerated Denials Workflows.” Check out the session details here.