Sponsored

Using Artificial Intelligence to Boost Charge Capture

by: Jason Williams, MBA, MEM, Vice President - Financial Assurance, Analytics and AI Change Healthcare

A recent survey from the National Association of Healthcare Revenue Integrity (NAHRI) shows charge capture is the number one supporting function for healthcare revenue integrity programs.1 Yet, many providers continue to fall short in fully documenting and coding the clinical encounter, leading to missed revenue and potential compliance issues.

All told, missing charges and associated reimbursement, combined with audit and recovery efforts, can have a tremendous impact on your bottom line. Moreover, there are regulatory risks if coding compliance isn’t what it should be.

While organizations are beginning to use technology to improve and streamline the coding process, there is still a high potential for human error. Given the frenetic pace of healthcare delivery, it’s not surprising providers can inadvertently miss documentation opportunities or make mistakes. Coding efforts present another chance for error as coding staff often are processing a high volume of care episodes within constrained timeframes.

How Can Artificial Intelligence Be Used to Improve the Process? 

AI uses computers to do work that usually requires human intelligence, thinking, or cognition. When these machines are trained to handle certain tasks, the tasks can be done as well or better than if humans performed them. So, for example, we can “train” computers to predict missing charges and make  ecommendations, helping organizations improve the accuracy and comprehensiveness of their claims, which translates into substantial revenue benefits.

Given the repetitive nature of charge-capture processes, there are significant opportunities to apply advanced technologies like artificial intelligence (AI) to increase the function’s accuracy and reliability.

Making AI Work for You

At Change Healthcare, we focus on the fundamentals that make AI more effective. Lots of companies are creating AI models, but the models’ recall, precision, and performance are not all the same.

Effective AI requires three things: rich data from which machines can learn; informed experts who can design the ideal learning models; and optimal delivery systems that present meaningful information at the point of decision-making.

Rich Data
First, let’s look at the data piece. AI learns what you teach it, and it requires a lot of information to learn. For example, while a toddler may learn to recognize pets after being exposed to a few of them, AI may need to see thousands of images before it comprehends and performs as well as the toddler. Since Change Healthcare owns one of the richest data repositories in healthcare—housing information from most hospitals, physicians, payers, and other stakeholders— we’re able to train our AI solutions on a breadth and depth of data that will enable them to perform with precision and accuracy.  

Expert Modeling
Next, consider access to data scientists—the experts who know the best ways to train AI solutions from the real-life experiences the data represent. Change Healthcare has invested in an in-house team of data scientists who are skilled in machine learning, neural networks, reinforcement learning, and a handful of other technical concepts that help them choose the best methods for building high-performance
AI models.

Delivering Predictions Pragmatically
Finally, let’s explore delivery strategies. Predictions are only as good as their availability at the point of decision-making. Change Healthcare uses a design-thinking discipline to determine how to get the AI results to the user, so he or she can act as close to real time as possible. This is not dissimilar to the way you receive information via your smartphone at a time and/or location that you can act on it.

How Charge Capture with AI Works

Change Healthcare’s Charge Capture Advisor is a cloud-enabled solution that identifies potentially missing charges in patient accounts prior to claims submission. Working with providers’ existing health information systems, the technology flags missed charges alongside providers’ existing coding and claim workflows.

Charge prediction results are delivered in real time, at the point of decision-making, so the user can act quickly to resolve discrepancies and help ensure a more complete claim prior to submission.

Commonly predicted charges include missed or incorrect diagnostics, images or tests, or implants or supply codes. Overlooked or deficient documentation for drugs and injection administration can also be flagged. To identify these missed items, the AI has been “taught” to recognize when claim information doesn’t look like it normally would, based on the real-world experiences the AI has been exposed to in the large data set.

Comparing AI to Rules-Based Predictions

AI uses a statistical approach to make predictions as opposed to rules-based methods, which have been used to aid in charge capture for some time. Rules-based systems report potential missing charges based on flags triggered by certain conditions. Unfortunately, the rules or flags are often either too aggressive or too conservative, with aggressive rules flagging too many issues and conservative ones failing to detect critical missteps. Also, rules-based systems are both time-consuming and resource-intensive to maintain due to constant chargemaster and clinical process updates, which require someone to adjust the relevant rule.

Conversely, an AI solution automatically learns over time based on charge predictions that are accepted or rejected, and will start to recognize updated patterns as well as any outliers as new data emerges—limiting the need for human intervention.

For more information about how Change Healthcare is using AI to transform charge-capture processes, visit our website or call 866-817-3813.


1National Association of Healthcare Revenue Integrity, 2018 State of the Revenue Integrity Industry Survey Report

This article was created in collaboration with the sponsoring company and our sales and marketing team. The editorial team does not contribute.