Medical Coding The Optimal Solution Is A Hybrid Approach With Deep Learning And AI Automation

The following is a guest article by Andrew Lockhart, Co-founder and CEO of Fathom.

The healthcare industry continues to evolve to serve patients, and with every advance, the system becomes more complex. This complexity is often evident in the medical codes for healthcare encounters. For example, ICD-10 has 69,000+ diagnostic codes, which is over four times the number of its predecessor, ICD-9. Naturally, more codes create a more complicated landscape for medical coding and billing departments to navigate.

Delayed time to payment, errors, and denials are major concerns for health systems. Considering that 68% of hospital denials are due to incorrect coding and denied claims cost hospitals roughly $262 billion per year, it’s crucial that your coding workforce is up-to-date.

Amid the constantly changing coding guidelines, training a coding workforce has become increasingly difficult for health systems. Staying informed on the plethora of payer-specific coding guidelines while simultaneously anticipating future revisions is no easy feat.

To address the growing intricacy of medical coding, many health systems are leaning into technology to automate a portion of their coding and billing processes. Because it has such a significant impact on the revenue cycle, it’s essential to identify what tasks are easily automated, which are not, and what blend of technology and human skill will yield the best results.

Often a hybrid approach that combines the power of automation for simple, routine tasks, and employees for more in-depth, complex jobs is the optimal solution.

How automation with deep learning and AI can transform the coding process

It is important to distinguish between simple, legacy tools such as computer assisted coding (CAC) and AI or true automation. CAC tools do not automate processes; they strictly impact workflow and productivity. Deep learning and AI combine large amounts of data with computer power and algorithms to mimic human intelligence and thus, automate processes—freeing staff to address other issues.

Routine tasks in the revenue cycle such as payment posting or insurance verification are easily automated using robotic process automation (RPA) or bots because there’s not a lot of variance in what comes through.

In the middle of the spectrum of easily automated tasks is coding. We’re finally at the point where computing power, AI, and data are sufficient to work at a level that exceeds what a human may be able to process.

For example, coding has made great strides regarding natural language processing (NLP) and recognizing what a physician says. Deep learning is excellent at picking up language around equivocation. So, if a clinician writes a note about a possibility, one shouldn’t necessarily code that.

Traditional coding software may have immediately flagged that as a code. In contrast, newer systems can recognize the subtleties in language and code or not code appropriately the same way a human might add a level of nuance when reading physician notes.

Leveraging a deep learning platform unlocks new levels of scalability within the coding process. This allows health systems to access a centralized best-in-breed machine that, once updated, applies its knowledge across their entire coding base, eliminating the strain of retraining coders individually.

It also addresses the fluctuating demand for coders. The U.S. Bureau of Labor Statistics projects that the overall employment of medical records and health information specialists will grow 9 percent from 2020 to 2030. However, the pandemic, employee burnout, and general seasonality have led to many health systems struggling to stay adequately staffed—a combination ripe for disaster if health systems don’t prepare. Luckily, deep learning and AI enable organizations to meet demand immediately.

In one instance, a provider organization had 20K of uncoded charts in their backlog. They leveraged a deep learning AI partner to code 90% of those encounters in hours without any human intervention.

A hybrid approach: The optimal solution

While the benefits are vast, deep learning and AI have limitations like any technology. In an ideal world, we could cover everything through automation, but because we can’t, humans are still needed to handle the more complex RCM issues.

For example, for any new use case, whether it’s a complex medical specialty or a corner case solution, machines require sufficient training data to teach themselves to code. Like any human, repetitions and experience translate pretty well into training data, which allows the machine to reach higher automation rates and accuracy levels.

So what tasks make sense to leave to your coding workforce?

Denials management is challenging to automate fully because of its complexity. This process often involves sifting through insurance information, codes, payer-specific guidelines, and any minutia that may have affected the denial.

Ensuring quick reimbursements is vital to a healthy revenue cycle, which makes efficient denials management essential. Due to its complicated nature, having experienced personnel handle this process yields the best results.

Auditing is another important task that can take a backseat when coders are bogged down by high volumes of uncoded charts. Auditing coded claims, either pre- or post-submission, allows management to keep an eye on higher-risk areas and gain insight into services rendered and how providers are documenting those services. Routine auditing also ensures that providers have access to regular feedback on their documentation practices. Coders can then provide targeted education so that providers are constantly learning and improving.

If you can utilize a solution that automates the large majority of your simple, routine coding encounters, you can use your coding workforce to focus on more complex tasks such as managing denials. This type of hybrid approach keeps your revenue cycle healthy in the face of an ever-changing coding landscape.

Make a hybrid approach work for you

To fully harness the power of deep learning and AI is to select the right solution, one that will help you reach your specific goals. Be sure to test all potential vendors using a large sample, one that truly represents your business. Understand how their services work in full production and test their solution by running an in-depth proof of concept that emulates your actual operations.

As mentioned, technology can be extremely beneficial, especially in the complex world of medical coding. Creating a hybrid approach that leverages the power of deep learning and AI to automate routine tasks allows your coding workforce to spend their time, energy, and expertise on more important tasks.

Use technology where it makes sense and let your workforce apply their expertise on jobs that move the needle. This hybrid approach will undoubtedly benefit your operations and positively impact your revenue cycle.

About Andrew Lockhart

Andrew Lockhart is a designer, entrepreneur and seed investor. He is the Co-founder and CEO of Fathom, a Tarsadia and Founders Fund backed company that uses deep learning to automate medical coding. At Fathom, Mr. Lockhart drives the corporate strategy, manages the commercial organization and leads hiring and recruiting.

   

Categories