Winter 2023

Coding Corner: Five Things Every Medical Coding Technology Should Deliver and How to Find the One That’s Right for You
By Amit Jayakar
For The Record
Vol. 35 No. 1 P. 28

Autonomous, artificial intelligence (AI)–driven medical coding technology solutions can revolutionize your organization’s revenue cycle. These solutions are catalysts for change; whether that change is positive or negative largely depends on the vendor and the type of technology it uses.

If you’re unfamiliar with solutions that automate medical coding, and even if you are familiar, finding a vendor with the right technology to help you reach your specific goals can be challenging.

There are steps you can take to make your search easier. For example, you’ll want to avoid these three common pitfalls when determining which solution to implement and ensure that the medical coding technology you chose delivers these five features.

Common Pitfalls to Avoid

Partial Solutions
Some coding solutions claim to do it all but don’t. When finding the right technology solution for your organization, it may be hard to discern which companies provide a full-autonomous coding solution and which are partial solutions in disguise. However, a good first step is recognizing that these partial solutions exist so you can stay vigilant during your search.

For example, some companies claim to automate coding through technology but actually rely on humans to code behind the scenes. This is not true automation; it’s essentially a tech-enabled offshore coding facility. If you are looking for a solution that genuinely automates portions of your medical coding through technology, look for one that does not take more than eight hours to return your coded encounters.

Computer-Assisted Coding Tools
Those who are not familiar with the nuances of AI coding automation technology may believe it’s the same as computer-assisted coding tools. However, the two are vastly different. Knowing the differences between these two technologies will help you avoid any disappointments regarding results after implementation.

Computer-assisted coding tools only affect workflows and productivity; they don’t automate processes. These simple tools can take years of rules customization to achieve the productivity rates they claim. Plus, rules-based technology tends to break very easily if any changes or updates occur and cannot scale up quickly in response to higher volumes of work.

True coding automation that uses deep learning AI doesn’t just work to boost productivity. It works in the same way as an actual human coder would. But, it works without any human involvement due to its ability to combine deep learning AI with large amounts of data to mimic human intelligence.

Unscalable Technology
The medical coding industry moves fast, and technology that can’t keep up isn’t going to provide the results you desire. Solutions that can’t quickly scale to meet demand will only hinder operations, not improve them.

For example, if you operate across multiple locations, implementing technology that takes more than a few months to support all sites is not time or cost effective. Or, if your organization experiences a sharp increase in patient volume, a solution needs to automatically meet that demand and act as an extension of your existing workforce. The inability to build a coding model quickly is indicative of a technology’s future inability to meet your changing requirements.

Five Expectations Where You Should Never Compromise
Seek a system that does the following:

1. Provides a Robust Proof of Concept
Before committing to a specific solution, ensure a vendor can deliver the results they promise through a robust proof of concept test. This will help you avoid accidentally implementing any of the unscalable, surface-level partial solutions mentioned above.

When going through this process, be sure to provide a vendor with uncoded encounters that are representative of your entire business—don’t cherry-pick a few small samples. It’s important to test the vendor on its ability to code large volumes of data across all chart types that they potentially would be responsible for so you can fully understand how its technology will work when it is in full production.

2. Implements Easily Across Coding Operations
Once you’ve found a vendor and tested its solution using your data, the next step is to dig into its implementation process. If it’s complicated, lengthy, and resource heavy, it will be hard to get buy-in from decision-makers within your company. With the right vendor, implementing an AI-driven solution to help your organization reach its coding goals can take as little as six weeks.

The first thing to look for is a process that does not rest squarely on the shoulders of your IT team and is primarily completed by the vendor. Let’s face it; your team already has a large workload, and asking them to take on a large task will set you up for failure.

Here’s an example of a standard implementation process when working with an AI-driven autonomous medical coding company that takes on a large portion of the implementation process.

Before it can start building your coding model, you’ll need to provide its team with a few pieces of information. Because this technology will act as another member of your coding team, treat this setup the same way you would with a newly hired employee. The vendor will need information detailing your workflow expectations. It is also important to outline any non-CMS custom guidelines specific to locations, payers, providers, or organizations and when to apply them.

Some information is needed from your internal IT department—mainly sample data, which it can provide in any format. A good AI-powered technology will use sample data in any format as fuel to build the coding model. So the more data the better, as they will give the deep learning technology a deeper understanding of your organization’s daily operations.

Once the AI technology builds the coding model, the next step is to ensure it works as it should. Your internal teams can use any EHR coding API, HL7, or SFTP file transfer to seamlessly transfer results into your system without any new manual workflows.

This entire process should not take more than three months, depending on the intricacy of your workflows and coding guidelines.

When searching for and choosing an autonomous coding solution, don’t overlook the importance of a fast setup that is mainly hands-off for your employees. If you find a company whose technology implementation takes longer than three months or requires you to take on most of the work, you may want to consider another vendor.

3. Offers Quick Scalability
Coding and payer guidelines are constantly evolving, which makes finding highly scalable AI technology of the utmost importance. Rigid, inflexible solutions that are primarily powered by manual rules or teams of analysts will not deliver the same results as their autonomous counterparts.

It is critical to recognize that while some vendors use an AI engine to build their coding models, the models are rules-based. As mentioned earlier, rules-based models tend to break easily and can’t deal with any variances in data formatting, physician language, or coding guidelines that sometimes occur. That’s why it’s essential to find a solution that leverages deep learning AI so it can understand varied, unstructured data and quickly make changes.

Combining AI with deep learning algorithms makes a solution extraordinarily flexible and scalable. Once an AI and deep learning powered coding model makes an update, it is automatically updated across its entire coding operations, even if they work across multiple sites. These seamless, automatic updates allow organizations to meet demand with high levels of accuracy.

4. Automates the Majority of Encounters
Automated medical coding solutions should enhance operational efficiencies where possible so that in-person staff can focus on more complex coding tasks such as handling denials or audits. Your coding staff’s expertise is best delivered when staff members do not have to review the majority of encounters.

The solution you partner with should automate more than 80% of encounters without human involvement, covering a broad range of medical coding specialties such as emergency medicine, risk adjustment, and radiology.

5. Guarantees Quality and Accuracy
Ensuring the quality and accuracy of your coding operations is vital to a healthy revenue cycle. This fact is abundantly clear when realizing that inaccurate medical coding costs medical organizations upwards of $20 billion per year, either in delayed or permanently lost reimbursement.

High accuracy and quality have the power to boost revenue, reduce time and cost inefficiencies, and create a better working environment for coding staff members.

That said, finding an autonomous coding solution that can guarantee a specified level of quality and accuracy is a surefire way to actuate real change throughout your department and, potentially, your entire organization.

Find the Solution That Works for Your Organization
Sifting through various vendors to find the right autonomous coding solution for your organization can be difficult. However, the search becomes less arduous and more straightforward when you know what to look for—and what to avoid.

An AI-driven medical coding technology that leverages deep learning is thoroughly tested against your own sample encounters, is easily implemented, offers quick and seamless scalability, automates the majority of your encounters, and can guarantee its quality and accuracy will ultimately yield the best results.

— Amit Jayakar is vice president of commercial operations for Fathom (www.fathomhealth.com), a deep learning natural language processing system that accelerates medical reimbursement and is backed by world-class investors, including Google Ventures, 8VC, and Stanford. Jayakar, who can be reached via Linkedin (www.linkedin.com/in/amitjayakar-00985911), earned his bachelor’s degree in economics from Georgetown University and his Master of Business Administration from The Wharton School.