What should medical coders know about artificial intelligence?

By David Fletcher
Vice President – Innovations, Streamline Health

Everywhere you look, our lives are becoming more automated. Machines can help us accomplish our most menial tasks, freeing us to devote more time to important, intense work that is profitable, engaging and beneficial.

Artificial intelligence (AI) and machine learning (ML) already assist us every day as we search online, stream music, read news and much more. In the healthcare industry, organizations have used AI/ML to anticipate patient health outcomes, better allocate resources, and efficiently schedule personnel.

As industry regulations push organizations to provide better care and keep more accurate clinical documentation, workloads will increase – unless organizations delegate menial tasks to automated systems and triage more complex work to the right people.

The U.S. Bureau of Labor Statistics reported that job growth for medical coders is growing at more than double the average rate across all industries. Between 2014 and 2024, the number of medical coding jobs is expected to grow by 15 percent. This influx of new workers is driven by the major update to ICD-10, which brought the total number of billable codes up to more than 70,000. Additionally, the broader adoption of APR-DRG over MS-DRG makes the coder’s job more complex. I recently heard from one of our clients that experienced coders who were handling 20 to 25 accounts per day with MS-DRG and ICD-9 coding are now finishing 10 to 12 with APR-DRG and ICD-10.

New coders, new codes and a constant need to process as many charts as possible is a recipe for missed revenue and costly compliance mistakes. Plus, the compounding issues presented by the Medicare Access and CHIP Reauthorization Act (MACRA) mean organizations could see negative payment adjustments for poor documentation.

It’s a lot to handle all at once, and many organizations will struggle to manage their finances if they continue to rely solely on human skill. Ultimately, medical organizations need an AI/ML coding solution if they hope to stay afloat in these turbulent times.

Medical coding is about to change rapidly

Today, some medical coders rely on computer-assisted-coding (CAC) to initially find a working set of codes. Instead of reading the whole medical record to learn the story of the patient’s encounter, the computer is proposing a “Cliff Notes” version of the story. Mistakes still occur. In fact, CAC fails to address a fundamental barrier to coding accuracy: a time constraint compounded by a mandate to process as many charts as possible, as fast as possible.

Audits are another key issue. Traditionally, they happen late in the revenue cycle which limits their usefulness. If an audit finds missed revenue several months after the bill has been paid, it may provide insights into coder accuracy but at that point it’s more costly to realize the additional revenue. Audits, and the accuracy that comes from them, must happen more frequently and quickly if organizations are to realize an efficient and stable revenue stream.

Working harder or faster isn’t the answer – and it takes years of experience to become an expert coder. An AI/ML solution is the perfect way to solve each of these problems in a single stroke.

Artificial intelligence and machine learning will augment medical coders

As a medical coding solution, artificial intelligence isn’t meant to replace coders, but rather augment their ability to code accurately and efficiently. Experienced coders shouldn’t have to spend hours each day coding simple charts, when they could better focus their efforts on complex tasks that no machine could complete.

“Real-time feedback helps coders improve faster.”

Consider this scenario: A new coder makes a small mistake as he quickly codes a chart. In real time, his AI assistant flags the mistake, recommends a solution, and informs the coder of the monetary difference the change will make. Alternatively, the AI assistant notification may go to a quality control reviewer but the point is to learn about the accuracy problem the same day, while the case is fresh and certainly before it goes to billing.

Too often accuracy is measured retrospectively through post-bill audit findings. While this follows the Check and Act steps of Deming’s quality improvement cycle of Plan-Do-Check-Act, it’s too late and it costs more to realize the additional revenue (or remove the risk of overbilling).

On a larger scale, and ML/AI solution could pinpoint common mistakes across an organization, tightening the floodgates against coding errors and improving documentation. This way, coders receive real-time feedback, so their skills improve faster. Likewise, top-tier coders can focus their efforts on complicated cases rather than on menial tasks.

Use our guide to learn how to optimize coding and financial performance prior to billing.

How Streamline Health® eValuator™ optimizes a coder’s workday

Streamline Health® eValuator™ is an intuitive AI/ML medical coding solution from Streamline Health which not only strengthens an organization’s revenue cycle, but also enhances the work day of each coder.

As shown in the previous scenario, eValuator™ can improve a coder’s skill by providing real-time feedback and support. This is all but impossible in a traditional audit process because by the time audits are finished, coders won’t remember the charts. (Can you remember the exact phrasing of an email you sent two months ago? Probably not.)

Best of all, eValuator™ won’t slow down a coder’s normal pace of work. Managers can decide what happens to flagged charts, so an alert doesn’t necessarily mean more work for the coder. For instance, if eValuator™ determines that the likelihood of a chart’s inaccuracy is 75 percent or greater, the system will automatically reroute the chart to a reviewer or auditor. If the chance of error is only 15 percent, the system will send it back to the coder – and if the chance of error is 5 percent, it will go ahead to billing.

Managers can adjust each of these values, and eValuator™ can take into account more than one variable. For instance, if a chart the chance of error is 75 percent, yet the potential financial gain is only $20, it’s not worth the investment of a second opinion, and therefore the system will send the chart along to the billing department.

This is only a small facet of what eValuator™ can do.