Autonomous inpatient coding. This may sound like a pipe dream for revenue cycle leaders, whose hopes were likely raised and crushed in the early aughts. But it’s coming sooner than most think.
Why is true inpatient coding automation the “holy grail” – as Kerry Gillespie, a former CFO at Intermountain Health and now an executive consultant at Warbird Consulting Partners, said to me at a Healthcare Financial Management Association roundtable last year?
For decades, inpatient services have been the backbone of many provider organizations and, simultaneously, a significant pain point for those managing hospital operations. First, the financial imperative is clear: although inpatient represents a much smaller share of patient volumes, it typically drives 60% of total revenues, according to MD Clarity.
Despite the financial magnitude, managing the revenue cycle for inpatient visits has proven more challenging. While providers have struggled with staffing shortages for medical coders of all stripes, certified coding capacity for inpatient admissions has been especially limited. According to recent Glassdoor data, this shortage is reflected in labor costs, with inpatient coder wages often 20%+ higher than outpatient wages. An MGMA Stat poll found that 34% of medical group leaders cited medical coders as the most difficult revenue cycle position to hire for.
With so much value on the line, coding accuracy and speed are much higher stakes. Inpatient coders must grapple with some of the most complex medical cases, typically involving longer stays and more complicated care plans – meaning a much larger volume of medical documentation to review and code. Fully automating coding for even a portion of inpatient cases would thus significantly reduce the burden on coders. Indeed, solving inpatient coding with AI, an elusive goal, would be a massive win for providers – perhaps one of the most impactful operational opportunities in a generation.
So what’s changed?
Tech advancements offer renewed promise
If you’re skeptical of meaningful automation on the inpatient side, you’d be right to think that way. But recent developments have upended what’s possible. To make sense of that, let’s take a quick look at what the past few decades have tried to deliver.
The history of coding automation is defined by four key phases:
- The arrival of natural language processing (NLP)
- The transition to ICD-10
- The advent of deep learning
- The development of large language models (LLMs)
NLP is a type of AI that can be traced back to the 1940s. It enables computers to understand, interpret, and generate human language in a natural and meaningful way. In the early aughts, pioneers like CodeRyte and A-Life (now owned by other companies) led the way in coding productivity based on NLP, achieving rates of 70% automation for low-complexity encounters and 30% automation for moderate complexity. But these capabilities hit their limit – and then coding got harder.
On October 1, 2015, ICD-10 went live in the US. And with it, the number of possible diagnosis codes ballooned from 13,000 to 69,000. This explosion of codes crushed the promise of early NLP, as it couldn’t handle the larger volume and complexity of codes. As medical coding teams struggled to keep their heads above water and contend with 5x the number of codes, automation rates from available tools plummeted, negatively impacting the revenue cycle.
A light on the horizon appeared in 2018 with advances in deep learning. This approach combined mountains of data and extraordinary computing power to create AI that figures out its own rules, translating into super-high automation rates, significant cost reductions, and broad specialty coverage for health systems.
Underpinned by deep learning, autonomous coding has rapidly expanded across high-volume outpatient specialties, delivering automation rates of 85-90%+ or even pushing 99% in specialties like radiology. Prompted partly by broader AI agendas, autonomous coding is becoming the norm for many outpatient settings. This same technology is now turning to inpatient care.
But to make inpatient autonomous coding robust, the last boon has been the arrival of LLMs. This generative AI technology – known as the basis of ChatGPT and other popular tools – complements deep learning to handle last-mile issues, enabling near 100% automation rates. With deep learning and LLMs working in tandem, providers can at last realize true inpatient coding automation.
What to expect and how to prepare
From my vantage point, true autonomous coding for inpatient will arrive later in 2024, providing hyper-accurate automation for the majority of patient admissions. In 2025 and beyond, inpatient capabilities will catch up to where outpatient is today.
This advancement is not too far away. So, how can health system leaders ready their organizations to take on meaningful inpatient automation? A great place to start is by bringing autonomous coding to outpatient departments. Besides reaping the financial and operational benefits – including heightened accuracy, reduced labor costs, improved revenue capture, and decreased administrative burden – getting started on the outpatient side enables leaders to build a relationship with a trusted vendor and sets the stage for expansion to inpatient. Completing implementation for one or more outpatient specialties will also build confidence in the approach and enable revenue cycle and HIM teams to move more quickly on inpatient down the road.
In addition to paving the way for inpatient automation, securing experience with autonomous coding for outpatient helps to promote the organization’s broader AI ambitions. As providers pursue different models for building AI competency – hiring Chief AI Officers, forming cross-functional committees, or appointing in-house experts – locking in concrete projects such as outpatient automation helps to increase momentum and skill-building. Visibility into these AI efforts may even help systems to attract top talent – who, according to BCG, increasingly expect AI to help with their day-to-day roles – in tough labor markets.
Capturing the ‘holy grail’
The impending arrival of autonomous inpatient coding is a remarkable breakthrough for health systems. Recent advancements in deep learning and LLMs mean this transformative technology is closer than ever before. And starting now, the provider organizations that proactively set themselves up for this monumental shift will reap the most benefits.
About Andrew Lockhart
Andrew Lockhart is CEO of Fathom, the leader in autonomous medical coding. Andrew earned his MBA from Stanford University and his BA from the University of Toronto. He is an avid speaker and has presented at HFMA, Academy Forum, Stanford Medical School, and HBMA events.
