Meet us at RBMA SYNC 25 in New Orleans | Oct 5-7 | Learn more
Meet us at RBMA SYNC 25 | Learn more
Join us for a broadcast with CHCS Consulting, featuring Brad Adams, VP Sales and Stacie Buck, Director of Coding Compliance as we explore Generative Artificial Intelligence in Medical Coding. Discover how this technology is not only reshaping the landscape of revenue cycle management but also enhancing the accuracy and efficiency of the medical coding practices. Don’t miss out on the opportunity to uncover transformative insights that AI is making in the healthcare industry.
It’s nearly impossible to turn on or read the news without hearing a story about artificial intelligence (AI). While it seems like it’s a completely new technology, in fact,we have been co-existing with bots for many years through the use of automated systems, interacting with chat bots when contacting customer service departments, and of course, with some of the most well known AI personalities, Siri & Alexa.
AI technology is reshaping the landscape of many industries, and it’s poised to have a significant impact on the healthcare industry. AI solutions are already being deployed throughout many areas of the revenue cycle – patient scheduling and registration, eligibility verification, charge capture, claims & denial management, payment collection, patient communication and of course medical billing and coding.
Over the past two decades, much of the medical coding has been performed using Computer Assisted Coding (CAC) technology. CAC is an early form of AI primarily utilizing natural language processing (NLP) to assign CPT, ICD-10-CM & HCPCS codes to encounters. While the use of CAC has continued to grow since its inception, autonomous coding solutions are now positioned to replace CAC technology.
In speaking with coding professionals across the country, I have noted that often autonomous coding is thought of as the next version of CAC, when in fact, the technology is very different from CAC in its design and functionality.
Let’s take a look at only a few key differences between CAC and autonomous coding.
1. CAC is a tool that relies heavily on human intervention to arrive at correct codes, while autonomous coding solutions can code many encounters in seconds with no human intervention
2. CAC requires reports with a predefined structure and cannot easily adapt to changes in language patterns. AI can process reports with unstructured data and can detect language patterns and consider context.
3. CAC is primarily driven by NLP, focusing on the identification of specific words and phrases to suggest codes, while relying heavily on rule-based coding. Autonomous coding solutions utilize machine learning, deep learning and algorithms to assign codes.
Machine learning is a process in which computers learn from data without being explicitly programmed. Deep learning takes this process a step further using multiple layers to recognize complex patterns in data. Deep learning models are taught using a large historical data set to train algorithms to ensure the coding engine produces the most relevant predictions, capturing the subtle variations and coding style of the provider. This process of using large amounts of historical data is what makes achieving a high direct-to-bill percentage possible.
4. In a typical CAC workflow, reports are sent to the coding engine, the engine reads the text and provides code suggestions. A coder then reviews the coding engine output and makes any necessary changes. If a repetitive error is noted, a rule may be created to correct the issue. This can be a time consuming process, resulting in hundreds and even thousands of rules being established over a course of time. With an autonomous coding solution, reports are sent to the coding engine, coded within seconds and sent directly to billing. Any encounters that cannot be coded by the engine, will be sent to a human coder for review. As coders modify an assigned code or add or delete codes, the model is trained and re-trained in real-time each time a coder makes these changes.
Conclusion
While traditional CAC workflows still require considerable manual intervention and rule-setting, autonomous coding solutions stand out with their real-time learning capabilities. Not only do they streamline the coding process, but they also continuously refine their accuracy with each human intervention. The convergence of technology and human expertise in this domain signifies a promising step towards more accurate and efficient coding in the future.
Maverick Medical AI is supported by an entire team of industry veterans of medical, regulatory, and AI/ML experts that not only had the vision to revolutionize medical coding, but are further revolutionizing the implementation phase with well defined, frictionless stages. VP of Product, Matan Neeman, shared some details about implementation when you’re making the transition to autonomous medical coding with Maverick.
I joined Maverick in early 2022 with a mission: To use machine learning and Generative AI technologies to automate the medical coding world. I spent my career in digital health, with a focus on information systems that support healthcare practitioners – technicians, physicians, nurses and administrators. I was part of a group who developed Cardiology PACS and Information Systems, then moved to the Radiology and Advanced Visualization space, and supported implementations of Enterprise software systems across multiple healthcare networks and systems.
As the leader of the product team at Maverick, I’m responsible for an important piece of our offerings: Product Management, skilled project management, implementation, training, and quality monitoring. Many vendors in this space leverage implementation specialists, and perhaps billing specialists, but Maverick leverages an entire team of product, project, implementation, revenue cycle experts, and beyond – all with in-depth medical and coding expertise. Folks who’ve been in the trenches and have an immensely in-depth understanding of the clinical environment.
The critical understanding and background our team has, paired with the way we collaboratively work with clients to implement autonomous medical coding to their practices and systems, differentiate Maverick.
Our team devised a meticulous process to minimize client implementation hours while validating historical data quality. Once we comprehend the implementation’s nature, such as shifting from less automated to more automated systems or from no automation to autonomous coding, we create a customized project plan. This planning stage is integral, offering change management assistance, identifying impacts on team members, stakeholders, systems, and processes, all for a seamless transition.
The key part of our implementation strategy that makes go-live frictionless? It’s a simple philosophy, but it’s critical: Preparing for it in advance. We engage organizations with a consultative and collaborative approach, leveraging their expertise, and yielding a plan that’s specifically for them and yields a much better outcome.
Collaboration with customers during the setup stage is critical to achieving the 85% direct-to-bill rate. We both have responsibilities that ensure the success of the implementation. One of these critical pieces to work and collect stakeholder input and analyze historical data to train the Maverick autonomous medical engine. Many vendors start at zero and add rules and automation after. Additionally, practices who’ve extensively customized their existing products may face challenges transferring this knowledge to a new system due to the lengthy accumulation of customizations by various individuals, many of whom are no longer with the organization. Unlike other vendors, Maverick eliminates the need for customers to manually provide these rules and customizations. Instead, Maverick automatically extracts this information from historical data, enabling a smooth and efficient transition. It’s beneficial and a lot less burdensome to our clients, especially in a replacement market where old tech has already become a burden.
Customers can expect the same degree of planning and careful monitoring to happen after go-live as they experienced beforehand. Our depth of knowledge in coding and not just implementation is key. Our folks have a greater understanding of the clinical environment and are adept at problem solving in a way that won’t frustrate clients who rely on folks who are only familiar with billing and implementation. Our decades of coding experience sets us apart. We relate to everyone, from SME to the clinical user to management and our resources pull from every department imaginable – all to serve our customers better without the unnecessary burden.
I encourage anyone who is interested in autonomous coding for their practice or health system to request a demo and learn about the advantages of a frictionless implementation of Maverick’s autonomous coding solution and how you can experience an 85%+ direct-to-bill rate.
Embark on a fascinating exploration at the cutting-edge intersection of Artificial Intelligence and medical coding, guided by Contempo Coding – Victoria Moll. Joining her are our visionary CEO and Founder, Yossi Shahak and our nationally recognized subject matter expert (SME) in medical coding, Stacie Buck. Together, they delve into how AI is catalyzing a revolution in medical coding methodologies, streamlining reimbursement and enhancing accuracy in this captivating video.
See how Maverick’s Autonomous Medical Coding solution can impact your practice.
In the face of medical coder shortages and burnout, Maverick Medical AI is dedicated to revolutionizing the healthcare industry. Research highlights a projected 7% annual growth in demand for medical coders, while a current 30% workforce shortage exacerbates the challenges faced. These shortages have far-reaching negative consequences, including increased claim denials, revenue delays, and inaccurate coding practices.
To tackle these administrative burdens, the implementation of AI-driven solutions presents an exciting opportunity. While many Healthcare Executives recognize the potential of AI, it is crucial to acknowledge that not all autonomous medical coding solutions are created equal in terms of accuracy and performance.
In the early days of medical coding, the process heavily relied on manual intervention and expertise. As technology advanced, rule-based systems emerged as a solution to automate certain aspects of coding. These systems utilized predefined rules and algorithms, aiming to streamline the coding process. However, they had limitations, including the need for extensive manual rule development and a lack of adaptability to handle complex and evolving coding scenarios.
The introduction of machine learning algorithms brought a significant shift in medical coding technology. These algorithms leveraged data to learn patterns and relationships, allowing for more automated and accurate coding. Nevertheless, human experts were still required to manually extract relevant data for these machine learning models.
At Maverick Medical AI, our cutting-edge technology harnesses the power of clinical language models, deep learning techniques such as Generative AI, and Generative Pre-trained Transformers (GPT). These advancements enable accurate and efficient coding of unstructured text, exceeding a remarkable 85% direct-to-bill rate and an exceptional accuracy rate of 97% upon implementation. Our commitment to pushing the boundaries of innovation sets us apart in the market.
How do we accomplish this?
Let’s shed a light on our remarkable technology stack.
By employing an ensemble architecture, we merge numerous models, specifically deep learning AI networks proficient in discerning patterns and excelling at various tasks, along with other algorithms. This approach harnesses the strengths of diverse models with the objective of enhancing the overall performance and adaptability of the system. Medical Coding is ideal for this architecture, where it can truly excel in accomplishing these tasks.
Developing a medical coding language model required a significant amount of effort to tune and adapt existing toolkits, incorporating our medical coding expertise and specialized models specific to the task. This has allowed us to customize, adjust, and enhance the technology according to our unique coding expertise.
Other language models cannot do this. ChatGPT, for instance, does not inherently solve the problem at hand. While ChatGPT is a powerful language model, it is not a medical coding platform. It has limited ability to generate accurate medical codes as it may not have the most current coding information nor does it understand the nuances of the practices, payors, and regulations. What point would there be to using ChatGPT in its current form for the purposes of medical coding, when the goal is to increase accuracy and reduce reliance on human coders in short supply and such high demand? The degree of auditing necessary would actually become an increased burden to coders.
Maverick’s autonomous medical coding AI engine is trained on a vast dataset of medical codes, enabling it to comprehend the intricate and dynamic rules and regulations governing our industry.
For effective machine learning, a substantial volume of consistent, accurate, and up-to-date data is required to train the system. Even the most exceptional providers with well-established workflows and practices may face challenges in maintaining data consistency and keeping it updated with the latest regulations over time.
The Maverick algorithm uses various sampling testing and techniques in machine learning to ensure that we are able to train the model with minimal human interaction and to iteratively improve it.
Understanding the provider’s historical data has been the success of Maverick’s achievement of 85% direct-to-bill rate at go-live.
As we all know, garbage in is garbage out. During implementation, large amounts of claims data, along with the reports, are analyzed and validated. Any coding quality issues are identified and corrected prior to go-live. And learning from historical data means no countless hours spent starting from scratch to rebuild coding rules.
Maverick employs advanced techniques to address scenarios with limited data. One such approach is the utilization of Generative AI models, which enable the generation of synthetic data resembling real-world examples. This synthetic data creation process allows Maverick to augment the available data and continuously enhance the system’s performance.
Additionally, Maverick utilizes transfer learning, a technique that leverages knowledge gained from pre-trained models on related tasks. By fine-tuning the pre-trained models using specific medical coding data, Maverick ensures consistent training and further refines the system’s accuracy in assigning appropriate codes.
By combining Generative AI for synthetic data generation and transfer learning for fine-tuning, Maverick optimizes the performance of its medical coding system, even in scenarios with limited data. This groundbreaking approach empowers Maverick to attain exceptional coding accuracy, enhances the overall efficiency of medical coding processes, and is the key factor behind our ability to achieve a guaranteed direct-to-bill rate of 85% upon go-live.
Maverick is continuously setting new benchmarks in accuracy, adaptability, and efficiency in the realm of medical coding. Our advanced solutions not only yield highly precise codes, surpassing 97% accuracy, but also fulfill the rigorous requirements of payors and regulatory agencies. With a guaranteed 85% direct-to-bill rate upon go-live, Maverick is redefining industry standards today and for the foreseeable future.
Many years ago, the company I worked for was an early adopter of Computer Assisted Coding (CAC) technology. At that time the vendor was claiming up to 75% of coded reports could be sent direct-to-bill and for a billing company, that claim was very promising. Unfortunately 3 years after implementation we were still reviewing 100% of the CAC coding output and never sent any claims direct-to-bill because we did not have confidence in the accuracy of the coding engine.
Fast forward two decades later, while there have been many improvements with CAC technology it still has not achieved high direct-to-bill rates once promised. It has limited capabilities using natural language processing (NLP) and relying on rules based coding. From my observations working with many providers across the country, CAC rarely exceeds a 50% direct-to-bill rate and many providers are not even hitting that threshold. Unfortunately, CAC has fallen short on desired outcomes for a large number of providers. Thankfully with the evolution of technology, artificial intelligence now has the ability to deliver better coding outcomes with a solution such as Maverick Medical AI’s autonomous coding solution.
When I initially began working with Maverick, they boasted of an 85% direct-to-bill rate at go-live. The skeptic in me said ‘No way, that threshold just isn’t possible.’ Much to my surprise Maverick has been able to deliver on that claim and I am excited to see how their technology will revolutionize the medical coding industry.
One of the main reasons why Maverick’s autonomous coding solution can deliver better outcomes than CAC is because it is powered by AI deep learning technology, rather than rules based coding. As a human coder interacts with the coding platform and makes coding changes, the platform is learning based on those changes and over the course of time the engine gets better and better at predicting codes. The success of Maverick in achieving an 85% direct-to-bill rate is largely due to applying the deep learning technology to a provider’s historical data.
Switching to a new coding solution may seem overwhelming and time-consuming, but the implementation process with Maverick is simplified. During the implementation phase, large amounts of claims data along with reports are processed and analyzed and then validated for coding accuracy. Maverick’s validation phase ensures that any coding quality issues are identified and corrected prior to go-live, reducing any future risks associated with coding quality. Much of this work is performed by Maverick’s coding experts, keeping the amount of time and resources allocated by clients during the implementation phase to a minimum. Additionally, there is no need to spend countless hours starting from scratch building your coding rules all over again since Maverick’s coding solution learns from your historical coding data.
As evidenced by the frenzy of new AI technology flooding the market, autonomous coding is the future of medical coding and CAC as we know it will become a thing of the past. With autonomous coding solutions becoming more prevalent in the market there are 3 key areas providers can address now to ensure the best outcomes with the autonomous coding solution of their choice:
1. Documentation improvement. Clear and concise documentation is the first key in achieving success with an autonomous coding solution. Coding is only as good as the documentation regardless of how coding is being performed. Utilizing standardized templates which contain all pertinent information to assign CPT, ICD-10-CM and quality codes are best.
2. Coding quality. Since a large component of the coding output is tied to historical coding practices, a provider should have confidence in its current coding quality. Providers should conduct regular coding audits (both internal and external) to validate coding integrity and correct any issues noted prior to implementing an automated solution.
3. Workflow. An efficient workflow is key to any successful implementation. It is essential to evaluate existing workflows and identify and resolve any issues that may be a barrier to implementation and proactively solve them before introducing any new coding solution.
Stacie L. Buck, RHIA, CCS-P, RCC, RCCIR, CIRCC
President & Senior Consultant, RadRx
