Re-envisioning medical coding with AI

In an era where healthcare organizations grapple with myriad challenges—from workforce shortages to overwhelming administrative burdens and complex billing issues—the need for innovation in medical coding has never been more critical. 

Hamid Tabatabaie, President and CEO of CodaMetrix, argues that the industry must move beyond the outdated fee-for-service coding paradigm. 

 He envisions a new approach where coding plays a pivotal role in supporting an array of critical initiatives, such as risk scoring for value-based contracts, engaging in research, and contributing to public health initiatives, especially in the aftermath of COVID-19.  

This shift is not just about addressing the immediate challenges of backlogs, inconsistencies, and untapped revenue that add to the burdens of healthcare providers and medical coders; it’s about reimagining the very foundation of healthcare administration to better serve the needs of today’s dynamic healthcare environment.  

CodaMetrix’s technology was designed, in part, to restore coding to its intended role—ensuring that medical procedures and treatments are necessary and appropriate, rather than merely facilitating billing processes. This ambition points toward a broader goal of improving healthcare efficiency and effectiveness by enhancing the way data is used and interpreted. 

“Unfortunately, we all have come to accept that claims data is notoriously suboptimal for any kind of clinical questioning. It’s great for being able to measure how many Medicaid patients were seen for pneumonia in the Northeast United States. You just follow the pneumonia codes and get your numbers,” Tabatabaie told DHI. But it doesn’t effectively gauge medical necessity. Machines, on the other hand, can learn both specific and necessary criteria without additional costs.” 

CodaMetrix has strategically aligned its operations with larger academic centers. The AI technology was developed at Mass General Brigham for 10 years before officially launching as CodaMetrix in 2019. Since then, CodaMetrix has partnered with other academic institutions including Yale Medicine, Henry Ford Health and University of Colorado Medicine.  

Academic centers often attract rare, complex medical cases which makes their data particularly valuable in training the AI.  

“Machines learn better with more diverse data. So, we start in these large health settings with rich data and then, as we gain more knowledge, we expand to smaller facilities in urban and suburban areas. This way, we can apply what we’ve learned more effectively,” Tabatabaie shared.  

CodaMetrix trains its AI using historical coding data provided by its healthcare system clients. The company refines and cleans the data to create a dataset that effectively trains a machine learning tool. The result is a tool that can perform the medical coding process that most hospitals and healthcare organizations still perform manually.  

The tool operates by analyzing patients’ medical reports, predicting the procedures and treatments administered, and determining the appropriate diagnostic and procedure codes for the care interaction. 

Beyond improving coding accuracy and efficiency, CodaMetrix’s AI solutions also aim to address significant industry challenges such as labor shortages and hospital revenue management. The shortage of medical coders is around 20-35%, Tabatabaie shared, and by automating coding processes, CodaMetrix helps reallocate human resources more effectively. 

The technology has also enhanced revenue management by reducing denial rates.  expediting claims submissions, and increasing the value of coding for participation in clinical trials and grants. 

“I’ve been in healthcare IT for nearly 40 years, and I believe what we’re doing now has the potential for the most significant impact,” Tabatabaie said. “Our work can change how we contract with payers, moving towards a system based on actual health outcomes rather than sickness. This has the potential to make a significant difference in the healthcare industry.” 

One concern about allowing AI and ML to automate healthcare processes is that they will perpetuate biases inherent in the data they were trained on. Can AI automation overcome human bias, and overcome data riddled with bias?  

“We offer configurability and control mechanisms for customers to address biases. Customers can designate certain cases or physicians for manual review when needed. We provide real-time analytics, highlighting cases where biases might exist, enabling customers to adjust configurations and reduce bias,” Tabatabaie explained. “Machines shouldn’t learn that part of human behavior.”