Leveraging AI Advancements for Coding and Clinical Documentation Improvement (CDI): Part II

In my last article, I provided thoughts on the promise of artificial intelligence (AI) to truly revolutionize coding as we know it today.

To get us grounded and lessen the confusion as we deal with endless promises from technology vendors, I provided some definitions. The diagram below will further clarify and unpack what ingredients are necessary to fuel advanced AI systems:

As we begin our journey, we need to identify our present pain points to illustrate for our senior leaders and stakeholders the challenges we are experiencing, and to define the need for leveraging AI technology to infuse our coding and clinical documentation integrity (CDI) functional areas and workflows.

The challenges I experienced while supporting large healthcare organizations included handling the evolving reimbursement and comparative quality models that require full, complete, and accurate code capture. These additional conditions may be documented outside out of the margins in which our current computer-assisted coding (CAC) technology can surface up insights and codes, as is the case with the social determinants of health (SDoH) codes, which can be documented by any clinician, not just a provider.

Also, as we compete for five-star ratings and our U.S. News and WorldReport rankings, ensuring that we capture and report the drivers that fuel our proxy systems for quality is essential. 

As we go beyond “capturing a DRG and moving on to the next case” approach of five or ten years ago, it is essential that we augment our resources for both CDI and coding to perform full code capture, while still ensuring that we don’t miss a beat with our drivers of reimbursement. Although CAC systems and improved electronic health record (EHR) workflows have slightly offset this resource drain, it hasn’t completely nullified this effect, which has resulted in backlogs and strains on our coding days. 

The professional fee coding component has also been subject to a multitude of challenges that stem from value-based care and risk adjustment, the addition of SDoH capture, challenges with obtaining timely provider documentation, and the plethora of payer rules requiring revisiting of coded accounts. 

Senior leaders are looking for ways to reduce the CDI and coding resource expenditure, while reducing the days/dollars held in coding. There continue to be challenges in finding and retaining outstanding coding talent in a remote marketplace, and the salaries for great coding talent continue to rise. Therefore, the business case for offsetting technology that can positively impact these metrics has never been stronger. As we begin our evaluation journey, the challenge I have had in my past organizations is explaining the current technology landscape versus what the advanced AI systems have to offer. Below is an excellent matrix to separate the current from the future state of the coding landscape:

The AI medical coding above is linked to a use case I will provide in my next article. In my past organizations, I did experience conventional CAC systems that made incremental advancements in deep machine learning and new AI-infused features to assist CDI and coding, serving as a basis for providing advanced AI use cases.

Many organizations have implemented advanced speech or text CDI provider nudges to surface up documentation opportunities, and to resolve nudges, similar to how queries are reconciled by CDI and coding staff. 

Major CAC vendors are also improving the rate of CDI professionals (CDIPs) and coders agreeing with codes surfaced up by the natural language processing (NLP) engine and are declaring these as “confident” codes assigned a higher rate/value for consideration by the coder. Also, I have helped to design and implement advanced coding sequencing to lessen the coding burden, and CDI worklists infused with AI that can prioritize the worklists for the CDIPs – and also auto-assign the DRG to lessen the CDI burden. 

Finally, there are also systems that can facilitate pre-bill scrubbing of each case to identify the DRG, coding, and potentially, cases that have a high likelihood of payer denial, based on pre-defined criteria.  However, these systems are bound by rules and complex configuration requirements. 

The current mission to incorporate documentation capture, CDI, and coding best practices, leveraging advanced AI, is our new north star. The first step is to focus on care delivery utilizing ambient experiences to enable the provider to focus on the patient and the care experience, instead of documenting in the EHR during and after the visit. Also, this technology can leverage nudges to notify actions by the provider. Over time, the need to nudge the provider and downstream interaction with the clinical documentation integrity specialist (CDIS) and the coder should decline, as deep machine learning kicks in for non-value-added nudges.

Also over time, the provider should incorporate the documentation tips into future documentation to reduce the nudge volume. AI can assist the CDI workstream by improving not only the worklist prioritization, but also reducing the volume of cases requiring a CDI review.

Finally, reducing the work types and specialties that require AI-assisted coding versus autonomous coding will be a focus of the next article, as we review successful use cases.

EDITOR’S NOTE:

Additional articles in this exclusive ICD10monitor series by Cassi Birnbaum include the following:

• Part III:  Envisioning Migration from Conventional CAC (NLP) to AI by Providing Coding and CDI Use Cases

• Part IV:  How to Prepare an AI Business Case for Coding and CDI, Choose a Partner, and Define Metrics

• Part V:  Setting up a Successful Implementation Infused with AI Governance