Computer-Assisted Coding – It’s Not Artificial Intelligence

CAC is not AI.

I know that numerous articles and even some books have listed the pros and cons of artificial intelligence (AI). Discussion of the technology in relation to computer-assisted coding (CAC) dramatically ramped up with the implementation of the 10^th version of the International Classification of Disease (ICD) clinical modifications and procedural coding system. 

I’m not one to rehash old information, and I’m certainly not against cutting-edge technology. I like my smartphone, I’m a super geek about my laptop, and when software and hardware deliver on promises made, I’m blowing the trumpet to let others know.

From a perspective of personal experience, though, I can honestly say that for me and many I work with, the jury is still out on the viability of this technology being rolled out in all areas of healthcare coding for physicians and hospitals across our country. 

Previously, in the 1990s, I was working as a coding manager and compliance liaison for a national emergency department physician practice management company. I was enlisted to roll out some new software and hardware within our organization to be used for our clients. This technology was sold by a company called A-Life Medical, and it was not yet referred to as CAC. While I don’t know the details of the invoicing, upper management assured me it was at least $1 million.

The software was based on something called “natural language processing,” and it was previously utilized to properly analyze and send emails related to U.S. military needs during the Operation Desert Storm operations in Iraq. During brainstorming sessions between the inventors and some investors, it was thought that the technology could properly read physician transcription to properly assign an evaluation and management (E&M) code and some minor procedure CPT codes. The software focused on appropriate template design, header information, and supposedly, some contextual information within the document.

Findings of the A-Life coded information showed that only 25 percent of records needed to be reviewed and/or touched up for additional or revised coding, and the other 75 percent were “error-free,” based upon agreed-upon guidelines for code assignment.

Fast forward about 10 years, and A-Life Medical was purchased by a major healthcare software and publication organization – and with an eventual technology update and changes in terminology, voila, we have CAC.

Okay, enough background. What is up with this technology today, and is it as wonderful or scary as people make it out to be? There are usually two questions I’m always asked about CAC, which are:

• Does it reduce hours related to coding?
• Does it increase productivity for coders?

Organizations selling CAC products will surely deliver a positive answer to these questions. But is it always the truth? I can quite honestly say – no! Let me share some personal responses to the utilization of this technology over the years, coming from individuals who have worked with it every day and in various setups for physicians and hospitals. These are simple and short statements:

• Frustrating
• No confidence in the product and technology
• Doesn’t increase quality and productivity
• Decreases productivity
• Too many mouse clicks to do normal coding abstraction
• Software doesn’t consider current coding guidance
• Software doesn’t understand basic and advanced coding concepts
• Software utilizes information from areas of the record that are not valid or relevant to the current service(s) being provided to the patient
• Software doubles the number of diagnosis codes to be assigned, meaning that coding staff review 50 percent more false positives to determine they are not appropriate or necessary
• Software cannot properly code trauma care for inpatients
• Procedure sequencing is problematic and must always be reviewed
• Present on admission (POA) is usually incorrect
• Scanned documents may not be analyzed appropriately
• Software updates and system changes necessary to correct problems
• System compatibility and implementation issues are one of the greatest challenges to address first, or the CAC experience will only be negative
• Coding accuracy actually decreased
• Workflow is hard to develop and establish
• Software updates can undo previous enhancements
• Attempts to help the software “learn” failed

You may have gathered that these are all negative in their tone. And while there are some positives related to the technology, they don’t rise to the level and weight of importance, advancement, and warm fuzzy feelings we’d all like to enjoy.

Adrienne Commeree in The Coder’s Guide to Physician Queries said:

“During AHIMA’s (the American Health Information Management Association’s) pilot testing of CAC software, the organization weighed in on some of the potential issues with using CAC software alone (with no human intervention). AHIMA noted that within specific areas of the pilot CAC testing in ICD-10, the coders did not accept 75 percent of the diagnosis codes presented, and they did not accept 90 percent of the procedure codes presented within the code sets.”

This has also been my personal experience when it comes to hospital outpatient and inpatient coding. Instead of the technology being one of many tools for coding staff to utilize in performing the functions they were hired to perform, they have become software auditors, and are now asked to also perform coding tasks, and inevitably, serve as proxy-vendor beta testers for non-working technology.

If a coder is assigned a 290-day inpatient stay and the patient has multiple acute and chronic conditions, along with multiple surgical interventions, CAC software increases the need for the coder to re-analyze what they would already analyze during their abstracting process. 

Can CAC help provide a starting place with some codes? Sure. Can CAC help provide some clinical documentation improvement (CDI) insight into the patient record? Sure. But do those minor benefits outweigh the costs of all of the negative issues that you also get? Maybe…maybe not. There are a host of things to consider, such as:

• Disparate competing systems that don’t interface
• Workflow design that’s insufficient for the needs of HIM and CDI
• CAC isn’t the “sole source of truth” for coding abstraction
• A trial period of the software should show true ROI in the form of the measured quality and productivity, and not the vendor claims, but anecdotal gains in quality, productivity, reduced denials, and compliance are insufficient reasons for investment
• Coders become auditors and editors, but many didn’t sign up for that job description
• Multiple documents within multiple systems must go through CAC for analysis
• You get what you pay for. Coding candidates whose only experience is with CAC solutions will likely never rise to the level of professional development of those who throughout their career continued to utilize their manuals and recognize the necessary but limited value of encoding tools. The thought processes for gathering complications and comorbidities, and major complications and comorbidities (CCs and MCCs), along with parsing notes of so many kinds, can’t be simply overcome by hope for AI and edit staff to run things.

Yes, of course, results will vary from organization to organization because of setup requirements, implementation, buy-in from staff, etc. But so far, I haven’t seen all this proof in the pudding that people speak of.

Where might this technology be most beneficial? 

• Office or clinic-based settings

In an office or clinic-based settings with short, succinct records and documents for the care provided, CAC could be useful.  The repeat disease processes or injuries would flow well through the technology and true measurable success could be realized. 

• Same-day surgery (SDS) providers

If you don’t have enough staff to perform all the SDS coding, the CAC might benefit your staff member(s) by providing a good starting place to complete the coding process.

• Freestanding ambulatory surgery centers (ASCs)

With a known list of procedures that can be performed, along with appropriate modifiers, I would think that CAC technology could easily handle an extensive list of patient procedures performed on a daily basis for this provider type. Challenges to be overcome relating to items mentioned previously with contextual analysis, coding guidelines, and references, and the ability to make the software to adapt to the client needs.

In a recent discussion concerning CAC, I was asked if I thought coders were afraid of losing their jobs to CAC. I stated that those who are knowledgeable are not afraid of any such thing. CAC is not AI. It isn’t alive and self-aware. It does not possess critical thinking and problem-solving skills and only performs at best upon what it has been programmed to accomplish.