Computer Assisted Coding: Is the Future Here?

Vol. 15 •Issue 5 • Page 22
Computer Assisted Coding: Is the Future Here?

We’re already coding with encoders and handheld charge capture devices. When we get to computer assisted coding (CAC), will we need coders?

Take a look at the North Star in a telescope, a mere 430 light years from earth. Maybe it makes you consider events that can happen in the space of time, and how far advanced we are technologically (some fields more so than others). On a much smaller scale, HIM has seen significant innovations within the past 20 years. Many HIM departments use encoders to help coders assign codes. Clinics and hospitals are seeing physicians use mobile charge capture devices that help them document and code at the point of care. Now take a look through HIM’s telescope. Is computer assisted coding (CAC) that far off? There’s already talk of using artificial intelligence (AI) to automatically assign codes. If this is the future, what role do coders play in it?

Stardate 2005: Encoders

It’s not uncommon today to encounter automation within the coding process. Consider the logic or rules-based encoders in hospital HIM departments today. While not automatically assigning codes, “The encoder is a tool you would use, just as a code book is a tool,” said Mary Stanfill, RHIA, CCS, CCS-P, professional practice manger, HIM products and services for the American Health Information Management Association (AHIMA). “It just so happens to be a much more slick, computerized tool.”

In fact, it’s a code book and a whole lot more.

“We have a team of nosologists or coding experts whose primary job is to incorporate clinical content in our encoder product,” explained Ann Frischkorn, MBA, RHIA, product line manager, coding and classification solutions, 3M Health Information Services in Salt Lake City. “Once the coder enters a diagnosis or procedure description, Frischkorn explained, he/she is prompted with a series of questions to answer to derive the best code. For example, “A coder might put in a diagnosis of pneumonia and the system will offer a list of specific pneumonia related diagnoses.”

The encoder is stocked with information from the American Hospital Association’s Coding Clinic, the American Medical Association’s CPT Assistant, Centers for Medicare and Medicaid Services (CMS) publications and information from 3M’s “nosology hotline,” which clients can call when they come across something new.

But if it can’t code for you, can it at least supplement coders’ knowledge?

“Even within an HIM department there are coders with varying levels of experience who can get value out of using encoder software,” said Frischkorn. “The key to successful coding is more than just the tool. Coding is an art and a science, the science of using the software product, but also the art of understanding what a physician is trying to say in the documentation. It’s almost like you’re a detective when you’re coding.”

Surely we can automate away such mysteries, right?

Stardate 2005: Charge Capture

Charge capture is talked about in relation to the revenue cycle, the idea being that the quicker documentation is captured and coded, the faster the provider gets paid, according to David Delaney, MD, vice president of business development for Boston-based MedAptus Inc.

“What our product does is provide a tool for clinicians to have an easy means of capturing the information to generate a charge at the point of service,” Dr. Delaney explained. “Another thing it does is check charges against correct coding initiative rules and LMRPs at the time of entry, flagging any issues for providers to correct, real-time.”

But charge capture also impacts coding, as demonstrated at the Lahey Clinic in Burlington, MA, where some 3,000 patients a day are treated on an outpatient basis by 600 providers, of whom 225 are using the MedAptus system.

“We have 48 coders located out in the clinical areas very close to where the physicians do the outpatient visits. They work elbow to elbow with the physicians, provide education and feedback, and do audits to measure for compliance,” said Cynthia A. Trapp, CHFP, CMPE, CPC, CCS-P, director of professional coding at Lahey. Her coders are using the MedAptus system alongside the physicians. “Physicians will code with the MedAptus software, and then coders will review any edits they find. They have a PC companion that mirrors pretty closely what the physician is doing.”

Although it’s not uncommon for physicians to code their own documentation in this setting, Trapp explained, the movement to this automation has streamlined the coders’ work. “Before we had this automated process, everything was done on paper. The physicians would have an encounter form or billing ticket they would check off with a finite list of [procedure and diagnosis] codes that would get batched up at the end of the day.” The coders would review the batches as necessary and the batches would get sent off to billing, Trapp added.

While the clinic hasn’t eliminated coders with the automated process, “We have been able to eliminate the charge entry function in these areas, thus allowing us to redeploy charge entry staff into other roles,” said Trapp. But it has reduced some of the coders’ remedial tasks. In addition, “We’ve seen an increase in communication and collaboration between the physician and the coder to be much more accurate and compliant. It has been a tool for both parties, to ensure accurate coding,” she observed.

According to Dr. Delaney, the MedAptus solution is used in inpatient settings as well. In addition, his company has numerous vendor partnerships, enabling end-users to benefit from integrated clinical applications such as e-prescribing and dictation.

But while this is integrated, automated workflow that helps coders, it’s not quite CAC.

AI: Is the Future Now?

AHIMA’s Stanfill thinks CAC is just a matter of time—the question is how much time. Given AHIMA’s recent workgroup study on the topic (see Journal of AHIMA/Nov-Dec 2004), she begins by explaining why the association prefers the CAC acronym to “automated or “automatic” coding.

“Automatic implies there’s no human intervention,” said Stanfill. Based on conclusions from volunteers including vendors, users, informaticists, physicians and individuals in the field of natural language processing (NLP), “They agreed that today there is no fully automated system to assign codes,” said Stanfill. “Even when you talk to the vendors developing software applications that can automatically suggest codes, they all recommend you get the codes reviewed and edited by a person.”

That said, there have been inroads on the AI end. Just talk to Andrew B. Covit, MD, CEO of Artificial Medical Intelligence Inc. in Eatontown, NJ. Discussing his EMscribe Dx product, he explained, “It scans a document, identifies key words or phrases that define a given coding opportunity and matches them to appropriate ICD-9 codes. In addition, EMscribe effectively avoids overcoding by applying the product’s built-in rules that effectively block the code match in certain key situations.”

By employing a variation of NLP, “We capture how physicians speak, their medical ‘slanguage’” said Dr. Covit. For example, “The standard dictionary has many different terms for ‘acute heart attack’ or ‘myocardial infarction,’ however most doctors do not speak using the official terminology. They may say that the patient is suffering from an ‘acute MI’ or a ‘new infarct’ but they are slang terms that are not listed in the ICD-9 terminology.” With AMI’s EMscribe product, “We’ve augmented the dictionary to be able to recognize those slang terms to make the solution realistic and useful in the real world,” Dr. Covit said.

“It’s taken us 3 years’ worth of development to get to the point where we are now capable of automating the coding process,” added Stuart Covit, executive vice president for marketing and administration for AMI. “We’re now confident about the reliability and computing powers this product offers.”

AMI says their technology is ripe for either the physician office setting or an acute care hospital. EMscribe DX is currently in beta testing at Robert Wood Johnson University Hospital in Brunswick, NJ, and the product was officially launched at the Healthcare Information and Management Systems Society conference in Dallas Feb. 13-17.

Of course, NLP-like solutions aren’t the only ones to hit the showroom floor.

Is the Future Structured?

Known as structured input or codified input, AHIMA’s Stanfill explained, “Structure doesn’t use NLP at all. With a structured input type of system, it’s driven by the health care provider who is documenting care.” These systems work on the same concept as macros,” she explained. For example, Stanfill offered, “Anybody who codes colonoscopies can tell you that doctors always do certain steps and in certain orders.” With preset menus for things such as anesthesia, prepping and draping, and insertion of scope, “The structured input approach capitalizes on that routine.”

The AHIMA workgroup found some structured input systems being utilized in specialty physicians groups such as gastroenterology. “They focused on CPT coding in this system because the procedures are very defined, fairly predictable steps,” said Stanfill.

As with NLP, Stanfill added, “Neither of these technologies are totally futuristic ‘Star Trek’ kinds of things, but their use is limited in health care,” said Stanfill, who added that one prerequisite to either technology is fully electronic text.

Beyond that is, of course, the electronic health record (EHR), something that both Stanfill and Frischkorn agree will move CAC forward.

Frischkorn sees it pushing structured coding. “I think with the movement toward EHRs, you’ll see more structured text output. Right now I’m seeing more NLP than structured text, but that will shift as we move forward.”

If the technology could go either way, one might question where this leaves the coders.

Domo Arigato, Coder Roboto

Technological innovations do change jobs. Trapp recalls when she introduced automated charge capture to her coding staff back in 2001. “At first the coders were afraid,” said Trapp. “They wanted to know: What is going to happen to our jobs?” Their roles have changed, but certainly they haven’t been eliminated. Nor does Trapp see a day when that would happen.

“I think because the intensity and complexity of coding is growing so much that, even with an automated system, you have to have skilled people who are going to review what is being [generated] in the system before it’s billed out.”

And after working with NLP, even AMI’s Covit doesn’t discount the role of the coder. “We believe there should always be some form of human intervention, be it at an administrator level or whatever. The fact of the matter is, there needs to be human intervention in this process at some point,” he said.

“I’ve seen NLP more in the professional setting as opposed to the inpatient setting,” commented 3M’s Frischkorn. “Depending on who you talk to, errors can range from 10 percent to 50 percent.” She compares NLP to speech recognition technology and the impact it has had on the job of MTs.

“Many MTs have gone from typing to becoming editors/experts on what is deemed a good transcribed report. The same thing will happen with coders. But I don’t see the validation role ever going away. Whether it’s structured or unstructured, I think the skill sets of the coders will need to expand even more than what they are now.” From Frischkorn’s perspective, “The computer will take care of coding standard mammograms and other repetitive procedures, but the coder will need to audit the coding and work with the physician, probably even more so than what’s required today.”

Nonetheless, “It will probably be a while before we see these systems in the inpatient environment,” said Stanfill. “We need electronic text and systems designed to be able to handle multiple inputs from multiple health care providers. Now we’re seeing them in specific pockets of outpatient reporting.”

Trapp added, “I haven’t seen a [CAC] system that’s gotten it perfect. The language might lead you to use one code, but to use codes you really have to understand that language.”

Stanfill agreed, but added, “We’ll get there. It’s not if, but when. I don’t know that I can say when that will be, but when we do, it’s going to be awesome.”

Linda Gross is an associate editor at ADVANCE.