An Answer to the ICD-10 Coder Productivity Fears?

As hospitals prepare to implement ICD-10, they increasingly are kicking the tires of computer-assisted coding software in an effort to take pressure off coders who expect to be overwhelmed with the massively larger sets of codes comprising the International Classification of Diseases and Related Health Problems 10th Revision.

Traditional coding software replaced coding books and acts like an electronic book. A coder no longer must leaf through a book, but scroll through lists to find an applicable code and enter it into the encoder.

ICD-9 has approximately 18,000 codes; ICD-10 has around 140,000, according to the Centers for Medicare and Medicaid Services. So scrolling, already an exasperating task with I-9, simply isn't going to cut it with I-10 without a major reduction in coder productivity-if the coders hang around. "We've already had two people retire with ICD-10 coming," says Sue Trewhella, associate vice president of revenue management at Geisinger Health System in Danville, Pa. "ICD-10 is the biggest issue bringing computer-assisted coding forward."

Computer-assisted coding, or CAC, software analyzes documents and generates codes based on specific phrases and terms within the document. The secret sauce is the use of natural language processing technology to identify key terms and phrases

The software also analyzes the context of the wording to determine if a particular instance requires coding. For example, the software can deduce if the term "cancer" is being used as a diagnosis and needs to be coded, instead of being used in a different context, such as a physician noting a family history of cancer.

Coders initially don't view computer-assisted coding with any less apprehension than they do ICD-10, Trewhella says. "The biggest issue with staff is that they were convinced they were going to be replaced by a machine," she says. "I had to convince them that computer-assisted coding was a way to support expansion and keep experienced staff." Another pitch was the expectation of reduced overtime to get back to a work-life balance. "We're not there yet, but it's something to look forward to."

 

Play it again

The same story played out at three-hospital Akron General Health System in Ohio, which is implementing CAC from OptumInsight and expects to be live in September, says Karyn Keay-Otte, program manager. "There is a myth it will replace coders and that's simply not true," she adds. "We had to be very clear with our coders and schedule several demos so they could see that the tool will enhance their experience."

Even "superstar" coders with interest in CAC at the University of Pittsburgh Medical Center had worries about the technology, but with a different twist, recalls Nancy Soso, executive director of health information management. They told me 'Whatever you're doing, don't slow me down."'

So far, the superstars are fine with CAC, but it's the non-superstars showing the biggest bump in productivity. UPMC has CAC software in three hospitals, and after two years coding productivity is up 21 percent. The case mix index-payments multiplied by the weight of patient cases with complex cases weighed higher-rose 8 percent, a $22 million increase in revenue, just in those three of the delivery system's 20 hospitals.

Gwinnett Hospital System in Lawrenceville, Ga., got its first-year return on investment within six months of go-live and its expected five-year ROI after the first year-and none of that related to case mix index improvement efforts.

Those improvements had previously been achieved and case mix revenue still rose 3.8 percent in the first year. The ROI came from less overtime and use of outsourced coders, quicker auditing that resulted in better justification of higher codes, and higher coder productivity, says Carol Fowler, director of health information management.

Productivity is the real ROI of computer-assisted coding, say those using or implementing the technology. Based on Canada's experience with implementing ICD-10, hospitals can expect a 50 percent drop in coder productivity when the code sets arrive, says Trewhella of Geisinger.

CAC won't mitigate all of the productivity losses, but it can cut off a big chunk. Geisinger, for one, saw a 20 percent increase in coder productivity. Trewhella is confident that the ROI will outweigh the cost of computer-assisted coding, but that analysis will take time: CAC is a unit-by-unit and hospital-by-hospital project, not something that can be done in a Big Bang, she says.

At four-hospital MultiCare Health System in Tacoma, Wash., the cost of CAC was just part of the migration plan to ICD-10, says Jenn Mykland, administrator of revenue integrity. "Rather than a 50 percent ongoing hit to coder productivity, this would be 25 percent ongoing hit because of the scope of ICD-10. And that's the ROI."

The cost for CAC is variable with each vendor, since some price per bed while others price by charges, says Keay-Otte at Akron General. "Did the cost hurt? No, because we had it budgeted in our ICD-10 plan prepared in late 2010."

A common misconception in hospitals about computer-assisted coding is that physicians also need to accept and train on the system. But doctors don't code on the inpatient side, they document, "so this is not a tool for physicians," says Terri Mayne-Jarman, an ICD-10 consultant at Point B Inc., and working with MultiCare.

The natural language processing technology in CAC has several facets, explains Fowler at Gwinnett Hospital System. The software first does pattern-matching, searching for specific words and suggesting one or more codes.

Gwinnett's vendor, OptumInsight, operates a national database of its clients' CAC coding experiences, which provides a baseline for statistical probability.

The database "learns" over time based on the information being fed into it, so the terminology in a hospital's NLP is continually tuned with knowledge gained from the hospital's CAC and the software at other Optum hospital clients. So if a patient comes into Gwinnett with chest pain, its CAC knows that hypertension and high cholesterol are common in Southern patients and will consider those factors in recommending diagnostic codes.

The learning also results in adding content to a hospital's NLP. A diagnosis for a rare heart disease inputted by another hospital into its NLP also will be added to the diagnoses being searched at Gwinnett.

As an example do the benefits of that function, the delivery system started coding open heart surgical cases in 2012 and the national database automatically assigned ICD-9 open heart codes to the operative report. Without the learning function, Gwinnett would have had to teach the vendor about the codes and have them built into the system, Fowler says.

CAC also conducts terminology matching and highlights appropriate terms to check for appropriate SNOMED CT and LOINC coding, among others. The software further applies rules by looking beyond common phrases for more context, such as a patient with lower abdominal pain who also recently had surgery in the region. In addition, the application includes productivity management tools. "I can tell how long it is taking my coders to work on one chart," Fowler notes.

A new feature in Gwinnett's software is the ability to abstract data within the CAC application. Coders previously had to log into another system to abstract particular data required by the delivery system or state, such as birth weight.

Two additional new features are coming soon-being able to stay within the CAC app and enter codes into the encoder, which works behind the scenes, rather than opening another window to log into the encoder; and a clinical documentation improvement module.

The module will enable quality management nurses and coders to review coded charts while patients are in the hospital to ensure entered codes are appropriate and check if better documentation or coding is needed. The CDI module also looks for documentation that is not present and whether there any supporting indicators that warrant checking to see if additional documentation is appropriate.

 

Making life easier

CDI software modules, part of an industry trend to get aggressive about clinical documentation improvement, make life much easier for the CDI teams looking at improving and correcting documentation, says Adele Towers, M.D., medical director of health information management at UPMC.

At UPMC, CDI software enables scanning of patient records to identify worklist items, so a CDI specialist "won't have to wade through 50 charts to find five that have opportunity," Towers adds.

UPMC has used computer-assisted coding technology since September 2008 from A-Life Medical, now part of OptumInsight, and is co-developing and testing an upgraded CDI module expected to launch in early 2013.

While the benefits of CAC can be significant, few would suggest rushing an implementation.

MultiCare took a three-phased approach to going live with computer-assisted coding software, from 3M Health Information Systems. In December 2011, it installed a CDI module to proactively analyze optimization of coding while patients are still in the hospital.

In January 2012, it turned on the database in the CAC software to tune the system by tracking what physicians were saying in their notes against how coders were interpreting the notes and then coding the information. Automated code suggestions started in July 2012. Like some other vendors, 3M hosts the the natural language processing engine to continuously train the software as multiple clients use the function.

The reason behind MultiCare's decision to buy CAC software in the first place was because it became clear with ICD-10 that the hit to coder productivity could reach 50 percent, says Jenn Mykland, the revenue integrity administrator. "No organization can double its coding staff."

At MultiCare, identifying and pulling data from the Epic EHR into the national language processor during implementation went smoother than expected, highlighting the importance of learning from peers who have done CAC installs, Mykland says, especially how they tackled the types of documents that go into the system and the order.

Documents have to go in the natural language processor in chronological order. Documentation for a patient presenting in the emergency department, then admitted, should have progress notes, medications, ordered tests, results and other pertinent information entered "from beginning to end, just like a book" says consultant Terri Mayne-Jarman of Point B Inc.

 

Doing it right

With any implementation of health I.T. come examples of decisions made in the early stages that turned out to be spot-on, and others that didn't stand the test of reality.

At Geisinger, getting the Precyse speech recognition-enabled dictating and transcription software along with computer-assisted coding-and storing the transcriptions as electronic text in the electronic health record-turned out to be an effective way to better leverage the use of data in the CAC software. Physician documentation, as a result, is in one place rather than in silos across the organization, Trewhella says.

As Akron General Health System ramps up testing in preparation for a September go-live, it has found a weakness in the basic functionality of CAC software that it didn't anticipate, says program manager Keay-Otte. The natural language processing technology works best on electronic documentation-discrete data elements or electronic transcriptions. But NLP does not read scanned and handwritten documents very well, which she estimates comprise at least one-third of documentation at Akron General. The hospital may have to go back to its cardiology information system vendor for an interface update because it's producing PDF files for reports instead of discrete data elements.

Akron knew going into the CAC implementation that NLP reading capability had some limitations, but knows in hindsight it should have approached vendors earlier about supporting textual electronic formats, Keay-Otte says.

When evaluating vendors, Akron saw significant differences in their competencies, she adds. "Not all NLP technologies are created equal, so it is important to get educated on NLP."

 

Information overload

During testing, MultiCare Health System found that too much documentation can diminish CAC's ability to recommend specific codes. "Understand the documents you can actually code from," says consultant Terri Mayne-Jarman. For instance, the EHR has nutrition codes, but coders don't cover those "so you don't want suggestions coming from nutrition notes," she explains.

When MultiCare first started working on CAC with 3M, it paired the vendor with coding and clinical documentation improvement teams, but didn't initially include the I.T. department. Mayne-Jarman advises others to not make that same mistake.

That exclusion caused headaches when the team was trying to get the Epic EHR to support CAC. For instance, the CDI team in its review process planned to view the electronic medication administration record from the EHR in the 3M system, but as it turned out, that would create a separate document every time eMAR is updated, making the information unwieldy and not helpful. It was a similar disconnect with vital signs: if they're taken every hour, the EHR generated 24 documents a day for coders and CDI personnel to wade through, when all they needed to see were major changes in the vitals.

 

Sidebar: Keeping Coders From Walking Out The Door

Four-hospital MultiCare Health System in Tacoma, Wash., went live with computer-assisted coding in early July 2012. The system invested plenty in coder training as part of the project, but executives now are worried that it turned its employees into accomplished ICD-10 coders who will take those skills to greener pastures, says Chief Financial Officer Vince Schmitz.

While it hasn't made any final decisions on how to avoid that, MultiCare knows other organizations are implementing coder retention policies, he adds. These include having coders enter into retention agreements and agreeing to pay back the cost of training if they leave before a certain time-such as before or within a year following the ICD-10 deadline-while also offering coders a bonus for staying a year beyond the deadline.

MultiCare was prepared to meet the October 2013 ICD-10 deadline. The pending change, which will tack on an additional year, gives it and other providers time to prepare at a slower pace and save resources, Schmitz says. The delivery system is ramping up CAC slowly, starting with using the technology for 10 percent to 15 percent of charts. A longer glide path will improve overall efficiency without overburdening the staff, he believes.

Schmitz expects coding productivity to increase by at least 40 percent as measured by how many charts are done in a day, with case mix revenue improvements of two or three percent. In a $1.5 billion delivery system, "the cost of CAC software is very, very small compared with the revenue improvement," he adds.

 

By Joseph Goedert, News Editor at Health Data Management