Don’t be scared off by the price tags. The higher your volume of claims, the more reasonable the pricing seems. And keep in mind the cost savings you might realize in reduced denials and faster revenue. Of course, if you already have few denials and low days in A/R, these might not be the systems for you.

As you shop, keep your users in mind. Knapper looked at some systems that she thought were too sophisticated for some of the billing staff she works with. Several members “of my staff don’t have the computer skills they would have needed to get the information in correctly.”

Make sure, too, that the system will allow you to create unique codes or modifiers, such as those sometimes required by state Medicaid programs.

Thinking solutions

Most revenue-cycle management programs run by following consistent rules: For Payer X always do Y. Vendors in the third category of coding tools — natural language processing software, including CodeRyte, PlatoCode, LingoLogix, and A-Life, among others — offer software that seems to “think” about the appropriate code based on the documentation.

The basic premise of natural language processing is “to make human beings interact with computers without forcing human beings to write in coding language,” explains Andy Kapit, CodeRyte’s CEO. Send the system any electronic document — Word, electronic medical records, you name it — via the Web and it gives back automatically assigned diagnoses and procedures codes, along with their definitions. Coders then confirm the code.

CodeRyte accomplishes this feat using statistics. The Web-based system identifies patterns of language that humans — actual coders — have decided are important in thousands of coded claims, and assigns a code based on the statistical consensus, Kapit says. “The engine doesn’t miss stuff so long as there is any indication in the notes; humans might not consider [certain language] important each time, but the engine would.”

If CodeRyte can’t identify a pattern, it alerts a human, which happens about 5 percent of the time, according to Kapit, who adds, “Usually if the engine couldn’t code it, a human can’t, either.” Typically, information is missing.

CodeRyte also uses a competency assessment model: If the computer can code it, it provides a statistical analysis of how many times a human at this practice location has ever changed the codes it offered for this sort of documentation. Then customers have the option of not even reviewing the codes, instead sending them straight to billing. Radiological Associates of Sacramento, Calif., a CodeRyte client, is now sending 20 percent of its notes directly to billing with no human intervention, according to Kapit.

Natural language processing seems to be the wave of the future. CodeRyte’s model isn’t the only one out there. For instance, instead of using statistics, PlatoCode’s software “reads” each word, sentence, and the document as a whole to formulate a response — a sort of artificial intelligence.

For now, though, the use of natural language software is somewhat limited, partly because it takes time to build knowledge of all the medical processes out there.

Can EMRs solve it all?

The fourth major type of product that can help your coding is the electronic medical record. Many EMRs include automated coding functions; they “read” your documentation and offer a coding suggestion. It’s particularly handy for those pesky E&M codes, and improved coding is one way practices get a return on investment from their EMR.

But coding and billing experts are a little wary about using an EMR exclusively.

Sharp’s physicians were using an EMR with coding capabilities — but they ended up turning off that functionality. The dictionary their system ran was simply too static to be very useful. Suppose a physician was documenting a laceration. The practice’s EMR wouldn’t provide a code; it could only identify the term “wound,” the term prioritized in CPT.