Accuracy of coding in ICD-10 pilot varies, report says

Medical coders participating in an ICD-10 coding pilot produced accurate coding using the complex new system less than two-thirds of the time, according to a report by two healthcare IT industry groups.

Accuracy rates varied widely by type of medical condition coded. For example, a case of acute gastritis without bleeding was coded accurately 100% of the time in one batch or “wave” of test results, while “chest pain, unspecified” was coded accurately in only 34% of records tested in a different batch.

The federal government is requiring that hospitals, office-based physicians and all other “covered entities” under the Health Insurance Portability and Accountability Act of 1996 must convert to ICD-10 by Oct. 1, 2014. The American Medical Association is pushing for a delay in the costly and difficult implementation, which has already been delayed twice dating to the late days of the George W. Bush administration.

The 54-page report (PDF) presents the results of the national pilot program begun in April and completed Aug. 30 that included all participants in the claims processing cycle—including more than 30 provider organizations, as well as software systems vendors, health plans, billing companies, claims clearinghouses and the government. The research was conducted by the Healthcare Information and Management Systems Society and the Workgroup for Electronic Data Interchange.

Lisa Gallagher, vice president of technology solutions at HIMSS, warned it would be unfair to blame the coders for any negative results. “It didn't mean the coder didn't know what they were doing,” she said. “It may have meant the medical documentation was not precise” and subject to differing interpretation. “It was a process. If we say the coder made a mistake, that's not an accurate way to portray it.”

In the real world, “there are all kinds of supporting functions that end up influencing how a document is coded,” Gallagher said. With ICD-10, “you're moving to a place where there is so much specificity in how things are coded. They may not know they need to provide documentation to a coder to reach a certain granularity.”

Healthcare providers from across the country donated de-identified patient medical records for use in the project. The records for a selected number of test cases were then coded using ICD-10-CM (clinical modification) and ICD-10-PCS (procedure coding system) by volunteer coders from various organizations who had been approved by the American Health Information Management Association. They produced a repository of ICD-10 coded records, representative of the most common medical conditions in healthcare, to be used as an answer key. Copies of the medical records for these test cases, stripped of their coding, were then accessed by pilot participants for their coders to read and code.

The pilot did not delve into the impact of ICD-10 on claim reimbursement rates or coder productivity attributable to the change from ICD-9 codes in current use to ICD-10.

“Among the most critical aspects of implementing the ICD-10 transition is the challenge and the complexity of end-to-end testing,” according to the report. “This is a challenge weighing heavily on the minds of payers, providers vendors and business partners alike as it is perceived to be costly and burdensome—requiring trained personnel, a well-planned and fully funded testing plan and process and possibly the assistance of external support.”

Gallagher said the takeaway from the pilot should be that the healthcare industry needs to work together to meet the ICD-10 challenge. “This is an example of how folks should collaborate to do their testing,” she said. “It's people working together. You can do testing however you want, but a collaborative effort is the only way we're going to get across the finish line.”

Some of the lessons learned from the pilot study include:
  • Coders often confused the number “0” (zero) with the letter "O,” and the number “1” (one) with the letter “l.”
  • Occasionally, coders only coded the diagnosis but forgot to code the procedures.
  • Most errors were functional – for example, records were not coded completely or codes were associated with the wrong medical test case numbers.
  • Some coders did not specify the type of some chest pains, not consulting information in the medical record that differentiated them from atypical pains.
  • Coders occasionally “went on auto pilot,” relying on their coding software instead of referencing their own code books.
Follow Joseph Conn on Twitter: @MHJConn



Loading Comments Loading comments...