Billing code-based algorithm helps identify PAD cases

Twitter icon
Facebook icon
LinkedIn icon
e-mail icon
Google icon
 - endovascular, peripheral vascular disease
Source: Spectranetics

Researchers at the Mayo Clinic designed and validated an algorithm using billing codes that accurately identifies people with peripheral artery disease (PAD) in a community-based sample.

The study, led by Jin Fan, MD, of the Mayo Clinic’s Gonda Vascular Center in Rochester, Minn., was published in the December issue of the Journal of American Medical Informatics Association.  Fan and colleagues used administrative data from the Mayo database to conduct a training and validation analysis. They then tested the algorithm in a community-based sample from the Rochester Epidemiology Project.

They drew the Mayo sample from patient encounter data from 1997 to 2008 and extracted encounters with PAD-related ICD-9-CM diagnosis/procedure codes and CPT-4 diagnostic procedural codes. Patients were randomized to either the training or validation groups. The training group included 11,356 patients, with 67.1 percent meeting the vascular laboratory criteria for PAD. The validation group also had 11,356 patients, 66.7 percent with PAD.

Fan et al determined that 13 billing codes—eight ICD-9-CM diagnosis/procedure codes and five CPT-4 diagnostic procedural codes—independently predicted PAD status. The model worked equally well in both groups, with a billing code score of 8 or greater providing the highest sensitivity and specificity, at 86 percent and 83 percent, respectively, in the training group.

The positive predictive values and negative predictive values were also similar in the two groups, with a positive predictive value of 91 percent and a negative predictive value of 74 percent for a score of 8 or greater in the training group. The researchers also compared the results to a simpler algorithm.

Both the model-based billing code algorithm and the simpler algorithm were applied to the community-based sample, which was split into scores of 0 (people without PAD), 1-7 and 8 or greater. The simpler algorithm was less sensitive than the model-based algorithm at detecting people with PAD in the community-based sample.

“Our approach provides a strategy for rapidly ascertaining PAD status for genetic association and epidemiological studies of PAD,” Fan and colleagues proposed. They cautioned that the codes were designed for billing and reimbursement uses and may be limited for clinical and research purposes.