Researchers at Mayo Clinic in Rochester, Minn., have developed an algorithm using billing code data to help identify potential peripheral arterial disease patients.
The historical cohort study used data from all patient encounters at Mayo, including billing codes, over the 11-year study period (July 1, 1997, to June 30, 2008), totaling 1.2 million patients and 14.5 million unique encounters. A total of 61,005 patients with potential PAD had at least one PAD-related billing code. A total of 22,723 of these patients were evaluated for PAD in Mayo's noninvasive vascular laboratory. The patients were randomized to create matched training and validation datasets, and the relationship between billing codes and PAD status was analyzed, using the vascular laboratory results as the gold standard for a PAD diagnosis.
The researchers then developed an algorithm to correlate patients' billing codes with the actual presence of PAD. An analysis revealed 13 billing codes independently predicted a patient's PAD status to a statistically significant degree. The codes with the closest correlation with verified PAD status included:
440.24 —atherosclerosis obliterans extremities with gangrene
440.21 — atherosclerosis obliterans extremities with intermittent claudication
440.23 — atherosclerosis obliterans extremities with ulceration
440.20 — atherosclerosis obliterans extremities unspecified
440.22 — atherosclerosis obliterans extremities with rest pain
84.11—amputation of toe
The researchers also developed a simpler algorithm using the presence of one of the predetermined PAD billing codes (440.20-440.29 in ICD-9). Results showed this simpler algorithm was also reasonably accurate for identification of PAD in patients referred to the vascular laboratory and did not provide any additional positive diagnoses than the model-based algorithm.
To evaluate the feasibility of applying the algorithms to the larger community, researchers identified a sample of 4,420 local residents who were not referred to the Mayo vascular laboratory, but who were evaluated at the clinic during the study period and who had at least one of the PAD-related billing codes. The model-based algorithm was reasonably accurate in identifying cases of PAD, though the sensitivity of the simpler algorithm was much lower in the community-based sample than it had been in the lab-evaluated sample, which was "somewhat surprising" to study author Iftikhar J. Kullo, MD, of Mayo's Division of Cardiovascular Diseases and the Gonda Vascular Center.
Dr. Kullo reports Mayo is currently using a version of the model-based algorithm to identify potential PAD patients in the community for an upcoming study. "It will help to mitigate the effort- and time-intensive process of manual abstraction for all potential cases and controls," he says.
The algorithms could be successfully used at other organizations to quickly identify cases of certain diseases, says Dr. Kullo, provided they are vetted appropriately. "Because billing practices and ordering of tests and procedures may differ by institution, one would expect that such algorithms would perform differently at different institutions," he says. "We would recommend that if institutions use these algorithms, to first perform an initial validation to test how the algorithm performs for their particular set of data."
More Articles on Billing Codes:
AMA Introduces Its First-Ever Physician App
Preparing for the Productivity Gap: Protecting Your Revenue When ICD-10 Goes Live
Good Samaritan Begins ICD-10 Implementation a Year Ahead of Deadline