Billing code algorithms to identify cases of peripheral artery disease from administrative data

No prior work had resolved the accuracy of billing code algorithms in identifying peripheral artery disease (PAD) from administrative data. Prior research has shown that administrative data can be used to approximate clinical diagnoses, but the reliability of such methods remains uncertain. This gap motivated the need to evaluate how well billing codes capture PAD cases. Existing methods rely on ICD-9 codes, but their performance in community settings is unclear. It was already known that PAD is underdiagnosed and underreported in many populations. The need for accurate case identification is critical for public health and clinical research. Administrative data offers a scalable solution but requires validation. This paper contributes by testing and comparing two billing code algorithms for PAD detection.

The aim of the study was to construct and validate billing code algorithms for identifying PAD cases from administrative data. The researchers focused on evaluating the performance of these algorithms in both clinical and community settings. They sought to compare a model-based algorithm with a simpler ICD-9 code-based method. The motivation was to determine which algorithm better captures PAD cases without over- or under-estimating prevalence. The study targeted patients evaluated in a vascular laboratory and a community sample. The researchers wanted to assess sensitivity and specificity in different populations. They also aimed to provide a practical tool for PAD case identification. This work addresses the need for reliable methods to use billing data in PAD research.

The researchers extracted encounters and billing details from Mayo Clinic Rochester between 1997 and 2008. They divided 22,712 vascular laboratory patients into training and validation sets. Logistic regression was used to create an integer code score from the training data. This model was then tested in the validation set for accuracy. The model-based algorithm was applied to patients in the vascular laboratory. A simpler algorithm using ICD-9 codes 440.20-440.29 was also tested. Both algorithms were applied to a community-based sample of 4,420 patients. A manual review was conducted to compare results across methods.

The logistic regression model performed well in training and validation datasets with a c statistic of 0.91. In the vascular laboratory, the model-based algorithm had better negative predictive value. The simpler algorithm showed reasonable accuracy but lower sensitivity and higher specificity. In the community sample, the simpler algorithm had 38.7% sensitivity versus 68.0% for the model-based method. Specificity was 92.0% for the simpler algorithm versus 87.6% for the model-based one. The model-based algorithm provided more accurate PAD identification in community settings. It also performed reliably in patients referred to the vascular laboratory. The simpler algorithm was less sensitive in the community but more specific.

The model-based billing code algorithm had reasonable accuracy in identifying PAD cases from administrative data. It performed well in both clinical and community settings. The simpler algorithm was accurate for PAD identification in vascular laboratory patients. However, it was significantly less sensitive in the community sample. The researchers propose that the model-based algorithm is more reliable for broader use. They suggest that the simpler algorithm may still be useful in specialized settings. Their findings support the use of validated billing code algorithms for PAD research. This approach can improve the accuracy of PAD case identification in administrative data.