Determination of colonoscopy indication from administrative claims data

Area of Science:

• Gastroenterology
• Health Services Research
• Biostatistics

Background:

• Colonoscopy quality indicators, such as polyp detection and complication rates, can vary based on the reason for the procedure.
• Accurate classification of colonoscopy indications is crucial for reliable quality assessment and outcome studies.

Purpose of the Study:

• To develop and validate algorithms for classifying colonoscopy indications using administrative data.
• To facilitate large-scale studies on colonoscopy quality and patient outcomes by leveraging Medicare claims data.

Main Methods:

• Linked 14,844 colonoscopy reports to Medicare Carrier and Outpatient File claims.
• Developed classification algorithms using classification and regression trees (CART) and linear discriminant analysis (LDA).
• Utilized ICD-9CM, CPT/HCPCS codes, patient demographics, and site of service as predictor variables.
• Validated algorithms on a separate test set of 7,329 procedures.

Main Results:

• Algorithms demonstrated moderate sensitivity and high specificity (over 95%) for identifying average-risk screening colonoscopies.
• Sensitivity for diagnostic colonoscopies ranged from 77% to 89%, with specificity between 55% and 87%.
• An LDA algorithm with 30 variables showed higher overall accuracy compared to CART or LDA with fewer variables.

Conclusions:

• Algorithms utilizing Medicare claims data offer a viable approach for classifying colonoscopy indications.
• These methods are valuable for research on colonoscopy quality and outcomes within the Medicare population.
• Further validation is recommended for application in diverse patient populations.