Salford Analytics and Data Mining Conference 2012

Insight For Data Enthusiasts • San Diego, CA • May 24-25
Training May 21-23 • Welcome Reception May 23

You are here:Home»Conference»Sessions»Algorithms To Identify Second Breast Cancer Events Using Administrative Healthcare Data

Algorithms To Identify Second Breast Cancer Events Using Administrative Healthcare Data

State tumor registries do not routinely ascertain cancer recurrences therefore epidemiologic studies of second breast cancer events rely on manual review of medical records which is expensive and time–consuming. Using CART, we sought to develop algorithms to identify second breast cancer events using automated healthcare data. We extracted automated data including procedures, diagnoses, and medications on 3152 women who were enrolled at Group Health Cooperative, an integrated healthcare delivery system, and were diagnosed with a first primary invasive early–staged breast cancer between 1993-2006.

We created nearly 500 potential predictors of a second breast cancer event from the automated data. The occurrence of a second breast cancer event was ascertained from medical record review and served as the gold standard. We specified a range of costs for misclassifying a true event as a non-event and vice versa in CART to develop a “menu&rdquo of algorithms that researchers can select from based on whether their priority is sensitivity, specificity, or positive predictive value when identifying second breast cancer events from automated data. The algorithms were parsimonious with good accuracy and have the potential to increase efficiency and reduce costs of epidemiologic and health services research on breast cancer treatment effectiveness and outcomes.

back to top