A method to predict breast cancer stage using Medicare claims

Area of Science:

• Oncology
• Epidemiology
• Health Informatics

Background:

• Cancer stage is crucial for predicting patient outcomes in epidemiological studies.
• Medical insurance claims datasets often lack cancer stage information, limiting their research utility.
• Developing algorithms to predict cancer stage from claims data is essential for expanding epidemiological research.

Purpose of the Study:

• To develop and validate an algorithm for predicting breast cancer stage using claims-based data.
• To enhance the use of medical insurance claims data for breast cancer epidemiology.

Main Methods:

• A cohort of 77,306 women aged 66+ with stage I-IV breast cancer was identified from the SEER-Medicare database.
• An algorithm was formulated using demographic, tumor, and treatment characteristics from claims data.
• Logistic regression models were used for prediction equations, with performance evaluated using sensitivity, specificity, PPV, and NPV in a validation set.

Main Results:

• The algorithm demonstrated high accuracy in identifying early-stage breast cancer (Stage I/II).
• The equation predicting Stage IV disease achieved 81% sensitivity and 89% specificity.
• The combined equations accurately identified 98% of patients as having Stage I or II disease.

Conclusions:

• A claims-based algorithm can effectively predict breast cancer stage, particularly for early stages.
• This prediction method significantly enhances the value of claims-based data for breast cancer research.
• The approach may be adaptable for developing algorithms for claims-based epidemiological studies of other cancers.