Breast cancer is the most common type of cancer, and the second-leading cause of death for women. Women with breast cancer have a far better chance of survival if the cancer is discovered early.

Currently, the most accurate test for breast cancer is an open surgical biopsy. But because this test has risks, as do all surgical procedures, doctors often use a less invasive and less expensive diagnostic test called fine-needle aspiration biopsy to determine whether a patient has cancer. The test involves extracting a small amount of tissue just below the surface of the skin for analysis. While it is more convenient, safer and cheaper than an open surgical biopsy, fine-needle aspiration has a greater risk of producing false results.

When analyzing the extracted cells, physicians look for different attributes, such as irregular size, texture and concavity. The presence or absence of cancer can be assessed by combinations of attributes, or rules. For example, one rule might be that a patient has cancer only if the cell is irregular, having a large size and a particular shape. But there are many possible rules, and no single rule is perfect, with each bearing a different risk of producing false results for certain types of patients.

Using data from nearly 600 patients, Professors Rajeev Kohli and Kamel Jedidi, working with Ramesh Krishnamurti of Simon Fraser University, developed a procedure for identifying the best predictive rules. The researchers set out to find the efficient frontier, which describes a set of rules that trade off between false positives and false negatives. The researchers found that the best predictive rules — those that resulted in the most accurate diagnoses — were those that identified a certain minimum number of cancer-predicting characteristics.