Predicting recurrence in patients with breast cancer from cumulative laboratory results: a new technique for the application of time series analysis.

We followed the cases of 26 consecutive postmenopausal patients operated on for primary breast cancer. Serum specimens were obtained each month for 1.5 years and stored at -80 degrees C until assayed for carcinoembryonic antigen (CEA) and other quantities. Ten patients developed recurrence, while 16 qualified as controls (no clinical recurrence for at least 1.7 years after the last venipuncture). Using the homeostatic autoregressive time series model, modified by us to be particularly sensitive to sustained deviations from the mean, we detected four recurrences by CEA without having any falsely positive alarms. Group-based reference limits and application of the unmodified homeostatic model were less effective (fewer detected and shorter lead time). Simulation studies, involving use of a mathematical model relating CEA concentration to tumor growth and using parameters estimated from patient data, verified this and indicated that at least five stable baseline values are needed to detect 100% of recurrences before they are detected by the group-based limit.