Prediction of the long-term survival in breast cancer patients according to the present oncological status

After a breast cancer diagnosis, single or multiple events can occur during follow-up (recurrence, metastasis, and death). An analysis of long-term survival should take into account not only the initial characteristics of the patient, but also her oncological status (that is, her history) after surgery. For this purpose, we used a technique proposed by Klein, Keiding and Copelan (1994), to predict the probability of a patient being alive 20 years after surgery for a breast cancer, based on data concerning her oncological status at time t. The first step of the model was to estimate the hazard function for each event of interest (recurrence, metastasis, and death) in a Cox model including initial patient characteristics (age, tumour size, number of involved axillary lymph nodes and the Scarff, Bloom and Richardson (SBR) histo-prognostic grade) and time-dependent covariates representing the occurrence of intermediate events (recurrence and metastasis). The second step was to use these estimations to calculate the conditional probability of being alive 20-t years later for a patient, given her oncological status at time t (t < 10 years). In this second step, the method presented by Klein, Keiding and Copelan was extended to include non-proportional hazards. This model has been applied to a population of 3180 patients operated on for a breast cancer at the Institut Gustave Roussy between I January 1954 and 31 December 1983. At the time of surgery, the probability of survival at 20 years is 0.78 for all patients. Ten years after surgery, if no recurrence or metastasis are observed, the probability of survival at 20 years will rise to 0.89. If only a recurrence is observed, the probability of a patient being alive at 20 years will drop to 0.72. If a metastasis and no recurrence is observed, the probability of survival at 20 years will be only 0.18. If both recurrence and metastasis are observed the probability of survival at 20 years will be equal to 0.09. In conclusion, the model used dynamically appraises the prognosis and represents a new approach for studying the outcome of breast cancer patients having undergone surgery. Copyright (C) 2002 John Wiley Sons, Ltd.