Predictive Models of 15-year Survival Developed Based on Reliable Data

Estimating survival for patients with early breast cancer for up to 15 years is possible, report researchers from Duke, who used a Markov model that incorporated data from a recent meta-analysis of the EBCTCG (Early Breast Cancer Trialist’s Collaborative Group) for this purpose. Further study will determine whether the model can estimate survival over a lifetime.

The ability to estimate long-term survival is important for making treatment decisions, and for healthcare policy and economics, including the drug approval process. At present, pathological complete response is considered a surrogate marker in the neoadjuvant setting, but there is no surrogate marker in the adjuvant setting for long-term survival of patients with early breast cancer, according to Christina I. Herold, MD, Duke Comprehensive Cancer Center in Durham, NC. Dr. Herold presented the data on the Markov model at a poster session at the San Antonio Breast Cancer Symposium.

One example of the need to model survival is the controversy about FDA approval of bevacizumab for the treatment of metastatic breast cancer based on improvement in progression-free survival (but not overall survival).

The investigators developed a Markov model to represent events that recur over time. They used discrete-time Markov models with intervals or cycles of 1 year. A mutually exclusive health state (dead or alive) was defined. Transition probabilities representing the chance of dying were then applied to each interval in succession. EBCTCG data provided breast cancer mortality rates for estrogen receptor (ER)-positive and ER-unknown patients randomized to receive tamoxifen or no tamoxifen in these trials. Then they estimated the 5-, 10-, and 15-year probability of dying using a specific formula.

Model 1 verified the model using 15-year EBCTCG data to model 15-year mortality rates on tamoxifen or no tamoxifen and compared these rates to the empiric 15-year EBCTCG mortality rates. Model 2 predicted extended survival using 5-year EBCTCG data to model 15-year mortality rates on tamoxifen or no tamoxifen and compared these rates to the empiric 15-year EBCTCG mortality rates. Model 3 estimated the effect of tamoxifen on 15-year life expectancy in women age 40-70 years incorporating age-sex-race-specific mortality and breast cancer-specific mortality.

The authors stated that further study is needed to evaluate the durability and relevance of these models of survival based on early survival data.