New method to estimate cancer survival

Download PDF Copy

Apr 17 2006

Cancer survival estimates are important to patients, physicians, clinical trials planning and resource allocation. Predictions by present statistical methods tend to lag behind current outcomes.

In the online version of Cancer, Dr. Mariotto and associates from the National Cancer Institute report a new statistical model to predict cancer survival.

The projection method described incorporates survival trends and as such generates more up-to-date estimates of long-term survival for newly diagnosed patients. This method consists of fitting a regression model across diagnosis years to each interval relative survival and extrapolating the rates beyond the data to the current calendar year.

The authors used the Surveillance, Epidemiology, and End Results Program (SEER) and the Connecticut tumor registry to investigate the predictive ability of the projection method. This was compared to predictions achieved with currently employed statistical methods such as the cohort method and the variable follow-up method.

The projection method was superior to existing statistical methods for predicting the survival of recently diagnosed patients. This was especially true for long-term survival in patients who had disease sites with an increasing and stable survival trend.

The 15-year cumulative relative survival (CRS) for patients diagnosed in 2003 was 61% for all sites combined, 97% for prostate cancer, 57% for colorectal cancer, 82% for breast cancer, and 53% for ovarian cancer.

Prostate cancer predictions are challenging to the model due to a rapid increase in all stages and local/regional survival rates observed from 1986-1991 and a slowing in trends after 1991. This is a result from the introduction of PSA screening with a greater proportion of patients detected with earlier stages and a better prognosis. The model overestimates CRS for men with prostate cancer diagnosed from 1985-1990 and it underestimates 10-year survival rates for patients who were diagnosed after 1991. For men diagnosed with prostate cancer after 1986, the 15-year CRS will continue to increase similar to the observed 10-year survival, but not at the same rate as predicted by the model, and the projected methods results may be overly optimistic.

By Christopher P. Evans, MD