Survival Estimation Methods

Appendices

In Figure L1 of this report, the most recent 5-year estimates of survival are for patients diagnosed in 1999. The estimates are slightly dated due to the lag time in cancer registry reporting of new cases and the time it takes to observe 5-year survival. Since complete follow-up is available only through 2003, the most recent estimates are based on data as follows:

Researchers at the National Cancer Institute (NCI) and elsewhere have been considering methods for extrapolating to obtain long-term survival estimates for cases diagnosed recently. Two such methods are the period method and the modeled method.

The period method, introduced by Brenner et al. (1) and slightly modified for use with Surveillance, Epidemiology, and End Results (SEER) data (2), considers only the patients' survival experience within the most recent calendar period (i.e., 2001-2003). For example, to estimate the 5-year survival in 2003, we use 0- to 1-year survival experience for cases diagnosed in (2001-2003), 1- to 2-year survival experience for cases diagnosed in (2000-2002), who survived at least 1 year, and so on up to 4- to 5- year survival experience for cases diagnosed in (1997-1999), who survived at least 4 years. The period 5-year survival is then calculated by multiplying these interval survival probabilities. The period method was not developed to provide survival trend but to give the most up-to-date estimate of survival experience observed in the data. The period method estimate is plotted as a filled diamond in the figure above.

The modeled method (3-4) consists of fitting a trend line across diagnosis years to each of the five observed interval survival probabilities (e.g., the 0- to 1-year survival for cases diagnosed in 1999 to 2003, 1- to 2-year survival for cases diagnosed in 1999 to 2002, etc.). These trend lines are then projected to the year of interest. The final estimate for a particular diagnosis year is obtained by multiplying the known and projected interval survival probabilities together.

For example, to estimate the 5-year survival rate for those diagnosed in 2003 using available data, known 0- to 1-year survival rates are combined with 1- to 2-, 2- to 3-, 3- to 4-, and 4- to 5-year projections from the model. The advantage of this method is that if survival is improving over time, the 4- to 5-year estimate of survival used for the 2003 computation will more accurately reflect the improved trend compared to the 1999 estimate used in the period method. The 5-year modeled survival estimates are plotted as open squares in the figure above. Because these projections combine known survival probabilities with projections, they are not a simple extrapolation of the last segment estimated using joinpoint regression.

These methods have the potential to provide clinicians, patients, cancer control analysts, and policy makers improved estimates of the long-term prognoses of recently diagnosed patients. As researchers continue to refine survival estimation methods and examine their assumptions and predictive ability, we ask for your input. Please click here to provide feedback.

References

1. Brenner H. Long-term survival rates of cancer patients achieved by the end of the 20th century: a period analysis. Lancet. 2002;360(9340):1131-5.
2. Cronin K, Mariotto A, Scoppa S, Green D, Clegg L. Differences Between Brenner et al. and NCI Methods for Calculating Period Survival. Technical Report # 2003-02, Statistical Research and Applications Branch, National Cancer Institute 2003 (http://srab.cancer.gov/reports).
3. Cronin K, Feuer E, Wesley M, Mariotto A, Scoppa S, Green D. Current Estimates for 5 and 10 Year Relative Survival. Technical Report # 2003-04, Statistical Research and Applications Branch, National Cancer Institute 2003 (http://srab.cancer.gov/reports).
4. Mariotto A, Wesley MN, Cronin KA, Johnson KA., Feuer EJ. Estimates of long-term survival for newly diagnosed cancer patients: a projection approach, Cancer, in Press.

Page last reviewed: December 12, 2007