Population-based Cancer Survival Statistics Overview
Cancer survival is the proportion of patients alive at some point subsequent to the diagnosis of their cancer, or from some point post-diagnosis (conditional survival). It is represented as the probability of a group of patients "surviving" a specified length of time (e.g. 3 years, 5 years, 20 years).
There are several Measures of Cancer Survival using different end points for survival depending on the question of interest. For example:
- Some may be interested in death from any cause, while others may be interested in just death from cancer.
- Policy makers and others may be interested in death from cancer where the confounding effects of death from other causes are removed (e.g. when comparing survival from cancer for different racial/ethnic groups with very different other cause mortality).
- Individuals may be interested in the both the probability of death from cancer and the probability of death from other causes, each estimated in the presence of the other.
There are various sources to obtain already-derived estimates of survival using SEER data (see Available Survival Statistics), as well as various software packages which allow one to estimate population-based survival (SEER*Stat), as well as to model differences between survival curves (CanSurv).
Important Facts About Survival
- Unlike incidence or mortality statistics where the total population is included in the denominator, only diagnosed patients are included in the survival calculations. In the past there has been some confusion when people use the term mortality to mean (1-survival). This is misleading, since mortality statistics include the entire population at risk, whereas survival (and 1-survival) only include diagnosed patients at risk. We use the term Cumulative Probability of Death for (1-survival).
- One problem inherent in estimating long- term survival is that only cohorts diagnosed a long time ago have enough follow-up to directly estimate these quantities. Thus direct estimates of long term survival may not be very relevant for newly diagnosed patients, especially in cancer sites where there have been dramatic improvements in survival. The section on Cohort Definition describes various approaches to defining which years of diagnosis are included in order to provide more up-to-date estimates of survival.
- Population-based survival derived from cancer registries differs in several important ways from survival derived in clinical trial settings. In a clinical trial there is a detailed review of the medical record to ascertain the cause of death, whereas in population-based registry settings one must depend on death certificates which have inherent inaccuracies. Approaches to Estimation describes various approaches to using death certificate cause of death as the endpoint. Another approach circumvents the problems inherent in using death certificate cause of death, if one can assume that the general population dies of causes other than cancer at the same rate as the cancer population. If this independent competing risk assumption is met, then one can use population life tables to statistically factor out the probability of death from cancer and other causes. Potential problems with using life tables include lack of availability for small geographic areas, certain racial/ethnic groups, and up-to-date tables.