How to Measure Cancer Survival

Cancer survival can be measured in three different ways:

• Observed all cause survival – This provides an estimate of the probability that patients will survive all causes of death.
• Net cancer-specific survival (statistic based on policy) – This provides an estimate of the probability of surviving cancer when other causes of death are not present. This measure is not affected by changes in mortality resulting from other causes. As such, it is very useful when it comes to tracking survival across time or to carry out comparisons between registries or between racial/ethnic groups.
• Crude probability of death (a measure of patient prognosis) – This provides an estimate of the probability of dying from cancer when other causes of death are present. This is more effective in evaluating the impact of cancer diagnosis when measured at an individual level because mortality resulting from other causes plays an important role. It records mortality patterns actually occurring in a cohort of cancer patients when other potential causes of death are working simultaneously. The crude probability of death is generally used to describe a cumulative probability of death resulting from cancer rather than survival.

Methods used to Measure Cancer-Specific Survival

There are two methods for estimating net cancer-specific survival and Crude Probability of Death – using information related to cause of death or expected survival tables. The use of cause of death information has attracted a lot of debate as to what is the right endpoint. In case the death certification was perfect, it would be appropriate to simply use the specific type of cancer as endpoint. But, in case the cancer metastasizes, there are occasions when the death certificate wrongly lists the metastatic site as the primary cause of death. In such cases, it is best to consider all types of cancers as the end point. This is especially true when the patient may have only one type of cancer. Work is currently underway to devise more sophisticated algorithms that can define endpoints on the basis of common sites of metastases for each specific cancer.

Irrespective of whether an approach based on the cause of death or the expected life tables is used, one must always consider the exclusions from the analysis. Boer et al published a technical report that describes several different approaches to exclusions used for survival analysis, and also the choice of endpoints while death certificate information is being used. The figure given on the right describes the survival statistics resulting from using a combination of the two measures and the two estimation methods. Below is a description of each one of them.

• Relative survival: Survival life tables are used to calculate cancer survival when other causes of death are not present. Relative survival describes the ratio of the proportion of observed survivors (includes all causes of death) in a cohort of cancer patients to the proportion of potential survivors, as applicable to a similar cohort of cancer-free individuals. This ratio is based on the supposition of independent competing causes of death. Since it is difficult to obtain a cohort of cancer-free individuals, expected life tables are used and it is assumed that deaths caused by cancer form only a negligible proportion of the overall deaths.
• Cause-specific survival: This is a net survival measure that represents survival of a particular cause of death when other causes of death are not present. Estimates are derived by citing the cause of death. Deaths resulting from causes apart from those mentioned are deemed to be censored.
• Crude Probability of Death obtained using Expected Survival: This crude measure utilizes expected survival (taken from the expected life tables) in order to estimate the probability of dying caused by other factors in each interval. Since it is difficult to obtain a cohort of cancer-free individuals, expected life tables are used and it is assumed that deaths caused by cancer form only a negligible proportion of the overall deaths.
• Crude Probability of Death obtained using Cause of Death Information: The probability of dying of cancer and from other causes, relevant to a cohort of cancer patients is derived using cause of death information.