Determining the Appropriate Cancer Survival Measure for a Specific Analysis

What is the difference between net survival and crude probability of death?

Cause-specific survival can be measured in the absence of competing causes of death (net measure) or the presence of competing causes of death (crude measure).

Net survival is a measure that is not influenced by changes in mortality from other causes and, therefore, provides a useful measure for cancer control over time. The leveling off of a net survival curve represents a biological cure from the specific cancer. This means that no additional deaths due to a specific cancer would be expected no matter how long a patient lives. The value at which net survival levels off is an estimate of the percent of the cancer cohort considered to be cured.  Statistically, this is the point where the level of risk for a cohort of cancer patients returns to the level of the general (cancer-free) population.

Crude probability of death provides a measure of the mortality actually experienced by a cohort of cancer patients. A crude measure of cause specific mortality estimates the risk of death from cancer when all causes of death are possible. This is reported as a probability of death rather than survival since the complement (1-crude cause specific probability of death) contains both the probability of surviving an interval and the probability of dying of other causes during the interval. The leveling off of the cumulative crude cause specific probability of death represents a personal cure for patients in a cancer cohort. A person cure is the point at which a patient is no longer at risk from dying of their cancer either because they have experienced a biological cure or they have died of other causes.

How do I choose between using cause of death information and expected survival?

The method that is preferred will depend on the analysis being performed. When choosing between the two options, the following limitations of each should be considered.

Cause of death information is based on the cause of death listed on the death certificates. The main limitations are listed below.

• Unknown/missing cause of death. The user defines how to handle unknown/missing cause of death in the Definition of Cause of Death. The user can define all unknown/missing cause of death as cancer or define as other causes, representing extremes in either direction with the correct assignment between the two extremes.

• Unreliable cause of death information. Cancers that have metastasized from the original site may be mis-coded on the death certificate. When using cause of death information one alternative is to include only individuals with a single primary cancer and consider deaths coded as cancer from any site as a cancer-specific death in the analysis. However this solution may introduce bias by excluding cases with multiple primary cancers.

Expected survival for a cohort of cancer patients is based on U.S. life tables. Expected survival for an individual cancer cases is matched to life tables on age, race, and sex. The main limitations are listed below.

• Violation of the assumption of independent competing causes of death. When this assumption is violated, the expected survival estimated from U.S. life tables may not be appropriate. One example of when this assumption would be violated is a situation were there is a risk factor (such as smoking) that affects survival from both the cancer (e.g lung cancer) and other causes (e.g. heart disease).

• Individuals with missing age, race, or sex data are excluded from the analysis. This exclusion may lead to bias in the survival calculation.

• Calculation of expected survival may be unstable for smaller subgroups in the population, such as specific race categories.

Why do we use survival as a net measure and probability of death as a crude measure?

Since the net measure is in the absence of other causes there are only two possible outcomes in a time interval. An individual can die of cancer or survive the interval. Therefore net survival has the clear interpretation of surviving the time interval.

When multiple causes of death are considered simultaneously, there are three possible outcomes in an interval. An individual can die of cancer, die of other causes, or survive the interval. Therefore, the probability of surviving cancer is equal to the probability of surviving all causes plus the probability of dying of other causes during the interval. Because of this awkward interpretation of crude survival, probability of death is estimated instead of survival.