Measures of Cancer Prevalence
There are several measures of cancer prevalence described below.
Limited-Duration Prevalence represents the proportion of people alive on a certain day who had a diagnosis of the disease within the past x years (e.g. x = 5, 10 or 20 years). Registries of shorter duration, less than 40 or 50 years of data collection, can only estimate limited-duration prevalence.
Limited-duration prevalence can be further classified into periods from year of diagnosis. Thus, the 20-years prevalence can be further classified into the prevalence of those diagnosed in the last 0 to < 5 years, 5 to < 10 years, 10 to < 15 years, and 15 to < 20 years.
NCI's SEER Program has information on cancer cases since 1975, thus a maximum of 42-year prevalence can be estimated from SEER cases diagnosed from 1975 through 2016.
Limited-duration prevalence statistics can be calculated using the SEER*Stat software.
Complete Prevalence represents the proportion of people alive on a certain day who were diagnosed with the disease, regardless of how long ago the diagnosis was made. Complete prevalence can be estimated from self-reported population-based surveys (Byrne et al., 1992), although one must be concerned with underreporting and misclassification of disease. Direct computation (the counting method) of complete cancer prevalence requires registry data that has been collected over a sufficiently long period of time to capture all prevalent cases of the disease.
In the United States, the only registry with sufficient incidence and follow-up data to approximate complete cancer prevalence is the Connecticut Tumor Registry (Connelly et al., 1968; Gershman et al., 1976). The Connecticut registry has information on cancer cases from as early as 1935, although computations do not usually include cases diagnosed prior to 1940. However, projecting estimates of US prevalence from Connecticut is not optimal because Connecticut is less representative of the United States than SEER. Moreover, from SEER, prevalence can be estimated by race and ethnicity. The completeness index (Capocaccia & De Angelis, 1997; Merrill et al., 2000), a statistical model to estimate complete prevalence from limited-duration prevalence, has been derived and used to estimate complete prevalence from SEER limited-duration prevalence, survival, and incidence data (see Chapter 29 of the SEER Cancer Statistics Review, 1973-1999 for a description of the methods).
Complete prevalence statistics can be calculated using the ComPrev software.
Prevalence estimates calculated from cancer registry data consider all patients previously diagnosed with cancer and alive at the prevalence date irrespective of whether the patient is under treatment or is considered cured. This use of the term prevalence may be justified because treatment for the disease (e.g., surgery or radiation) may lead to long-term or permanent mental and physical impairment, as well as changes in one's socioeconomic and cultural status. However, the definition may also be used simply because of the difficulty of determining when a person is cured or when, using population-based data, treatment ends.
Care Prevalence is an estimate of prevalent cases that are still under care. Since population-based cancer data does not include follow-up information on cancer care, estimation of care prevalence is problematic. The SEER-Medicare linked data allow for longitudinal tracing of individuals with cancer using information from the Medicare claims. Mariotto et al., (2003) have estimated the prevalence of patients with colorectal cancer age 65 and older who are under care in the US.
Non-cure Prevalence is an estimate of prevalent cases that have not been cured of disease. Statistical approaches (e.g. assuming survival models which are a mixture of cured and uncured patients) have been applied to model cure prevalence. (See Capocaccia & De Angelis, 1997; Coldman et al., 1992)