Abstract
PL03-01
Epidemiology studies the occurrence, distribution, determinants, and control of diseases in populations. Epidemiology is fundamentally divisible into descriptive and analytic studies. Descriptive epidemiology identifies the distribution of disease, i.e., person, place, and time. Analytic epidemiology assesses the determinants of disease, i.e., why and how. Descriptive studies use the observational method and are externally valid (or generalizable) when population-based. Analytic studies use the scientific method and are internally valid when controlled. Chance, bias, and confounding threaten the validity of both descriptive and analytic epidemiologic studies. Much of the confounding in descriptive epidemiology is related to complex interactions between age, calendar-period, and birth-cohort (APC) effects or artifacts. These variables are “linearly dependent” through the relationship: age at diagnosis = year of diagnosis (calendar-period) - year of birth (birth-cohort). Age effects reflect cancer biology, i.e., early-onset versus late-onset disease. Calendar-period effects reflect time or secular trends, impacting all age groups at a given point in time, i.e., changing screening practice patterns, case ascertainment, diagnostic drift, etc. Birth-cohort effects reflect time trends, impacting all age groups in a given generation, i.e., changing exposure patterns (parity, body mass index (BMI), oral contraceptives, hormone replacement therapy, etc). The linear dependency of the APC framework produces a “non-identifiability” problem. That is, it is not possible to identify the independent effects for a 3rd variable when the other two and their difference are in the same linear model. Non-identifiability is especially problematic for breast cancer, which is characterized by complex age, period, and cohort interactions as illustrated with population-based data from the National Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) program. Breast cancer incidence and mortality rates with age-at-diagnosis on the independent scale (x-axis): Overall age-specific incidence rates increase rapidly until age 50 years, pause, and then continue to rise at a slower pace, implying that key carcinogenic events occur during early reproductive life. The pause in rates has been termed Clemmesen’s hook and has been attributed to menopause. Though overall age-specific incidence rates are similar for Blacks and Whites, age-specific rates are higher among Blacks than Whites before Clemmesen’s menopausal hook, and then rates cross, after which rates are higher among Whites than Blacks. In contrast, the Black-White ethnic crossover for mortality rates occurs after (rather than before) Clemmesen’s hook. Overall breast cancer risk is greater for nulliparous than parous women. Notwithstanding this overall effect, nulliparity compared to parity reduces breast cancer risk among women ages <35 years then rates cross, after which nulliparity increases risk among older women. Obesity has a similar crossing pattern. Obesity is protective for early-onset breast cancers and a risk factor for late-onset tumors. The reversal (or crossing) of age-specific incidence and mortality rates is a “qualitative” age interaction. If there was no age interaction, age-specific rates by race, parity, etc would be parallel, i.e., rate ratios would be constant for all age groups. Quantitative interactions have rate or slope changes in magnitude but not direction. Qualitative interactions have rate changes in magnitude and direction with the crossing (or reversal) of rate ratios. Parenthetically, though similar overall age-specific rate patterns suggest similar types of breast cancer for Blacks and Whites, Blacks appear to have more of the “worse” (early-onset) and less of the “best” (late-onset) type of breast cancer. Qualitative age interactions for parity suggest that risk factor profiles are different for early-onset versus late-onset types of breast cancer. Breast cancer incidence and mortality rates with calendar-period (or birth-cohort) on the independent scale (x-axis): For decades, overall incidence rate time trends have been greater for Whites than Blacks, but these overall trends vary by age at diagnosis and tumor characteristics. For example, among women ages 50+ years, incidence rates have been greater for Whites than Blacks for all recorded time periods. Among women ages 40-49 years, rates were greater for Whites than Blacks prior to 1980, and then rates merged. Among women ages <40 years, incidence rates have been greater for Blacks than Whites for all time periods. Additionally, incidence rates for ER positive tumors have been greater among Whites than Blacks for all age groups and all time periods. In contrast, incidence rates for ER negative tumors have been greater among Blacks than Whites for all age groups and all time periods. Prior to 1980, overall mortality rate time trends were greater for Whites than Blacks then rates crossed, after which rates were greater for Blacks than Whites and this disparity may be widening. This overall racial mortality trend varies by age at diagnosis. Among women ages 50+ years, Black-White rates crossed during the 1980s, similar to overall rates. Among women ages <50 years, mortality rates always have been greater for Blacks than Whites. Prior to 1980, overall incidence rates for high grade tumors were more common than rates for low grade tumors then rates crossed, after which low grade tumors were more common than high grade lesions. Among women ages 40+ years, the transition point (or crossover) occurred in earlier time periods with advancing age. Among women ages <40 years, incidence rates for high grade tumors were more common than rates for low grade tumors for all time periods. The overall effect of age and grade were similar for Whites and Blacks, though the transition point occurred in earlier time periods for Whites than Blacks. Similar to crossing age-specific rate patterns, the reversal (or crossing) of temporal trends is a qualitative age interaction, probably due to changing risk factor and/or screening practice patterns where screening mammography preferentially detected tumors of low malignant tumors among older women. Mammographic screening in the United States was initiated around 1980, and its usage increased rapidly thereafter. The earlier transition from high to low grade tumors for Whites than Blacks is consistent with the fact that Whites have larger fractions of the “best” breast cancer types (low grade and late-onset) and may be screened more than Blacks. Clemmesen’s hook and qualitative age interactions: Qualitative interactions suggest etiologic heterogeneity in an otherwise homogeneous cancer process. Of note, the reversal of rates consistently occurs near Clemmesen’s menopausal hook, i.e., approximately age 50 years. Indeed, emerging data suggest the Clemmesen’s phenomenon reflects the superimposition of two different rate curves according to age at onset and hormone dependence. The 1st breast cancer has a peak frequency near age 50 years and is hormone-dependent. The 2nd breast cancer has a peak frequency near age 70 years and is hormone-independent. In addition to these two age incidence patterns, descriptive data demonstrate bimodal age distributions for most histopathologic subtypes of breast cancer and even the so-called intrinsic breast cancer phenotypes, i.e., luminal A, luminal B, HER2+, and basal subtypes. A demonstration project using SEER’s Residual Tissue Repository appears to confirm bimodal incidence patterns according to molecular subtype. Bimodal age distribution_the occurrence of two incidence peaks or modes_further suggests etiologic heterogeneity due to at least two biological subtypes, causal pathways, or risk factor profiles. Although bimodal age incidence is acknowledged for cancers such as Hodgkin’s lymphoma, bimodality is not established for breast cancer. However, where the two modes for Hodgkin’s disease are widely separated in early and late adult life, the modes for breast cancer are obscured by a narrower age range of 50 to 70 years. In sum, population-based descriptive epidemiology of breast cancer suggests that different breast cancer populations--defined by race, tumor characteristics, molecular subtypes, etc--appear to reflect mixtures of two main breast cancer types (early-onset and high grade versus late-onset and low grade) with constant peak frequencies or modes near ages 50 and 70 years. Bimodal breast cancer is subdivided by Clemmesen’s menopausal hook and distinguished by qualitative age interactions or effect modifications. Parenthetically, bimodal breast cancer may be the population-based reflection of “Knudson’s” two-hit hypothesis, separating early-onset and late-onset cancers in the general population.
First AACR International Conference on the Science of Cancer Health Disparities-- Nov 27-30, 2007; Atlanta, GA