Birth Cohort Analysis

A birth cohort is a group of individuals born during a given calendar time period within a specified geographical region. For example, the U.S. 1950 annual birth cohort refers to the group of people born in the United States during the calendar year 1950, and the 1950 to 1952 birth cohort identifies those born during the period covering the three consecutive calendar years 1950, 1951, and 1952. Birth cohort analysis is an observational cohort analysis (as opposed to an experimental cohort analysis such as clinical trials) of an entire birth cohort or of a selected sample of the birth cohort.

Data for birth cohort analysis are best depicted by a Lexis diagram with the horizontal axis denoting calendar time and the vertical axis denoting age wherein each individual live birth is represented by a lifeline with 458 inclination from the horizontal axis. Starting at birth on the horizontal axis, each individual is followed continuously as he or she passes through different ages during all or part of the life span, that is, until death or censoring (such as out-migration), whichever occurs first. Events of interest occurring to an individual during his or her lifetime are marked on his or her lifeline. The events of interest typically include exposure to risk factors (such as start of smoking) and occurrence of health outcomes (such as incidence or recurrence of a disease or medical condition) as well as vital events (such as giving birth, change of marital status, and death). Each marked lifeline represents the complete life history of an individual and constitutes a sample path. So the life histories for all individuals (all marked lifelines) in the birth cohort constitute the sample space.

Since we start at the beginning of life at which time one is susceptible to almost all risk factors, events, and health outcomes, almost any of them can be studied by birth cohort analysis. In particular, the recorded birth cohort data as described above allow calculation of both cumulative incidence and incidence density and so can be used to perform the following cohort analyses retrospectively.

Event History Analysis Using Survival Time Data. When sample size is not too large, this can be done by applying likelihood methods and martingale theory to produce statistical inference (maximum likelihood or martingale estimates and hypothesis testing comparing different birth cohorts) of disease incidence rates and of effects of risk factors on disease incidence.

Construction of Cohort Life Tables Using Age Group Data (for Large Populations). Such tables include attrition life tables and the most general incrementdecrement life tables to obtain transition probabilities from one state to another and to obtain an estimate of the expected duration of stay in each state. In addition to comparing two different cohort life tables, one can also compare a cohort life table with a period life table, both constructed on the same base period, to see if there exists any period-cohort effect or birth cohort effect.

Comparing the Health Outcome or Vital Event of Interest (Say, Mortality) of the Birth Cohort With That of the Corresponding General Population to See if the Special Life Event Experienced by the Birth Cohort Has Caused a Significant Change in Mortality Level. This may be done by calculating the standardized mortality ratio (SMR) and testing the null hypothesis SMR = 1 using the asymptotic unit normal test statistic Z = ln SMR= ffiffiffiffiDp, where SMR= D= E and D and E are the observed deaths and expected deaths (based on the mortality of the corresponding general population), respectively, in the cohort.

...