‘Age Selection of Mortality from Tuberculosis': A re-analysis of Frost on the 70th anniversary of publication

Introduction: W.H. Frost's 1939 publication of strong cohort effects in tuberculosis mortality was revolutionary. Frost relied on graphs to illustrate effects; statistical approaches to age-period-cohort effect estimation remain desirable yet elusive, and have been a source of debate for over fifty years. This is due partially to the co-linearity among age, period, and cohort and partially to the epidemiological misconception that cohort effects can be estimated independently of age and period effects. Based on Greenberg's conceptualization of cohort effects as a partial interaction between age and period, we used a novel approach to re-analyze the Frost tuberculosis data and quantify the cohort effects.

Methods: A three-phased method was used; first, classical graphical presentation and inspection; second, median polish to remove additive components of age and period effects; third, linear regression of residuals from the median polish to quantify the cohort effect. Frost data was extracted from the contingency table in the 1939 publication.

Results: We found significant interaction of age and period effects, indicating the presence of cohort effects. Compared to a reference cohort (1851-1860), those born 1911-1920 had 0.77 (95% C.I. 0.61-0.97) times the risk of tuberculosis mortality, and those born 1921-1930 had 0.54 (95% C.I. 0.39-0.73) times the risk.

Discussion: The median polish is a novel approach to quantifying cohort effects in contingency table data. Re-Analysis of the Frost data confirms a strong cohort, whereby later born cohorts exhibit a lower risk of tuberculosis mortality compared to earlier-born cohorts.