The heterogeneity of human populations is a common consideration in describing and validating their various age-related features. Heterogeneity, in particular, amongst other factors, is used to explain the variability of mortality rates across the lifespan and deviations from an exponential growth at young and very old ages. A mathematical model that combines the population heterogeneity with the assumption that the mortality of each constituent subpopulation increases exponentially with age, has recently been shown to successfully reproduce the entire mortality pattern across the lifespan as well as its evolution over time. Furthermore, the analysis of time-evolution of the mortality pattern, performed by fitting the model to actual data of consecutive periods, confirms the applicability of the compensation law of mortality to each subpopulation and concludes on the evolution of the population towards homogenisation.
In this work we aim to show that the heterogeneity of human populations is not only a convenient consideration for fitting mortality data but is indeed the actual structure of the population as reflected by the dynamics of its mortality over age and time. In particular, we demonstrate that the model of heterogeneous populations fits mortality data better than most of the other models if the data are taken for the entire lifespan and better than all other models if we consider only old ages. Also, we show that the model can reproduce seemingly contradicting observations in late-life mortality dynamics namely deceleration, levelling-off and mortality decline. Assuming that heterogeneity is reflected in genetic variations within the population, using Swedish mortality data for 20th century we show that the homogenisation of the population, observed in the model fits, can be associated with the evolution of allele frequencies.