BACKGROUNDEstimates of mortality at older ages, even above age 70, are a major concern for demographers and have important applications and consequences in other areas such as actuarial science and economics. In general, mortality estimates at older ages are limited by small numbers both in the exposure and events OBJECTIVEIn this paper, we propose a mixture-based model for mortality modeling for the elderly (+70 years). METHODSThe proposed model is compared with 4 other widely studied and used models: the Beard, Gompertz, Makeham, and Perks models. We apply our proposed method to two populations of different data quality: Brazil and Japan. RESULTSThe mixture-based model captures the decrease in mortality force at older ages, which is a characteristic observed in several populations. CONCLUSIONSIn the comparative study for the Japanese population, our model presented a better fit to the data, obtaining an absolute mean percentage error of less than 7%, while the other models presented values greater than 30%.

We propose a novel method to detect deceleration in mortality patterns. For a gamma-Gompertz frailty model, we suggest maximizing a penalized likelihood in a Bayesian setting as an alternative to traditional likelihood inference and hypothesis testing. We compare the performance of the two methods on simulated and real mortality data.