Fitting poisson models sequentially to explore cancer mortality trends in Argentina.

NICLIS CAMILA; POU SONIA; OSELLA ALBERTO; DÍAZ MARÍA DEL PILAR

Florianópolis

Conferencia; XXVth International Biometric Conference; 2010

International Biometric Society

Age-period-cohort models are usually used in descriptive epidemiology to analyze time trends in mortality. Unlike the classical analyses, this approach provides three dimensions or components of time, such as age-group, period of death and cohort of birth, in order to explore the trends. However, the exact linear dependence between these three effects produces non-identifiable functions for estimation procedures. In this work, Poisson models fitted sequentially were used to deal with this problem. First, an age-cohort model was fitted. After that, a period-only model, using the log-fitted values from previous age-cohort model as offset, was fitted sequentially. Following biological plausibility criteria, age was defined as the major time scale, cohort of birth as the secondary time scale and period of death as the residual time scale, conditional on the estimated age and cohort effects. Mortality rates corresponding to high incidence cancers in Argentina from 1986 to 2006 were analyzed. Cancer mortality data were tabulated in four periods of death (1986-1990, 1991-1995, 1996-2000 or 2001-2006), and five-years age groups. Cohorts of birth were identified by the central year of birth, resulting in ten-years cohorts overlapping each other by five-years. Models were fitted by using STATA v10. The estimates obtained by this sequential procedure showed a strong age effect, with an exponential rise in mortality rates in both sexes in all cancers considered. Instead, the other two effects differed by cancer site. Period effect reflected increasing and decreasing tendencies for men and women, respectively, in colorectal cancer. Breast cancer mortality risk showed a slight increase around 1998. No period effect was observed for bladder and prostate cancer. The cohort effect showed differences by gender (increasing and decreasing trends for male and female, respectively) in colorectal cancer mortality. Though, in bladder cancer this effect exhibited a similar pattern for both sexes characterized by a decreased risk of death, particularly in post-1931 generations. A decrease since 1946 cohort of birth in prostate cancer was also registered, whereas in breast cancer a significantly lower risk of dying was observed in women born in cohorts centered in 1956 and 1966. Fitting Poisson models as used here avoid confusion about the mechanics of the age-period-cohort model and allowed to draw valuables interpretations in the field of cancer epidemiology in our region.