Effect of Age, Birth Cohort, and Period of Death on Cerebrovascular Mortality in Spain, 1952 Through 1991
Background and Purpose The continued decrease in cerebrovascular disease in Spain remains unexplained. Age-period-cohort analysis enables description of birth cohort and period-of-death components. This study sought to describe these effects on the decline of stroke mortality in Spain.
Methods Deaths due to cerebrovascular diseases in the period from 1952 through 1991 and the corresponding population figures were grouped into 11 age groups and 8 5-year periods, from which age-specific mortality rates for 18 birth cohorts were then computed. These were plotted for graphical presentation purposes and fitted to Poisson regression models to assess age, period, and cohort effects.
Results An exponential age effect was present for both sexes regardless of cohort or period. A definite downward period effect was observable from 1962 to 1991, except for a sharp fall and peak in the periods 1967 to 1971 and 1972 to 1976, respectively, which was possibly ascribable to changes in diagnostic and coding practices. Age- and period-adjusted stroke mortality increased for earlier cohorts and decreased for generations born between 1892 and 1940. For post-1940 generations, there was an increasing risk of stroke mortality.
Conclusions The results suggest that a decrease in incidence coupled with an increase in survival may account for the observed decline in stroke mortality, but further studies on the Spanish population are needed to assess these findings. Although not yet definitive, there are signs of an increase in incidence among the more recent generations. If the decreasing period effect fails to offset this increase, future years may see a deceleration in the current decline in stroke mortality.
In Spain, cerebrovascular diseases constitute the first and second leading causes of death in women and men, respectively.1 These diseases mainly affect the elderly and can prove to be a source of neurological sequelae and disability. Accordingly, as population and longevity increase, cerebrovascular diseases become a major public health problem.
Time-trend analysis of stroke mortality in Spain reveals that age-adjusted rates have decreased steadily since the beginning of the century and more sharply still since 1975.2 The reason for this decline remains unknown, a possible explanation being that case fatality and survival have increased because of major improvements in treatment of the acute phase of the disease. On the other hand, the decline in mortality could also be attributable to a decline in incidence. Such a decrease in incidence could be due to antihypertensive treatment, changes in lifestyle, and improvements in socioeconomic status.3
Pinpointing the distinction between the respective contributions of survival and incidence to mortality trends is of interest from a public health point of view: if incidence is not decreasing yet survival is increasing, the result will be increased prevalence and thus a growing number of disabled old persons requiring both medical and social support. Unfortunately, however, there is a dearth of information on stroke incidence and survival in Spain.
Traditional analyses of mortality time trends by year of death, even when stratified by age, fail to describe differences in mortality between generations. In contrast, age-period-cohort analysis of mortality time trends measures the component explained by birth cohort and risk of dying vis-à-vis circumstances prevailing at the time of death, ie, the period effect. Cohort effect relates to lifelong exposure to disease risk factors shared by a whole generation, and period effect relates to factors that simultaneously affect the whole population, such as the introduction of widely available medical care and changes in diagnostic or cause-of-death coding practices.
This study sought to describe time trends in cerebrovascular mortality and assess age-period-cohort effects on stroke mortality in Spain to explain the decline in mortality and provide a basis for future projections.
Age- and sex-specific cerebrovascular deaths (International Classification of Diseases [ICD], 6th and 7th revisions, codes 330 to 334 for pre-1968 deaths; 8th and 9th revisions, codes 430 to 438 for post-1968 deaths) were taken from the official vital statistics published by the Spanish National Statistics Office for the years 1952 to 1991. Deaths were grouped into eight 5-year periods from 1952-1956 to 1987-1991 and into eleven 5-year age groups ranging from ages 35 to 39 years up to 85 years and over. Year of birth was deduced from year of death and age at death. The 5-year age and period grouping yielded 18 partially overlapping 10-year birth cohorts. Cohorts were defined by their respective mid-years, starting with 1867.
Parallel population groups were constructed using mid-year official population estimates, and age-specific death rates for each birth cohort were computed and plotted for the purposes of graphical presentation.
The above-mentioned matrices, namely age- and sex-specific rates, were used to estimate the effects of age, birth cohort, and period of death on stroke mortality. Log-linear Poisson models were fitted using standard GLIM-package procedures, along with specific program macros to correct for extra-Poisson variation. The correction model used was based on a generalization of Breslow's4 procedure II for Poisson models. This was equivalent to fitting a negative binomial model.
The predictive variables (age, cohort, and period) were sequentially included in the model. Age was considered first because stroke mortality increases remarkably with age. Thereafter, the “age+drift” model, which assumes a constant change in the rate logarithm between adjacent cohorts or periods, was fitted. In the next step, two-factor “age+period” and “age+cohort” models were considered. Finally, the model came to include all three factors (age, period, and cohort) and—as proposed by Osmond and Gardner5 and Decarli and La Vecchia6 —from among the infinite possible solutions, the one selected was that which minimized the sum of the Euclidean distances from the two-factor models.
The three independent variables considered are arithmetically interrelated (age at death=year of death−year of birth), thus rendering it impossible for their relative effects on the general trend to be separately identified (the so-called “identifiability problem”).7 Because Osmond and Gardner's solution to nonidentifiability implies an arbitrary constraint, an additional analysis was run on the deviation of each output value from its general linear component, thereby allowing variations in time to be interpreted regardless of the general trend of the effect.7 Results are presented under three different assumptions: Osmond and Gardner's solution, which distributes net drift between cohort- and period-effect estimates; and the two extreme situations, whereby trend is attributed wholly to cohort or wholly to period effect. Irrespective of the solution adopted, the shapes of the curves do not change.
Output values for age effect are interpretable as age-specific death rates per million person-years, adjusted for period and cohort. Owing to such adjustment, these estimates may diverge from the real rates. Cohort and period effects were averaged to unity using a linear scale; hence, these values can be interpreted as rate ratios or relative risks, with their weighted mean being taken as reference.
From 1952 through 1991, age-adjusted death rates for cerebrovascular diseases decreased from 138 to 89 per 100 000, a decline of 36% in men and 37% in women. This decline was more evident from 1975 onward and occurred across all age groups (Table 1⇓).
Figs 1⇓ and 2⇓ depict age-specific rates on a logarithmic scale with respect to the generation's mid-year of birth for males and females, respectively. Rates for all age groups have been joined by lines, as described in the figure keys. This presentation enables graphical assessment of the effects of age, period, and cohort on mortality rates. It should be noted that because 1952 is the initial date of the study, rates for the earliest cohorts are available for only the oldest age groups. Similarly, by year-end of 1991, recent cohorts had not attained ages older than 44 years. Almost complete follow-up was feasible for central cohorts only.
As envisaged, mortality increased markedly with age in all cohorts for men and women alike, with rates being somewhat higher in men, especially in the younger generations. For any given age group, mortality was observed to vary with cohort, increasing in the older cohorts and decreasing in the more recent ones. The negative slope was less evident for men, the younger age groups, and the more recent cohorts.
A period effect was also to be observed, since mortality peaked for all age groups in consecutive cohorts, meaning that the downward shift occurred in the same period for all birth cohorts and age groups. From age and year of birth, it can be deduced that this period must have been circa 1974 (1972-1976). Regression analysis confirmed the results of the graphical approach. Table 2⇓ shows that a significant reduction in deviance (a gauge of the efficacy of models describing stroke mortality) was achieved when shifting from the age model to models including age+period or age+cohort. This indicates that both cohort and period effects should be taken into account, even though the cohort model evinced a better fit. Similar results were obtained for men and women. The final three-factor model was used to estimate all three effects.
Estimates for age-, period-, and cohort-specific effects on male and female stroke mortality are set out in Fig 3⇓. Age- and period-adjusted stroke mortality increased for earlier cohorts and decreased for generations born in the period from 1892 to 1940. This downturn in the cohort curve was clearly observable, irrespective of which solution was adopted regarding identifiability. Warranting special interest was the fact that the downward trend not only leveled off but was even reversed in the youngest generations, especially in men. This finding must be interpreted carefully, however, since cohort values for the most recent periods are based on fewer age-specific rates and fewer deaths and are therefore less reliable than central-cohort values. Nevertheless, this trend was consistently observed for the most recent consecutive cohorts (five male and three female).
Independently of age and cohort, a definite downward period effect was observable from 1962 to 1966 onward. The decline was evident even when the general downward trend was wholly attributed to cohort (upper line of the period-effect graph). The respective fall and peak in the 1967-1971 and 1972-1976 periods may reflect changes in diagnostic and coding practices in those years, ie, they coincide in time with the implementation in Spain of the 8th revision of the ICD and the introduction of CT.
Lastly, an exponential age effect was present for both sexes, regardless of cohort or period.
Although this analysis does not allow for distinction between the relative importance of cohort versus period effect on the decline of stroke mortality, our results suggest that both effects are present in the decline of stroke mortality in Spain.
Dating from the beginning of the century, successive generations in Spain evince a decreasing risk of dying of stroke, irrespective of age and year of death, something which can be interpreted as a lowering in the probability of exposure to the risk factors for cerebrovascular mortality. Similar cohort effects have been described in Italy, England and Wales, Australia, Canada, and the United States.8 9 10 11 12 There are no data on stroke incidence trends in Spain, yet incidence decreased in other developed countries until the 1980s.13 14 15 16 From 1960 onward, industrial development in Spain brought with it massive rural-urban migration and a sharp socioeconomic rise, which could go some way to explain a subsequent decline in cerebrovascular risk. However, major changes in lifestyle were also introduced, including an increase in caloric and lipid intake, smoking, a shift in alcohol consumption patterns, and sedentariness.17 Antihypertensive treatments were not widely used in Spain until 1980 and thus cannot be related to any reduction in stroke risk in most of the cohorts considered in this study.2 Although the latter are believed to be the main cause of the decline in mortality observed in recent years, other factors aside from hypertension control must be involved in the epidemiology of stroke and must be affecting mortality.
Regardless of sex, cohort, or age, the period of 1962 through 1991 registered a continuous decline in stroke mortality, and this held constant even when the general trend was assigned wholly to the cohort effect. Such a declining period effect must mean that some positive influence simultaneously affected and subsequently continued to affect the entire Spanish population, thereby reducing the risk of death from cerebrovascular disease. This is consistent with the decrease in case-fatality rates that has been described almost worldwide.18 19 The general availability of qualified medical attention in Spain from 1965 onward may well explain a substantial improvement in case fatality and a subsequent declining period effect.20
Mortality trends analyzed by period of death revealed a fall in the 1967-1971 period and a subsequent peak in the 1972-1976 period. The former can be linked to the implementation in 1968 of the ICD 8th revision, when new cerebrovascular codes were introduced and others suppressed, whereas the peak could be associated with the increase in diagnostic sensitivity brought on by the introduction of CT in Spain in 1974.2 20 In all events, the impact of CT on mortality is imprecise, since it may have resulted in better case ascertainment or, on the other hand, enhanced diagnostic specificity and a drop in nonvascular deaths that would have otherwise been attributed to stroke.21
The most striking result of this study is the reversal seen in the decline of cohort effect. This means that post-1940 generations (those now aged 55 years or under) face an increasing risk of stroke mortality regardless of age or sex. This result should not be taken as definite, since mortality information for these cohorts is still incomplete. Moreover, since the subjects concerned have not yet reached the typical ages prone to ischemic stroke (the most frequent cause of cerebrovascular death), the increase may be reflecting an age-cohort interaction due to the differential in survival between hemorrhagic and ischemic stroke. Nevertheless, the increase is preceded by a flattening of the curve among the older cohorts. Moreover, similar findings have been reported by Wolfe and Burney9 in England and Wales for post-1920 cohorts and by D'Avanzo et al8 in Italy for post-1940 cohorts. There is evidence that from the mid 1980s, stroke incidence, particularly among more recent generations, has ceased to decrease and/or has even been increasing in other western European countries, the United States, and New Zealand.13 14 15 22 23 24 25 Possible reasons for this rise in risk lie in increased survival in diabetes and heart disease, the drift toward a more atherogenic diet, or the high prevalence of smoking in the cohorts in question. If the trend holds and the decline in the period effect proves inadequate to counteract it, cerebrovascular mortality may increase in years to come.
To sum up, both period and cohort effects are present in the decline of cerebrovascular mortality in Spain during the period of 1952 through 1991. This can be interpreted to mean that the observed decline in stroke mortality may lie in a decrease in incidence coupled with an increase in survival. Although not yet definitive, there are signs of a probable increase in incidence among the more recent generations. If survival continues to increase, the prevalence of older persons with cerebrovascular disease will rise. Moreover, if the decreasing case fatality fails to offset this increase in incidence, future years may see a deceleration in the current decline in stroke mortality. Consequently, prevention must still be regarded as a priority.
Age-period-cohort analyses are no substitute for the lack of population-based information. The results of this study emphasize the need to undertake more in-depth studies on stroke epidemiology in Spain.
This work was partially supported by a grant from Spain's Fondo de Investigaciones Sanitarias (FIS–Health Research Fund). We are grateful to Margarita García of the National Statistics Office for providing mortality data, Michael Benedict for the English language review of the manuscript, and Gracia Romasanta for secretarial assistance.
- Received July 29, 1996.
- Revision received October 10, 1996.
- Accepted October 14, 1996.
- Copyright © 1997 by American Heart Association
Martínez Aragón MV, Llácer A. Mortalidad en España 1991. Bol Epidemiológico Semanal. 1995;3:33-38.
Decarli A, La Vecchia C. Age, period and cohort models: review of knowledge and implementation in GLIM. Rev Stat App. 1987;20:397-409.
Wolfe C, Burney P. Is stroke mortality on the decline in England? Am J Epidemiol. 1992;136:558-565.
Al-Roomi K, Dobson A, Hall E, Heller R, Magnus P. Declining mortality from ischemic heart disease and cerebrovascular disease in Australia. Am J Epidemiol. 1989;129:503-510.
Wolf PA, D'Agostino RB, O'Neal MA, Sytkowski P, Kase CS, Belanger AJ, Kannel WB. Secular trends in stroke incidence and mortality: the Framingham study. Stroke. 1992;23:1551-1555.
Feigin VL, Wiebers DO, Whisnant JP, O'Fallon WM. Stroke incidence and 30-day case-fatality rates in Novosibirsk, Russia, 1982 through 1992. Stroke. 1995;26:924-929.
Tuomilehto J, Bonita R, Stewart A, Nissinen A, Salonen JT. Hypertension, cigarette smoking, and the decline in stroke incidence in eastern Finland. Stroke. 1991;22:7-11.
Banegas JR, Rodríguez Artalejo F, Graciani A, Hernández Vecino R, Rey Calero J. Trends in ischemic heart disease mortality and its determinants in Spain, 1940-1988. Eur J Public Health. 1995;5:1-6.
Modam B, Wagener DK. Some epidemiological aspects of stroke: mortality/morbidity trends, age, sex, race, socioeconomic status. Stroke. 1992;23:1230-1236.
Lázaro P. Evaluación de los Servicios Sanitarios: la Alta Tecnología Médica en España. Madrid, Spain: Fondo de Investigación Sanitaria; 1990.
Sandercock P, Molyneux A, Warlow C. Value of computed tomography in patients with stroke: the Oxfordshire Community Stroke Project. Br Med J. 1985;290:193-197.
Jorgensen HS, Plesner AM, Hubbe P, Larsen K. Marked increase of stroke incidence in men between 1972 and 1990 in Frederiksberg, Denmark. Stroke. 1992;23:1701-1704.
Harmsen P, Tsipogiani A, Wilhelmsen L. Stroke incidence rates were unchanged, while fatality rates declined, during 1971-1987 in Goteborg, Sweden. Stroke. 1992;23:1410-1415.
Lindenstrom E, Boysen G, Nyboe J, Appleyard M. Stroke incidence in Copenhagen, 1976-1988. Stroke. 1992;23:28-32.