Background and Purpose Variations in stroke mortality could be explained by changes in factors that act around the time of death (period effect) and by risk factors that are present in early life (cohort effect). The aim of this study was to analyze mortality rates for stroke in Sweden during the period 1969 through 1993 and to predict mortality trends until the year 2003, taking into account age, cohort, and period effects.
Methods Age-period-cohort models were used to analyze stroke mortality in Sweden between 1969 and 1993 and to predict age-specific death rates and total number of deaths for the periods 1994 through 1998 and 1999 through 2003.
Results Mortality rates in the age group 25 to 89 years decreased from 203 to 143 per 100 000 for men and from 185 to 113 per 100 000 for women over the study period (average annual decrease of 1.3% for men and 1.9% for women). The decline was present in all age groups. The full age-period-cohort model provided an acceptable fit in both sexes. Predictions based on these models gave a mortality rate of 122 and 92 per 100 000 for the period 1999-2003 in men and women, respectively. Despite an aging and increasing population, the total number of stroke deaths in Sweden is predicted to decline by approximately 10% in both men and women from 1989-1993 to 1999-2003.
Conclusions Both factors, cohort and calendar period, contain relevant information to explain the decline in stroke mortality trends in Sweden. Predictions indicate that the decline of both age-specific and total mortality will continue.
During the last decades there has been a declining mortality from heart disease and stroke in the United States and western Europe, whereas an increase in mortality has been observed in many countries in eastern Europe.1 2 A high degree of concordance in mortality trends for heart disease and stroke is reported,3 indicating changes in common causes for the two groups of diseases. The changes in mortality may be due to changes in disease incidence or case-fatality rate. Incidence in turn is affected by changes in the cardiovascular risk-factor profile in the community. Variations in case-fatality rate can be explained partly by changes in medical care. Changes in risk-factor profile could also affect case fatality indirectly by changing the severity of the disease.
The age-period-cohort model4 5 6 has been used to describe variations in mortality and incidence over time. The effect of birth cohort includes risk factors and environmental exposures that are present in early life.7 Period effects contain factors that act around the time of death. These include short-term effects of primary and secondary prevention measures, modified or new medical care procedures, and modifications in certification practices.7
The aim of this study was to analyze the variation in mortality rates for stroke in Sweden during the period 1969 through 1993. Specifically, it was of interest to identify and measure effects of the three interrelated factors—age, calendar period of death, and birth cohort—on mortality from stroke. In addition, the estimated models for mortality rates were used to make projections of future mortality rates and total number of deaths from stroke, thereby assessing effects of changing mortality on public health and healthcare planning.
Subjects and Methods
Mortality and Population Data
Data on number of deaths, grouped by sex and 5-year age group (25 to 89 years old) from 1969 through 1993 were obtained from the Central Bureau of Statistics, Sweden. Stroke was defined by codes 430 through 438 in the 8th and 9th revisions of the International Classification of Diseases (ICD). Data on mid-year population estimates and forecasts for population statistics for the years 1994 through 2003 were obtained from publications of the Central Bureau of Statistics.
Annual mortality rates for men and women were age-adjusted with the Swedish population in 1981 as a standard population and are presented by calendar period and sex. Age-specific mortality rates are presented by sex and birth cohort in 5-year age groups. Furthermore, the time trends were analyzed with an age-period-cohort model.4 5 6 In brief, the mortality rate in a given age group and time period is a function of the factors of age, calendar period, and birth cohort.
Parameters of the model were estimated with the method of maximum likelihood using the Gauss software package.8 In addition to the full model with all three factors age, period, and cohort, various submodels were estimated to evaluate which factors were important. Goodness of fit of the models was evaluated by the deviance measure. Different models were compared using difference in deviance. Pearson's residuals were calculated and analyzed to assess the assumptions of the models.
The final model was used to forecast the mortality rates for stroke for the next two 5-year periods (1994-1998 and 1999-2003). The unknown values for period and cohort effects were estimated by linear regression using arbitrarily chosen numbers of the most recent period and cohort values.9 The estimated mortality rates were applied to predictions of population estimates to predict the total number of deaths for the next two periods. The linear dependency between the factors of age, calendar period, and birth cohort implies that estimates of the parameters in the model are not unique. However, estimating future mortality rates with an age-period-cohort model is not affected by the problem of nonidentifiability.9
Description of Trends
During the study period 1969-1993, a total of 237 357 deaths from stroke were available in the routine mortality register in Sweden. Fig 1⇓ shows age-adjusted mortality rates for stroke in men and women, aged 25 to 89 years, for the study period. Mortality rates for stroke decreased from 203 to 143 per 100 000 for men and from 185 to 113 per 100 000 for women (average annual decrease of 1.3% for men and 1.9% for women). The relative risk between men and women increased steadily over the study period, being 1.09 and 1.27 in 1969 and 1993, respectively (P<.0001 for test of trend).
Fig 2⇓ shows the age-specific stroke mortality rates by sex and birth cohort. For all age groups, the mortality rate decreased with increasing birth cohort. The relative decrease in mortality rate with increase in birth cohort was greater for the younger age groups. For the youngest age group, the mortality rate decreased from 2.73 to 1.26 per 100 000 (2.7% decline per year) among women and from 2.42 to 1.25 per 100 000 (2.4% decline per year) among men. In the oldest age group, the mortality rate decreased from 2403 to 1994 per 100 000 (0.9% decline per year) among women and 2335 to 1966 per 100 000 (0.8% decline per year) among men.
Age, Period, and Cohort Effects
For stroke mortality, the full age-period-cohort model provided an acceptable fit in both sexes (deviance [D]=45.0, P=.08 for men; D=41.3, P=.15 for women; df=33). Residuals from the models did not reveal any systematic patterns. The full model was an improvement over both an age-period and an age-cohort model (men: difference in deviance [ΔD]=71.0, P<.0001, df=3 for age-cohort model and ΔD=177.5, P<.0001, df=15 for age-period model; women: ΔD=148.2, P<.0001, df=3 for age-cohort model and ΔD=489.2, P<.0001, df=15 for age-period model). In addition, the age-period and age-cohort models provided a significantly better fit than a model with the factors age and linear effect of calendar period.
Predictions of Future Mortality Rates
Predictions of future mortality rates for stroke were based on the full age-period-cohort model because it could explain variation in rates significantly better than age-cohort and age-period models. Predictions were sensitive to numbers of values included to predict unknown period values, although in all cases predictions indicated a continued decreasing trend. Predictions were less sensitive to the numbers of values included to predict cohort effects because cohort values to be predicted affect only the youngest age groups, in which the numbers of deaths are small. We report results from calculations based on the three most recent period values and the five most recent cohort values because the regression models used provided the best fit with these numbers. To evaluate how accurate predictions from the models would be, we used only a subset of the data (periods 1969-1973, 1974-1978, and 1979-1983) to estimate a full age-period-cohort model, and we used it to predict mortality rates for periods 1984-1988 and 1989-1993. The predictions were accurate for men. Among elderly women, the observed mortality rates were approximately 6% and 16% higher than the predicted rates for the periods 1984-1988 and 1989-1993, respectively. When projecting mortality rates and total number of deaths for women, we also report values that were revised for this underestimation.
The Table⇓ shows the age-specific and age-adjusted observed and predicted mortality rates from stroke for men and women. Age-standardized mortality from stroke was predicted to decrease from 195 to 122 per 100 000 for men and from 177 to 92 per 100 000 for women between 1969-1973 and 1997-2003. Predictions of future age-specific mortality rates showed a decreasing trend for all age groups.
Prediction of number of total deaths was made by applying the predicted age-specific mortality rates to the projected age-specific population sizes. The observed and predicted total numbers of deaths from stroke in the age group 25 to 89 years are shown in Fig 3⇓. As seen in Fig 3⇓, total number of deaths is not going to increase, despite an increasing and aging population. The total numbers of deaths were 19 700 and 24 500 for men and women, respectively, in the period 1989-1993. The predicted numbers of deaths for 1999-2003 were 18 400 and 21 200 for men and women, respectively. An alternate higher total number of deaths is also shown for women in Fig 3⇓; these numbers are revised to account for the underestimation mentioned above.
Stroke accounts for approximately 12% of all deaths in Sweden, which is comparable with other industrialized countries. The declining trend of age-adjusted stroke mortality has been observed in most industrialized countries, and in some populations there has been a more rapid decrease since the 1960s.7 10 In this study, mortality rates from stroke in Sweden were found to have steadily declined in both sexes during the 1969-1993 period; the decline in age-adjusted mortality has been almost linear. The decrease in mortality rates has been somewhat more marked for women. The phenomenon of increasing mortality ratio between men and women has been observed worldwide,11 but the reasons for this are unclear.
Changes in mortality rate over the years could be explained by changes in incidence rate or by changes in case-fatality rate. Niessen et al12 concluded that stroke incidence has decreased, and they projected a continuing decrease in The Netherlands. On the other hand, Malmgren et al13 projected an increase in the number of patients with first-ever strokes in England and Wales. Studies from the United States and the Scandinavian countries indicate that changes in stroke incidence are insufficient to explain the declining stroke mortality.14 15 16 17 Reports from the World Health Organization MONICA study show that the incidence of stroke in Northern Sweden has not been changing in women, whereas in men there has been a moderate shift toward higher ages during 1985-1991.18 Case-fatality rates in northern Sweden have been declining among subjects with first-ever stroke.18 Other studies from Sweden report a decreasing case fatality during the 1970s and early 1980s.19 20 Declining case fatality can result from better medical management or a change in the severity of stroke. Changes in risk-factor levels in the population could lead to a redistribution of stroke subtypes, resulting in less severe clinical manifestations. An increase in public awareness of stroke could also explain stroke becoming a less severe disease. Studies that support the hypothesis that stroke is becoming a less severe disease have been published.18 21
In this study, the age-period-cohort model was used to describe variations in mortality rates over the years. The effect of birth cohort can be seen as the risk factors and environmental exposures that are present in early life or are typical for a given generation.7 The period effects include factors that act around the time of death. Variations in mortality trends for stroke were best described by the full age-period-cohort model for both men and women. This means that when explaining variations in rates over the years, the cohort factor contains information even if one adjusts for period effects, and the period factor contains information even after adjustment for cohort effects. It is not possible to conduct a formal test to determine which factor, cohort or period, contributes the most to explaining the changing rates.4 Still, the age-period-cohort model provides a method to assess the significance of the three interrelated factors (age, calendar period of death, and birth cohort) on mortality. Furthermore, parameters in the full model cannot be estimated uniquely without further restrictions or assumptions. That makes the interpretation of the parameter values difficult and thereby comparisons between different cohorts and periods complicated.
Projected mortality rates based on the age-period-cohort models indicate that the observed declining trend is to continue. When combining predicted mortality rates with predicted population statistics, the total future number of deaths can be estimated. Our analysis showed that if the rate of decline in stroke mortality follows the same pattern as it has until now, it will overcome the increase in numbers due to the aging and the increasing size of the population. This will result in no increase in the total number of deaths, which will remain approximately at the same level as today or even decrease somewhat.
The use of age-period-cohort models for predictions of future mortality rates is based on the assumption that the progress we have seen in cohort and period effects is to continue. Therefore, it is not possible to take into account factors, such as new medical treatments or changes in risk-factor profile in the community, that do not agree with the observed trends. In addition, the predictions are sensitive to the number of values chosen to predict period effects, an arbitrary choice. Nevertheless, the advantage of using age-period-cohort models to forecast future age-specific rates is that they simultaneously take into account the effects of age, calendar period of death, and birth cohort. It should be noted that predictions of total number of deaths include an extra source of uncertainty, namely, predictions of population growth.
The validation of prognostic ability of the present models using only a subset of data showed that for the oldest age groups among women, predicted values declined more than was actually observed. We were not able to find a reason for this, since residuals from the model did not reveal any systematic pattern and the fit of the model was good. If this applies to periods analyzed with the full set of data, the decline among women will not be as great as reported here. This could indicate that the rate of decline among women is actually flagging.
Changes in coding practices of diseases could affect mortality statistics. In Sweden, the ICD coding practice was changed once during the study period (from the 8th to the 9th revision in 1986), but differences in the 8th and 9th revision of ICD codes are small for the stroke diagnoses codes 430 through 438.22 In our study, mortality rates during the period 1969-1985 with the 8th revision of ICD and 1986-1993 with the 9th revision showed similar trends. The broad definition of stroke used in the study, including ICD codes 430 through 438, improves the validity of comparing data from the two different time periods.
If the incidence rate remains stable, and severity of stroke is decreasing, this will result in more survivors after stroke and increased costs for secondary prevention. However, the total costs of stroke are dominated by costs for in-hospital care and long-term institutional care.23 In terms of total costs for society, future costs will approximately follow the time trend for the total number of stroke deaths in the community24 ; therefore, based on our results the total cost of stroke will remain at the present level or decrease somewhat.
In summary, stroke mortality in Sweden has been declining in both sexes and in all age groups. The relative risk of death from stroke between men and women has increased. Both cohort and calendar period factors contain information relevant to the explanation of variations in stroke mortality trends. Predictions of mortality rates indicate that the decline in stroke mortality is to continue. The total number of deaths due to stroke is not going to increase, despite the aging and increasing population.
This study was supported by grants from the Swedish National Public Health Institute, the Swedish Medical Research Council (27X-07192), the Heart and Chest Fund, the Swedish Stroke Association, the Joint Committee of the Northern Sweden Health Care Region, and the Norrbotten and Västerbotten County Councils.
- Received May 2, 1996.
- Revision received July 9, 1996.
- Accepted July 10, 1996.
- Copyright © 1996 by American Heart Association
Bonita R, Stewart A, Beaglehole R. International trends in stroke mortality. Stroke. 1990;21:989-992.
Thom TJ, Epstein FH. Heart disease, cancer, and stroke mortality trends and their interrelations: an international perspective. Circulation. 1994;90:574-582.
Wolfe CD, Burney PG. Is stroke mortality on the decline in England? Am J Epidemiol. 1992;136:558-565.
Aptech Systems Inc. Gauss, Release 3.0. 1992.
Osmond C. Using age, period and cohort models to estimate future mortality rates. Int J Epidemiol. 1985;14:124-129.
Klag MJ, Whelton PK, Seidler AJ. Decline in US stroke mortality: demographic trends and antihypertensive treatment. Stroke. 1989;20:14-21.
Zhang XH, Sasaki S, Kesteloot H. Changes in the sex ratio of stroke mortality in the period of 1955 through 1990. Stroke. 1995;26:1774-1780.
Niessen LW, Barendreg JJ, Bonneux L, Koudstaal PJ. Stroke trends in an aging population. Stroke. 1993;24:931-939.
Malmgren R, Bamford J, Warlow C, Sandercock P, Slattery J. Projecting the number of patients with first ever strokes and patients newly handicapped by stroke in England and Wales. BMJ. 1989;298:656-660.
Jørgensen HS, Plesner AM, Hübbe P, Larsen K. Marked increase of stroke incidence in men between 1972 and 1990 in Fredriksberg, Denmark. Stroke. 1992;23:1701-1704.
Tuomilehto J, Sarti C, Narva EV, Salmi K, Sivenius J, Kaarsalo E, Salomaa V, Torppa J. The FINMONICA Stroke Register: community-based stroke registration and analysis of stroke incidence in Finland, 1983-1985. Am J Epidemiol. 1992;135:1259-1270.
Alfredsson L, von Arbin M, de Faire U. Mortality from and incidence of stroke in Stockholm. Br Med J. 1986;292:1299-1303.
Stegmayr B, Asplund K, Wester PO. Trends in incidence, case-fatality rate, and severity of stroke in northern Sweden, 1985-1991. Stroke. 1994;25:1738-1745.
Harmsen P, Tsiogianni A, Wilhelmsen L. Stroke rates were unchanged, while fatality rates declined, during 1971-1987 in Göteborg, Sweden. Stroke. 1993;23:1410-1415.
Terént A. Increasing incidence of stroke among Swedish women. Stroke. 1988;19:598-603.
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.
Statistiska centralbyrån. Klassificering av dödsorsaker i svensk statistik (Classification of Causes of Death in Swedish Statistics). Stockholm, Sweden: Statistics Sweden; 1990:3.
Terént A, Marké L-Å, Asplund K. Costs of stroke in Sweden: a national perspective. Stroke. 1994;25:2363-2369.
Asplund K, Peltonen M, Stegmayr B. From the 20th to the 21st century: health care and socioeconomic issues. In: Ginsberg MD, Bogousslavsky J, eds. Cerebrovascular Disease: Pathophysiology, Diagnosis and Management. Oxford, UK: Blackwell Scientific Publications Inc. In press.