Published online before print June 19, 2003, doi: 10.1161/01.STR.0000078661.72578.0A
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
(Stroke. 2003;34:1610.)
© 2003 American Heart Association, Inc.
Original Contributions |
Improvements in Treatment of Coronary Heart Disease and Cessation of Stroke Mortality Rate Decline
From the Department of Public Health, Faculty of Medicine, Erasmus Medical Center, Rotterdam, Netherlands.
Reprint requests to A. Peeters, PhD, Department of Public Health, Erasmus Medical Center, PO Box 1738, 3000 DR Rotterdam, Netherlands. E-mail peeters{at}mgz.fgg.eur.nl
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionAppendixReferences |
|---|
Background and Purpose— Many countries observed rapidly declining stroke mortality rates during 1970–1990, followed by a slowing or a cessation of this decline. This slowing was seen for both sexes and all ages. Here we test the hypothesis that improvements in coronary heart disease (CHD) survival can explain this slowing through an increase in the number of CHD survivors at an increased risk for stroke.
Methods— We created multistate life-table models based on the survival experience of 46 years of follow-up of the Framingham Heart Study cohort. Improvements in survival after CHD were modeled by decreasing mortality rates for those with CHD. We analyzed whether improved CHD survival could result in a >3% increase in annual stroke mortality rates, which would be enough to eliminate the previously observed decline.
Results— CHD survival improvements led to an increase in the number of stroke deaths but also a concomitant increase in the total population size. Under no circumstances was there an annual increase in stroke mortality rates approaching 3% for both sexes and for younger and older age groups.
Conclusions— The hypothesis that increases in the numbers of people with CHD, as a consequence of improvements in CHD survival, explain the observed slowing of the stroke mortality rate decline must be rejected. The true explanation is also likely to be a factor that changed markedly around 1990, but with more direct effects on stroke mortality.
Key Words: coronary heart disease • incidence • models, theoretical • stroke
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendixReferences |
|---|
Many countries observed rapidly declining stroke mortality rates during 1970–1990, followed by a lessening or a cessation of this decline after approximately 1990.1–3 In the United States there were reports of a slowing of the 3% to 4% annual decline in stroke mortality rates, with similar changes reported for both sexes and for younger and older age groups (Figure 1). 1–3 Because the cause is unknown, predictions cannot be made regarding future trends or regarding optimal strategies to restimulate a decline. Three major hypotheses for the decreasing decline in stroke mortality rates are as follows: (1) that the minimum stroke mortality rates have been obtained4; (2) that improvements in hypertension control have slowed and other risk factors such as diabetes are increasing2; and (3) that there is an increase in the general level of risk of stroke in the population resulting from more survivors with coronary heart disease (CHD), following significant improvements in survival with CHD in the 1970s and 1980s.1,3 Here we focus on the latter. We use models to analyze whether increases in the number of people with CHD, resulting from improvements in survival, could lead to a great enough increase in population stroke mortality rates to eliminate the previously observed decline.
|
See Editorial Comment, page 1615
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendixReferences |
|---|
To estimate the annual rate of decline in the stroke mortality rate in the United States during 1981–1991, we used log-linear regression on the number of stroke deaths (obtained for International Classification of Diseases, Ninth Revision [ICD-9] codes 430 to 438 from the Compressed Mortality Database, 1979–19985) in a specified sex and age group, using the midyear population as offset.
To identify whether improvements in survival after CHD could plausibly lead to the annual increase in stroke mortality rates required to eliminate the stroke mortality rate declines observed in the United States, we performed 2 analyses. First, we used a simple model (Figure 2) to identify factors affecting the degree of increase in stroke mortality rates associated with improved post-CHD survival. We analyzed the independent effects of (1) the absolute stroke mortality risk in those without CHD, (2) the relative risk of stroke mortality after CHD, (3) the initial CHD prevalence, and (4) the magnitude of the increase in the population with CHD (such as would occur as a result of improved survival after CHD).
|
Second, we analyzed the effect of improved survival after CHD in a more realistic setting, in which altered survival at one age affects mortality at subsequent ages. We constructed a multistate life table, using data from the United States to enable comparison with the changes in the American population stroke mortality rates. This model was similar to the aforementioned model with the addition of an age dimension and transitions between health states. The model has the states "no CHD," "CHD," and "death," and transitions to death can be stroke or nonstroke mortality. We constructed the multistate life table using age-specific transition rates derived from 46 years of follow-up of the Framingham Heart Study.6,7 We estimated each set of age-specific transition rates ("no CHD to CHD," "no CHD to nonstroke death," "no CHD to stroke death," "CHD to nonstroke death," and "CHD to stroke death") for ages 40 to 90 years using Poisson regression with age entered continuously, using the best-fitting polynomial representation (STATA 7) (for further explanation, see the Appendix, which can be found online at http://stroke.ahajournals.org). During follow-up of the 5209 original participants, there were 3678 (1812 men and 1866 women) nonstroke deaths, 281 (114 men and 167 women) stroke deaths, and 1907 (1055 men and 852 women) cases of incident CHD. Life-table models were constructed separately for each sex and represent ages 40 to 85 years, starting with a cohort free of known CHD at age 40 years. To analyze the consequences of improved survival after CHD, we modeled a constant 30% decrease in the risk of nonstroke mortality from the state of CHD for all ages. This was the relative mortality risk (at both 28 days and up to 4 years [95% CI, 0.54 to 0.83 in men and 0.54 to 0.98 in women]) after myocardial infarction in patients in 1980 compared with 1970 in the United States.8 We analyzed the change in stroke mortality in the year after the improvements in post-CHD survival. We also analyzed the change in the second and third years after the improvements in post-CHD survival. In essence, if a proportion of 40-year-old persons with CHD are prevented from dying, in the next year there will be more 41-year-old persons alive and at risk of dying from stroke. If the improvements in survival after CHD were the cause of the higher than expected annual stroke mortality rates, we would expect to see an annual increase in stroke mortality rates in the life-table models of the necessary magnitude.
The outcomes estimated from the life tables were the annual changes in the following: the number of stroke deaths at a given age, the stroke mortality rate at a given age, and the age-standardized stroke mortality rate between ages 50 to 74 years and ages 75 to 84 years, for comparison with the literature. Age standardization was conducted with the use of the 1990 US population.9 All outcomes were estimated separately for men and women.
The sensitivity of the results to the underlying transition rates was tested through analyses with the use of transition rates representing the 95% confidence limits. We estimated the outcomes of a "worst case scenario" using the combination of the transition rate limits giving the greatest increase in the stroke mortality rate. We also performed sensitivity analyses on the degree of improvement in survival after CHD using a mortality decrease of 50% (based on the limit of the described 95% CI8). We estimated the maximum theoretical change between ages 40 and 85 years associated with improvements in survival after CHD by analyzing a life table in which the effects have flowed through all the ages and the population has reached equilibrium. For example, changes in the stroke mortality rate for 85-year-old persons are derived from changes at all ages before and including age 85 years. Theoretically, it would take 45 years for the full effect derived by this life table to be seen in 85-year-old persons because it depends partly on changes flowing through from those aged 40 years.
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendixReferences |
|---|
The crude annual rates of stroke mortality decline during 1981–1991 in the United States were 4.1% (95% CI, 4.0 to 4.3) for men aged 35 to 74 years, 3.6% (95% CI, 3.5 to 3.7) for men aged
75 years, 4.1% (95% CI, 4.0 to 4.2) for women aged 35 to 74 years, and 3.1% (95% CI, 3.0 to 3.2) for women aged 75 years. Because the stroke mortality rates did not change (or slightly increased) in these groups during 1992–1995 (Figure 1), the objective of our analysis was to determine whether improvements in survival after CHD could plausibly lead to the 3% to 4% annual increase in stroke mortality rates required to eliminate the observed declines.
In a population homogeneous except for CHD status (Figure 2), the stroke mortality risk increases as the proportion of those with CHD increases (Table 1) (as long as the stroke mortality risk is greater in those with CHD). The degree of increase is independent of the underlying absolute stroke risk in the population free of CHD. It is dependent on the CHD prevalence and the relative risk for stroke associated with CHD (Table 1). To see a 3% increase in the stroke mortality risk, a combination such as a 25% increase in the population with CHD, relative risks of stroke after CHD of 2, and an initial CHD prevalence of 20% is needed (Table 1). These results suggest that increases in the stroke mortality rate of >3% could only be expected in a very limited range of age and sex groups. High CHD prevalence (>20%) and a large increase in the population with CHD (25%) are more likely in older, male populations because they require both high absolute risks of CHD and an accumulation of survival benefits. In contrast, relative risks for stroke associated with prior CHD of 2 are restricted to younger age groups (data not shown). Finally, we would not expect such large increases in the population with CHD to be likely in a single year, suggesting that if stroke mortality rate increases of 3% due to improved CHD survival were possible, they could only occur after a number of years of accumulation rather than consecutively each year.
|
To take into account the accumulation of survivors with CHD across ages, we created a multistate life table representing the situation before improvements in post-CHD survival. The age-standardized population stroke mortality rates derived from this model were 81/100 000 (ages 40 to 74 years) and 534/100 000 (ages 75 to 84 years) for men and 67/100 000 (ages 40 to 74 years) and 438/100 000 (ages 75 to 84 years) for women. These figures can be compared with the US national stroke mortality rates during 1989–1993, which were 42/100 000, 526/100 000, 33/100 000, and 436/100 000, respectively.3 The total mortality rates were also comparable to the US white population rates for 1989–199110 (data not shown). For men, the CHD prevalence was 20% at age 65 years and 29% at age 75 years. The relative risk of stroke mortality after CHD was 2.0 at age 65 years and 1.6 at age 75 years.
For men, a 30% reduction in nonstroke mortality from the state of "CHD" resulted in a 1-year increase in the number of stroke deaths of 1% at age 75 (Table 2). The stroke mortality rate increased by only 0.3%. This discrepancy arises as a result of the simultaneous increase in the number of stroke deaths and the number of person-years lived (Figure 3). The age-standardized stroke mortality rates increased by 0.2% for ages 40 to 74 years and by 0.3% for ages 75 to 84 years. For women, the increases were even less (Table 2). Lesser rate increases were seen for the second and third years after the introduction of the change.
|
|
These results were not very sensitive to the nonstroke mortality rate or the CHD incidence rate. A survival improvement of a 50% decrease in the post-CHD nonstroke mortality rate led to annual increases in the age-standardized stroke mortality rates of 0.4% for men and 0.2% for women. A worst case scenario using the upper 95% CI limits for the transitions "no CHD to CHD," "no CHD to nonstroke death," "CHD to nonstroke death," and "CHD to stroke death" and the lower 95% CI limits for the transition "no CHD to stroke death" was still associated with less than a 1% increase in the annual stroke mortality rate for men and women at all ages (Table 2). Varying the type of polynomial representation for age within the Poisson models predicting the transition rates had very little effect on the outcome. Analysis of the maximum hypothetical effect after a 30% reduction in nonstroke mortality from the state of "CHD" for 45 years resulted in increases in the male age-standardized stroke mortality rates of only 2.0% for ages 40 to 74 years and 2.6% for ages 75 to 84 years (Table 2).
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendixReferences |
|---|
We have shown, using models, that changes in survival after CHD are unable to cause the 3% to 4% annual increase in stroke mortality rates required to eliminate the stroke mortality rate decline observed in the United States before 1991. We found that a 30% to 50% decrease in nonstroke mortality rates in those with CHD would lead to a small increase in stroke mortality rates and that a variety of factors affect the degree of this effect. However, we found no reasonable scenarios in which the increase in stroke mortality rate was >1% per year. This is because while increased numbers of CHD survivors increase the number of stroke deaths in the population, they also inflate the population size.
This is the first study to test this hypothesis. To examine the effect of improved CHD survival independently from the many other population changes, it was necessary to use models. To evaluate the robustness of our conclusions requires consideration of the sources of variance in the estimates, the limitations of this model type, and the degree to which the base model represents the situation in the United States before 1991.
Unfortunately, it is not straightforward to produce CIs for our outcomes because the variance needs to be combined from the 5 (dependent) regression analyses and the life tables, which also have internal dependence between transition types and across ages. Uncertainty is also found in the estimation of the degree of CHD survival improvement. Therefore, we chose to explore the impact of possible variance through a range of uncertainty analyses. We found that reasonable (and extreme) variations (based on estimated CIs) in the transition rates and in the degree of survival improvement did not alter the conclusions.
To estimate the annual effect of changes across all ages, we used life-table models. Other, more "2-dimensional" models, such as time series analyses, suffer from the difficulty of linking changes in one age group with outcomes in all future age groups, with appropriate time intervals taken into account. The life table does this automatically. An empirical approach closer to the life table would be to examine whether improved survival from CHD in middle age was linked to increased stroke mortality rates at older ages within a given cohort. However, it would still be difficult to exclude the effect of all other population changes affecting stroke mortality rates. One principal assumption of the life table is that within each category the population is homogeneous, including the markovian assumption that transitions are independent of prior history. This is appropriate for analyzing the effect of an increase in the proportion of the population with CHD. However, it is important to recognize that such a model does not evaluate possible changes in the stroke mortality risk of those with CHD associated with improvements in treatment.
One other potential limitation of this model is that the transition rates are estimated from a long follow-up period within the Framingham Heart Study during 1948–1998. We demonstrated that important factors affecting the degree of change in stroke mortality rates were the relative risk of stroke mortality after CHD and the prevalence of CHD. Therefore, insofar as these have not increased greatly over time, we would expect models based on other, more recent, data to produce similar results. A CHD prevalence of 25% in 70-year-old men is similar to the prevalence of 26% observed for heart disease in 65- to 74-year-old men in the United States in 1989.11 However, for women the prevalence of CHD at age 70 years was only 14%, compared with a reported prevalence of 21%.11 While this may have led to underestimation of the increase in the stroke mortality rate in women, we would still not expect the increase to be greater than that seen for men. We also showed that the age-specific total and stroke mortality rates derived from the models were similar to American mortality rates around 1991, with stroke mortality rates more similar for the older than younger age groups. The similarity of the mortality rates arises because the majority of Framingham deaths occur later in the follow-up period and because the population is a selected, relatively healthy population.7,12 Those dying young were more likely to have died in the earlier follow-up period and therefore reflect 1990 rates less closely. A further limitation of the Framingham population is that it consists of primarily white Americans. However, since the decline in stroke mortality rates in the United States has been the greatest for the white population,4 our results make the tested hypothesis even less likely.
These results suggest that improvements in CHD survival may not be accompanied by as large an increase in the burden of disease in the elderly as has been feared.1 While the absolute number of stroke deaths (and before them incident strokes) will increase, the age-specific population rates will not change substantially. Furthermore, these results suggest that we may be able to restimulate the stroke mortality rate decline if it is not an unavoidable consequence of a desired treatment improvement. This coincides well with the data showing that the stroke mortality rate in the United States is once again declining (Figure 1). Further understanding of the factors driving the changes in stroke mortality rates is required if we are to optimize this decline.
We demonstrate that increases in the population with CHD cannot alone explain the higher than expected stroke mortality rates in the 1990s. The remaining possibilities are that stroke mortality rate changes were caused by factors other than improvements in survival with CHD (such as the worsening of prevalence and control of other risk factors) or that with improvements in CHD survival the risks of stroke mortality in the population with CHD also increased. Competing theories will have to explain why the slowing of the decline occurred around 1990 and was observed to a similar extent in all sex and age groups (Figure 1) and in widely different healthcare systems such as those in the United States, the Netherlands, and Sweden.1,3 They will also have to explain why the stroke mortality rates appear to be declining once again. An additional hypothesis fitting these conditions is the international introduction of a new therapy with immediate effects on stroke mortality. The introduction of thrombolytic therapy for myocardial infarction, with its associated hemorrhagic side effects,13 is one possible and testable candidate.
|
Appendix |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendixReferences |
|---|
The transition rates were estimated separately for each sex with the use of the following equations:
Non-CHD to CHD: ha=Exp(
+ß1a+ß2a2+ß3a3)
Non-CHD to nonstroke death: ha=Exp(+ß1a+ß2a2)
Non-CHD to stroke death: ha=Exp(+ß1a)
CHD to nonstroke death: ha=Exp(+ß1a+ß2a2)
CHD to stroke death: ha=Exp(+ß1a)
where ha is the transition rate at a given age; , ß1, ß2, and ß3 are the regression coefficients; and a represents age (single year at follow-up).
To derive these equations, each individual was divided into a number of consecutive data entries, representing each year of age between study entry and exit. Each entry included CHD status and occurrence or not of CHD, stroke death, or nonstroke death at that age. On the basis of these data, for each transition we performed Poisson regression on the number of events, with the person-years lived as the offset, using single year of age as the independent variable.
|
Acknowledgments |
|---|
This study was supported by grants from the Netherlands Heart Foundation and the Netherlands Organization for Scientific Research. The authors would like to acknowledge the Framingham Heart Study coordinators for access to the original data set. The Framingham Heart Study is conducted and supported by the National Heart, Lung, and Blood Institute (NHLBI) in collaboration with the Framingham Heart Study Investigators. The Netherlands Epidemiology and Demography Compression of Morbidity Research Group (NEDCOM) includes, in addition to the listed authors, the following: F. Janssen, A. Kunst, C. de Laet, A.A. Mamun, W. Nusselder, and F.J. Willekens. The authors would like to thank Professor P.J. Koudstaal for critical review of the manuscript and Dr C. Looman for helpful discussions.
Received November 20, 2002; revision received February 5, 2003; accepted March 5, 2003.
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendixReferences |
|---|
1. Bonneux L, Looman C, Barendregt J, Maas PVD. Regression analysis of recent changes in cardiovascular morbidity and mortality in the Netherlands. BMJ. 1997; 314: 789–792.
2. Cooper R, Cutler J, Desvigne-Nickens P, Fortmann S, Friedman L, Havlik R, Hogelin G, Marler J, McGovern P, Morosco G, et al. Trends and disparities in coronary heart disease, stroke, and other cardiovascular diseases in the United States. Circulation. 2000; 102: 3137–3147.
3. Sarti C, Rastenyte D, Cepaitis Z, Tuomilehto J. International trends in mortality from stroke, 1968 to 1994. Stroke. 2000; 31: 1588–1601.
4. Howard G, Howard VJ, Katholi C, Oli MK, Huston S. Decline in US stroke mortality: an analysis of temporal patterns by sex, race, and geographic region. Stroke. 2001; 32: 2213–2220.
5. Centers for Disease Control and Prevention. Compressed Mortality Database, 1979–1998. http://wonder.cdc.gov/mortsql.shtml. Accessed October 2002.
6. Dawber T, Meadors G, Moore F. Epidemiological approaches to heart disease: the Framingham Study. Am J Public Health. 1951; 41: 279–286.
7. Peeters A, Mamun AA, Willekens F, Bonneux L. A cardiovascular life history. Eur Heart J. 2002; 23: 458–466.
8. McGovern P, Folsom A, Sprafka M, Burke G, Doliszny K, Demirovic J, Naylor J, Blackburn H. Trends in survival of hospitalized myocardial infarction patients between 1970 and 1985: the Minnesota Heart Survey. Circulation. 1992; 85: 172–179.
9. Centers for Disease Control and Prevention. Census State Population Projections, 1990–2020. http://wonder.cdc.gov/popu00.shtml. Accessed October 2002.
10. National Center for Health Statistics. U. S. Decennial Life Tables for 1989–91: Volume I, Number I, United States Life Tables. Hyattsville, Md: US Dept of Health and Human Services; 1998.
11. National Center for Health Statistics. Current Estimates From the National Health Interview Survey, United States, 1989. Hyattsville, Md: US Dept of Health and Human Services; 1989.
12. Leaverton P, Sorlie P, Kleinman J, Dannenberg A, Ingster-Moore L, Kannel W, Cornoni-Huntley J. Representativeness of the Framingham risk model for coronary heart disease mortality: a comparison with a national cohort study. J Chron Dis. 1987; 40: 775–784.[CrossRef][Medline] [Order article via Infotrieve]
13. Vaitkus PT, Berlin JA, Schwartz JS, Barnathan ES. Stroke complicating acute myocardial infarction: a meta-analysis of risk modification by anticoagulation and thrombolytic therapy. Arch Intern Med. 1992; 152: 2020–2024.
This article has been cited by other articles:
 |
|
 |
|
  Y. Gerber, S. J. Jacobsen, R. L. Frye, S. A. Weston, J. M. Killian, and V. L. Roger Secular Trends in Deaths From Cardiovascular Diseases: A 25-Year Community Study Circulation, May 16, 2006; 113(19): 2285 - 2292. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C. S. Anderson, K. N. Carter, M. L. Hackett, V. Feigin, P. A. Barber, J. B. Broad, R. Bonita, and on behalf of the Auckland Regional Community Strok Trends in Stroke Incidence in Auckland, New Zealand, During 1981 to 2003 Stroke, October 1, 2005; 36(10): 2087 - 2093. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 L. Wilhelmsen, M. Koster, P. Harmsen, and G. Lappas Differences between coronary disease and stroke in incidence, case fatality, and risk factors, but few differences in risk factors for fatal and non-fatal events Eur. Heart J., September 2, 2005; 26(18): 1916 - 1922. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||